|
|
|
// Copyright 2020 The Abseil Authors.
|
|
|
|
//
|
|
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
|
|
// you may not use this file except in compliance with the License.
|
|
|
|
// You may obtain a copy of the License at
|
|
|
|
//
|
|
|
|
// https://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
//
|
|
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
// See the License for the specific language governing permissions and
|
|
|
|
// limitations under the License.
|
|
|
|
|
|
|
|
#include "absl/strings/cord.h"
|
|
|
|
|
|
|
|
#include <algorithm>
|
|
|
|
#include <atomic>
|
|
|
|
#include <cstddef>
|
|
|
|
#include <cstdio>
|
|
|
|
#include <cstdlib>
|
|
|
|
#include <iomanip>
|
|
|
|
#include <ios>
|
|
|
|
#include <iostream>
|
|
|
|
#include <limits>
|
|
|
|
#include <ostream>
|
|
|
|
#include <sstream>
|
|
|
|
#include <type_traits>
|
|
|
|
#include <unordered_set>
|
|
|
|
#include <vector>
|
|
|
|
|
|
|
|
#include "absl/base/casts.h"
|
|
|
|
#include "absl/base/internal/raw_logging.h"
|
|
|
|
#include "absl/base/macros.h"
|
|
|
|
#include "absl/base/port.h"
|
|
|
|
#include "absl/container/fixed_array.h"
|
|
|
|
#include "absl/container/inlined_vector.h"
|
|
|
|
#include "absl/strings/cord_buffer.h"
|
|
|
|
#include "absl/strings/escaping.h"
|
|
|
|
#include "absl/strings/internal/cord_data_edge.h"
|
|
|
|
#include "absl/strings/internal/cord_internal.h"
|
|
|
|
#include "absl/strings/internal/cord_rep_btree.h"
|
|
|
|
#include "absl/strings/internal/cord_rep_crc.h"
|
|
|
|
#include "absl/strings/internal/cord_rep_flat.h"
|
|
|
|
#include "absl/strings/internal/cordz_statistics.h"
|
|
|
|
#include "absl/strings/internal/cordz_update_scope.h"
|
Export of internal Abseil changes
--
ac1df60490c9583e475e22de7adfc40023196fbf by Martijn Vels <mvels@google.com>:
Change Cord constructor(string_view) to explicit make_tree and Cordz tracking
This CL changes the ctor to use an easier to maintain model where Cord code explicitly invokes Cordz update or new / tree logic, which avoids the ambiguity of the 'branched' InlineRep::set_tree code. This removes the need to equip InlineRep with 'MethodIdentifier' or other necessary call info, and also is a cleaner model: InlineRep is carrying too much code now that should plainly sit in Cord, especially with all internal abstractions having moved to InlineData.
See child CL(s) for desired state
PiperOrigin-RevId: 369433619
--
b665af7f586e6c679a8b27d4f78d5a1d2b596058 by Abseil Team <absl-team@google.com>:
Rename the 'Compare' template type to 'LessThan', as the passed-in function is expected to act like operator<. It is worth avoiding confusion with std::compare, which returns an int (-1/0/1), as due to implicit casting this can lead to hard-to-spot bugs.
PiperOrigin-RevId: 369391118
--
c3c775269cad0f4982ec63f3616dd78bb9e52dca by Martijn Vels <mvels@google.com>:
Integrate CordzUpdateTracker into CordzInfo
PiperOrigin-RevId: 369348824
--
771d81ed357496c117179e1daec76eba5155932d by Martijn Vels <mvels@google.com>:
Replace mutex() with Lock() / Unlock() function
Mini design future tracking of CordzInfo sampled cords: CordzInfo holds a CordRep* reference without a reference count. Cord is responsible for synchronizing updates for sampled cords such that the CordRep* contained in CordzInfo is at all times valid. This is done by scoping Lock() and Unlock() calls around the code modifying the code of a sampled cord. For example (using the future CL CordzUpdateScope()):
CordzInfo* cordz_info = get_cordz_info();
CordzUpdateScope scope(cordz_info, CordzUpdateTracker::kRemovePrefix);
CordRep* rep = RemovePrefixImpl(root);
set_tree(rep);
if (cordz_info) {
cordz_info->SetCordRep(rep);
}
On CordzInfo::Unlock(), if the internal rep is null, the cord is no longer sampled, and CordzInfo will be deleted. Thus any update resulting in the Cord being inlined will automatically no longer be sampled.
PiperOrigin-RevId: 369338802
--
5563c12df04a1e965a03b50bdd032739c55c0706 by Martijn Vels <mvels@google.com>:
Add UpdateTracker to CordzStatistics
PiperOrigin-RevId: 369318178
--
6b4d8463722a3e55a3e8f6cb3741a41055e7f83e by Martijn Vels <mvels@google.com>:
Add kClear, kConstructor* and kUnknown values and fix typo
PiperOrigin-RevId: 369297163
--
041adcbc929789d6d53371a8236840fc350e1eeb by Derek Mauro <dmauro@google.com>:
Switch from malloc to operator new in pool_urbg.cc
so it can only fail by throwing/aborting
PiperOrigin-RevId: 369274087
--
5d97a5f43e3f2d02d0a5bbe586d93b5751812981 by Benjamin Barenblat <bbaren@google.com>:
Correct Thumb function bound computation in the symbolizer
On 32-bit ARM, all functions are aligned to multiples of two bytes, and
the lowest-order bit in a function’s address is ignored by the CPU when
computing branch targets. That bit is still present in instructions and
ELF symbol tables, though; it’s repurposed to indicate whether the
function contains ARM or Thumb code. If the symbolizer doesn’t ignore
that bit, it will believe Thumb functions have boundaries that are off
by one byte, so instruct the symbolizer to null out the lowest-order bit
after retrieving it from the symbol table.
PiperOrigin-RevId: 369254082
--
462bb307c6cc332c1e2c3adb5f0cad51804bf937 by Derek Mauro <dmauro@google.com>:
Add a check for malloc failure in pool_urbg.cc
GitHub #940
PiperOrigin-RevId: 369238100
GitOrigin-RevId: ac1df60490c9583e475e22de7adfc40023196fbf
Change-Id: Ic6ec91c62cd3a0031f6a75a43a83da959ece2d25
4 years ago
|
|
|
#include "absl/strings/internal/cordz_update_tracker.h"
|
|
|
|
#include "absl/strings/internal/resize_uninitialized.h"
|
|
|
|
#include "absl/strings/str_cat.h"
|
|
|
|
#include "absl/strings/str_format.h"
|
|
|
|
#include "absl/strings/str_join.h"
|
|
|
|
#include "absl/strings/string_view.h"
|
|
|
|
|
|
|
|
namespace absl {
|
|
|
|
ABSL_NAMESPACE_BEGIN
|
|
|
|
|
|
|
|
using ::absl::cord_internal::CordRep;
|
|
|
|
using ::absl::cord_internal::CordRepBtree;
|
|
|
|
using ::absl::cord_internal::CordRepCrc;
|
|
|
|
using ::absl::cord_internal::CordRepExternal;
|
|
|
|
using ::absl::cord_internal::CordRepFlat;
|
|
|
|
using ::absl::cord_internal::CordRepSubstring;
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
using ::absl::cord_internal::CordzUpdateTracker;
|
|
|
|
using ::absl::cord_internal::InlineData;
|
|
|
|
using ::absl::cord_internal::kMaxFlatLength;
|
|
|
|
using ::absl::cord_internal::kMinFlatLength;
|
|
|
|
|
|
|
|
using ::absl::cord_internal::kInlinedVectorSize;
|
|
|
|
using ::absl::cord_internal::kMaxBytesToCopy;
|
|
|
|
|
|
|
|
static void DumpNode(CordRep* rep, bool include_data, std::ostream* os,
|
|
|
|
int indent = 0);
|
|
|
|
static bool VerifyNode(CordRep* root, CordRep* start_node,
|
|
|
|
bool full_validation);
|
|
|
|
|
|
|
|
static inline CordRep* VerifyTree(CordRep* node) {
|
|
|
|
// Verification is expensive, so only do it in debug mode.
|
|
|
|
// Even in debug mode we normally do only light validation.
|
|
|
|
// If you are debugging Cord itself, you should define the
|
|
|
|
// macro EXTRA_CORD_VALIDATION, e.g. by adding
|
|
|
|
// --copt=-DEXTRA_CORD_VALIDATION to the blaze line.
|
|
|
|
#ifdef EXTRA_CORD_VALIDATION
|
|
|
|
assert(node == nullptr || VerifyNode(node, node, /*full_validation=*/true));
|
|
|
|
#else // EXTRA_CORD_VALIDATION
|
|
|
|
assert(node == nullptr || VerifyNode(node, node, /*full_validation=*/false));
|
|
|
|
#endif // EXTRA_CORD_VALIDATION
|
|
|
|
static_cast<void>(&VerifyNode);
|
|
|
|
|
|
|
|
return node;
|
|
|
|
}
|
|
|
|
|
|
|
|
static CordRepFlat* CreateFlat(const char* data, size_t length,
|
|
|
|
size_t alloc_hint) {
|
|
|
|
CordRepFlat* flat = CordRepFlat::New(length + alloc_hint);
|
|
|
|
flat->length = length;
|
|
|
|
memcpy(flat->Data(), data, length);
|
|
|
|
return flat;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Creates a new flat or Btree out of the specified array.
|
|
|
|
// The returned node has a refcount of 1.
|
|
|
|
static CordRep* NewBtree(const char* data, size_t length, size_t alloc_hint) {
|
|
|
|
if (length <= kMaxFlatLength) {
|
|
|
|
return CreateFlat(data, length, alloc_hint);
|
|
|
|
}
|
|
|
|
CordRepFlat* flat = CreateFlat(data, kMaxFlatLength, 0);
|
|
|
|
data += kMaxFlatLength;
|
|
|
|
length -= kMaxFlatLength;
|
|
|
|
auto* root = CordRepBtree::Create(flat);
|
|
|
|
return CordRepBtree::Append(root, {data, length}, alloc_hint);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Create a new tree out of the specified array.
|
|
|
|
// The returned node has a refcount of 1.
|
|
|
|
static CordRep* NewTree(const char* data, size_t length, size_t alloc_hint) {
|
|
|
|
if (length == 0) return nullptr;
|
|
|
|
return NewBtree(data, length, alloc_hint);
|
|
|
|
}
|
|
|
|
|
|
|
|
namespace cord_internal {
|
|
|
|
|
|
|
|
void InitializeCordRepExternal(absl::string_view data, CordRepExternal* rep) {
|
|
|
|
assert(!data.empty());
|
|
|
|
rep->length = data.size();
|
|
|
|
rep->tag = EXTERNAL;
|
|
|
|
rep->base = data.data();
|
|
|
|
VerifyTree(rep);
|
|
|
|
}
|
|
|
|
|
|
|
|
} // namespace cord_internal
|
|
|
|
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
// Creates a CordRep from the provided string. If the string is large enough,
|
|
|
|
// and not wasteful, we move the string into an external cord rep, preserving
|
|
|
|
// the already allocated string contents.
|
|
|
|
// Requires the provided string length to be larger than `kMaxInline`.
|
|
|
|
static CordRep* CordRepFromString(std::string&& src) {
|
|
|
|
assert(src.length() > cord_internal::kMaxInline);
|
|
|
|
if (
|
|
|
|
// String is short: copy data to avoid external block overhead.
|
|
|
|
src.size() <= kMaxBytesToCopy ||
|
|
|
|
// String is wasteful: copy data to avoid pinning too much unused memory.
|
|
|
|
src.size() < src.capacity() / 2
|
|
|
|
) {
|
|
|
|
return NewTree(src.data(), src.size(), 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
struct StringReleaser {
|
|
|
|
void operator()(absl::string_view /* data */) {}
|
|
|
|
std::string data;
|
|
|
|
};
|
|
|
|
const absl::string_view original_data = src;
|
|
|
|
auto* rep =
|
|
|
|
static_cast<::absl::cord_internal::CordRepExternalImpl<StringReleaser>*>(
|
|
|
|
absl::cord_internal::NewExternalRep(original_data,
|
|
|
|
StringReleaser{std::move(src)}));
|
|
|
|
// Moving src may have invalidated its data pointer, so adjust it.
|
|
|
|
rep->base = rep->template get<0>().data.data();
|
|
|
|
return rep;
|
|
|
|
}
|
|
|
|
|
|
|
|
// --------------------------------------------------------------------
|
|
|
|
// Cord::InlineRep functions
|
|
|
|
|
|
|
|
#ifdef ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL
|
|
|
|
constexpr unsigned char Cord::InlineRep::kMaxInline;
|
|
|
|
#endif
|
|
|
|
|
Export of internal Abseil changes
--
07240ca7822d007cdcc79f2c40bd58b2c2010348 by Abseil Team <absl-team@google.com>:
Correct the comment from "AlphaNum" to "Arg".
PiperOrigin-RevId: 416139192
--
adcba4a6b3763626e1db7b1e8c108b3114903557 by Martijn Vels <mvels@google.com>:
Fix NewExternalRep() to require data being non-empty, and remove nullptr return.
PiperOrigin-RevId: 416135865
--
c0d14cd918fb16f15d1d84de9284b5c5ecc1f8f2 by Abseil Team <absl-team@google.com>:
Fix doc comment for absl::ascii_isprint().
The comment was incorrectly saying that it includes all whitespace.
It doesn't; the only whitespace char it includes is ' '.
PiperOrigin-RevId: 416112524
--
d83327800159c07002b6865e21232a12463e02dd by Abseil Team <absl-team@google.com>:
Internal change
PiperOrigin-RevId: 416099978
--
baf11e9ca42ca9140cdbf8075f971db8d65b1195 by Ilya Tokar <tokarip@google.com>:
Prevent compiler from optimizing Group_Match* benchmarks away.
Currently we benchmark single store of precomputed value.
Not all affected benchmarks show performance changes:
BM_Group_Match 0.53ns ± 1% 0.53ns ± 0% -0.42% (p=0.038 n=10+10)
BM_Group_MatchEmpty 0.26ns ± 1% 0.26ns ± 1% ~ (p=1.000 n=10+10)
BM_Group_MatchEmptyOrDeleted 0.26ns ± 1% 0.26ns ± 1% ~ (p=0.121 n=10+10)
BM_Group_CountLeadingEmptyOrDeleted 0.26ns ± 1% 0.45ns ± 0% +70.05% (p=0.000 n=10+8)
BM_Group_MatchFirstEmptyOrDeleted 0.26ns ± 0% 0.44ns ± 1% +65.91% (p=0.000 n=8+9)
But inspecting the generated code shows the difference,
e. g. BM_Group_MatchFirstEmptyOrDeleted
Before:
add $0xffffffffffffffff,%rbx
jne 30
After:
pcmpeqd %xmm0,%xmm0
pcmpgtb -0x30(%rbp),%xmm0
pmovmskb %xmm0,%eax
add: 0x23$0xffffffffffffffff,%rbx
jne 40
PiperOrigin-RevId: 416083515
--
122fbff893dc4571b3e75e4b241eb4495b925610 by Abseil Team <absl-team@google.com>:
Put namespace guard in ABSL_DECLARE_FLAG to make declaring a flag in a namespace a compiler error instead of a linker error.
PiperOrigin-RevId: 416036072
--
020fd8a20f5fa319e948846e003391fcb9e03868 by Ilya Tokar <tokarip@google.com>:
Make Cord::InlineRep::set_data unconditionally zero out memory.
Currently there is a single case where we don't zero out memory
as an optimization. Unconditional zeroing doesn't show any changes
in benchmarks, except for the unrelated improvement:
BM_CordPartialCopyToCord/1M/1 12.6ns ± 4% 12.6ns ± 4% ~ (p=0.857 n=16+19)
BM_CordPartialCopyToCord/1M/128 44.9ns ± 7% 45.0ns ± 3% ~ (p=0.468 n=18+17)
BM_CordPartialCopyToCord/1M/1k 64.5ns ± 4% 61.4ns ± 4% -4.82% (p=0.000 n=19+17)
BM_CordPartialCopyToCord/1M/8k 139ns ± 3% 128ns ±15% -7.76% (p=0.009 n=17+20)
BM_CordPartialCopyToCord/1M/16k 193ns ± 6% 168ns ± 6% -13.17% (p=0.000 n=17+17)
BM_CordPartialCopyToCord/4M/16k 199ns ± 4% 177ns ± 4% -11.36% (p=0.000 n=17+18)
BM_CordPartialCopyToCord/4M/32k 275ns ± 3% 250ns ± 4% -9.00% (p=0.000 n=18+18)
BM_CordPartialCopyToCord/4M/64k 291ns ± 4% 266ns ± 5% -8.53% (p=0.000 n=18+16)
BM_CordPartialCopyToCord/4M/128k 322ns ± 5% 291ns ± 4% -9.43% (p=0.000 n=20+18)
BM_CordPartialCopyToCord/8M/32k 281ns ± 5% 251ns ± 4% -10.38% (p=0.000 n=20+16)
BM_CordPartialCopyToCord/8M/64k 293ns ± 6% 267ns ± 4% -8.87% (p=0.000 n=16+19)
BM_CordPartialCopyToCord/8M/128k 334ns ± 3% 305ns ± 2% -8.56% (p=0.000 n=17+16)
This is clearly an alignmnet effect since number of the executed instructions is the same:
M_CordPartialCopyToCord/1M/1 155 ± 0% 155 ± 0% ~ (all samples are equal)
BM_CordPartialCopyToCord/1M/128 446 ± 0% 446 ± 0% ~ (p=0.332 n=36+39)
BM_CordPartialCopyToCord/1M/1k 473 ± 0% 473 ± 0% ~ (p=0.969 n=40+40)
BM_CordPartialCopyToCord/1M/8k 808 ± 0% 808 ± 0% ~ (p=0.127 n=40+39)
BM_CordPartialCopyToCord/1M/16k 957 ± 0% 957 ± 0% ~ (p=0.532 n=40+40)
BM_CordPartialCopyToCord/4M/16k 952 ± 0% 952 ± 0% ~ (p=0.686 n=39+39)
BM_CordPartialCopyToCord/4M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.690 n=40+40)
BM_CordPartialCopyToCord/4M/64k 1.23k ± 0% 1.23k ± 0% ~ (p=0.182 n=40+39)
BM_CordPartialCopyToCord/4M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.711 n=40+40)
BM_CordPartialCopyToCord/8M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.697 n=40+40)
BM_CordPartialCopyToCord/8M/64k 1.23k ± 0% 1.23k ± 0% +0.00% (p=0.049 n=40+40)
BM_CordPartialCopyToCord/8M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.507 n=40+40)
This makes code simpler and doesn't regress performance.
PiperOrigin-RevId: 415560574
--
37305b2690b31682088749e4d62f40d7095bdc54 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 415558737
--
86aaed569b9e743c1eb813a5f48def978a793db3 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415515201
--
6cdb8786cdcb4fa0b8a4b72fc98940877d1fdeff by Abseil Team <absl-team@google.com>:
Update SubmitMutexProfileData to accept wait_cycles instead of wait_timestamp
PiperOrigin-RevId: 415360871
--
9f979d307aa16ad09f214e04876cbe84395c0901 by Abseil Team <absl-team@google.com>:
absl::flat_hash_set compiles with -Wconversion -Wsign-compare
PiperOrigin-RevId: 415357498
--
9eceb14174708f15e61259d449b214a8a4c7f9e7 by Abseil Team <absl-team@google.com>:
Fix AddressIsReadable for the corner case of (aligned) addr == NULL.
PiperOrigin-RevId: 415307792
--
1a39ffe55898375e2d7f88c17c99db5a1b95b313 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415162872
--
64378549b110d5f5762185a5906c520fba70f0e7 by Abseil Team <absl-team@google.com>:
Fix a typo in the comments
PiperOrigin-RevId: 415088461
--
41aae8322e913b82710153c22b97c611fdb6e1fb by Abseil Team <absl-team@google.com>:
Switch from `connect` to `rt_sigreturn` -- the latter is much less problematic
for system call sandboxes.
PiperOrigin-RevId: 415073965
--
870c5e3388b6a35611bff538626fe7a1c8c87171 by Abseil Team <absl-team@google.com>:
Add ABSL_HAVE_HWADDRESS_SANITIZER and ABSL_HAVE_LEAK_SANITIZER
PiperOrigin-RevId: 414871189
--
f213ed60a66b58da7ac40555adfb1d529ff0a4db by Derek Mauro <dmauro@google.com>:
Remove reference to __SANITIZE_MEMORY__, which does not exist
It appears to have been copied by pattern matching from the ASAN/TSAN
code blocks.
https://github.com/gcc-mirror/gcc/blob/f47662204de27f7685699eeef89aa173ccf32d85/gcc/cppbuiltin.c#L79-L126
PiperOrigin-RevId: 414806587
--
b152891e73ab515f397ceb53f66c8ee2f33863ea by Abseil Team <absl-team@google.com>:
Rollback previous commit: SYS_open is not defined in certain environments.
PiperOrigin-RevId: 414521820
--
5a1cbb282331023902e1374dd0d920c4effbe47f by Abseil Team <absl-team@google.com>:
Use syscall(SYS_open, ...) instead of open() to avoid possible symbol
interposition.
Also add some warning notes.
PiperOrigin-RevId: 414508186
--
1824d6593612710aafdc599a89b0adced7d787f6 by Abseil Team <absl-team@google.com>:
Correct aarch64 macro check
The macro is __aarch64__, not __arch64__.
PiperOrigin-RevId: 414446225
--
a1536a57b64dfd53945d33a01cfc08b18c99c97b by Abseil Team <absl-team@google.com>:
Fix backwards comment in the last commit.
PiperOrigin-RevId: 414281214
--
11ac021ba779513667a31cf2563ddafc57d6d913 by Abseil Team <absl-team@google.com>:
AddressIsReadable() didn't work correctly on ARM when the given pointer was
misaligned at the end of the page.
Fix that by aligning the pointer on an 8-byte boundary before checking it.
PiperOrigin-RevId: 414203863
GitOrigin-RevId: 07240ca7822d007cdcc79f2c40bd58b2c2010348
Change-Id: If5f129194d59f5c9e5d84efd8cd9e17a70e072ab
3 years ago
|
|
|
inline void Cord::InlineRep::set_data(const char* data, size_t n) {
|
|
|
|
static_assert(kMaxInline == 15, "set_data is hard-coded for a length of 15");
|
|
|
|
|
Export of internal Abseil changes
--
07240ca7822d007cdcc79f2c40bd58b2c2010348 by Abseil Team <absl-team@google.com>:
Correct the comment from "AlphaNum" to "Arg".
PiperOrigin-RevId: 416139192
--
adcba4a6b3763626e1db7b1e8c108b3114903557 by Martijn Vels <mvels@google.com>:
Fix NewExternalRep() to require data being non-empty, and remove nullptr return.
PiperOrigin-RevId: 416135865
--
c0d14cd918fb16f15d1d84de9284b5c5ecc1f8f2 by Abseil Team <absl-team@google.com>:
Fix doc comment for absl::ascii_isprint().
The comment was incorrectly saying that it includes all whitespace.
It doesn't; the only whitespace char it includes is ' '.
PiperOrigin-RevId: 416112524
--
d83327800159c07002b6865e21232a12463e02dd by Abseil Team <absl-team@google.com>:
Internal change
PiperOrigin-RevId: 416099978
--
baf11e9ca42ca9140cdbf8075f971db8d65b1195 by Ilya Tokar <tokarip@google.com>:
Prevent compiler from optimizing Group_Match* benchmarks away.
Currently we benchmark single store of precomputed value.
Not all affected benchmarks show performance changes:
BM_Group_Match 0.53ns ± 1% 0.53ns ± 0% -0.42% (p=0.038 n=10+10)
BM_Group_MatchEmpty 0.26ns ± 1% 0.26ns ± 1% ~ (p=1.000 n=10+10)
BM_Group_MatchEmptyOrDeleted 0.26ns ± 1% 0.26ns ± 1% ~ (p=0.121 n=10+10)
BM_Group_CountLeadingEmptyOrDeleted 0.26ns ± 1% 0.45ns ± 0% +70.05% (p=0.000 n=10+8)
BM_Group_MatchFirstEmptyOrDeleted 0.26ns ± 0% 0.44ns ± 1% +65.91% (p=0.000 n=8+9)
But inspecting the generated code shows the difference,
e. g. BM_Group_MatchFirstEmptyOrDeleted
Before:
add $0xffffffffffffffff,%rbx
jne 30
After:
pcmpeqd %xmm0,%xmm0
pcmpgtb -0x30(%rbp),%xmm0
pmovmskb %xmm0,%eax
add: 0x23$0xffffffffffffffff,%rbx
jne 40
PiperOrigin-RevId: 416083515
--
122fbff893dc4571b3e75e4b241eb4495b925610 by Abseil Team <absl-team@google.com>:
Put namespace guard in ABSL_DECLARE_FLAG to make declaring a flag in a namespace a compiler error instead of a linker error.
PiperOrigin-RevId: 416036072
--
020fd8a20f5fa319e948846e003391fcb9e03868 by Ilya Tokar <tokarip@google.com>:
Make Cord::InlineRep::set_data unconditionally zero out memory.
Currently there is a single case where we don't zero out memory
as an optimization. Unconditional zeroing doesn't show any changes
in benchmarks, except for the unrelated improvement:
BM_CordPartialCopyToCord/1M/1 12.6ns ± 4% 12.6ns ± 4% ~ (p=0.857 n=16+19)
BM_CordPartialCopyToCord/1M/128 44.9ns ± 7% 45.0ns ± 3% ~ (p=0.468 n=18+17)
BM_CordPartialCopyToCord/1M/1k 64.5ns ± 4% 61.4ns ± 4% -4.82% (p=0.000 n=19+17)
BM_CordPartialCopyToCord/1M/8k 139ns ± 3% 128ns ±15% -7.76% (p=0.009 n=17+20)
BM_CordPartialCopyToCord/1M/16k 193ns ± 6% 168ns ± 6% -13.17% (p=0.000 n=17+17)
BM_CordPartialCopyToCord/4M/16k 199ns ± 4% 177ns ± 4% -11.36% (p=0.000 n=17+18)
BM_CordPartialCopyToCord/4M/32k 275ns ± 3% 250ns ± 4% -9.00% (p=0.000 n=18+18)
BM_CordPartialCopyToCord/4M/64k 291ns ± 4% 266ns ± 5% -8.53% (p=0.000 n=18+16)
BM_CordPartialCopyToCord/4M/128k 322ns ± 5% 291ns ± 4% -9.43% (p=0.000 n=20+18)
BM_CordPartialCopyToCord/8M/32k 281ns ± 5% 251ns ± 4% -10.38% (p=0.000 n=20+16)
BM_CordPartialCopyToCord/8M/64k 293ns ± 6% 267ns ± 4% -8.87% (p=0.000 n=16+19)
BM_CordPartialCopyToCord/8M/128k 334ns ± 3% 305ns ± 2% -8.56% (p=0.000 n=17+16)
This is clearly an alignmnet effect since number of the executed instructions is the same:
M_CordPartialCopyToCord/1M/1 155 ± 0% 155 ± 0% ~ (all samples are equal)
BM_CordPartialCopyToCord/1M/128 446 ± 0% 446 ± 0% ~ (p=0.332 n=36+39)
BM_CordPartialCopyToCord/1M/1k 473 ± 0% 473 ± 0% ~ (p=0.969 n=40+40)
BM_CordPartialCopyToCord/1M/8k 808 ± 0% 808 ± 0% ~ (p=0.127 n=40+39)
BM_CordPartialCopyToCord/1M/16k 957 ± 0% 957 ± 0% ~ (p=0.532 n=40+40)
BM_CordPartialCopyToCord/4M/16k 952 ± 0% 952 ± 0% ~ (p=0.686 n=39+39)
BM_CordPartialCopyToCord/4M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.690 n=40+40)
BM_CordPartialCopyToCord/4M/64k 1.23k ± 0% 1.23k ± 0% ~ (p=0.182 n=40+39)
BM_CordPartialCopyToCord/4M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.711 n=40+40)
BM_CordPartialCopyToCord/8M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.697 n=40+40)
BM_CordPartialCopyToCord/8M/64k 1.23k ± 0% 1.23k ± 0% +0.00% (p=0.049 n=40+40)
BM_CordPartialCopyToCord/8M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.507 n=40+40)
This makes code simpler and doesn't regress performance.
PiperOrigin-RevId: 415560574
--
37305b2690b31682088749e4d62f40d7095bdc54 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 415558737
--
86aaed569b9e743c1eb813a5f48def978a793db3 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415515201
--
6cdb8786cdcb4fa0b8a4b72fc98940877d1fdeff by Abseil Team <absl-team@google.com>:
Update SubmitMutexProfileData to accept wait_cycles instead of wait_timestamp
PiperOrigin-RevId: 415360871
--
9f979d307aa16ad09f214e04876cbe84395c0901 by Abseil Team <absl-team@google.com>:
absl::flat_hash_set compiles with -Wconversion -Wsign-compare
PiperOrigin-RevId: 415357498
--
9eceb14174708f15e61259d449b214a8a4c7f9e7 by Abseil Team <absl-team@google.com>:
Fix AddressIsReadable for the corner case of (aligned) addr == NULL.
PiperOrigin-RevId: 415307792
--
1a39ffe55898375e2d7f88c17c99db5a1b95b313 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415162872
--
64378549b110d5f5762185a5906c520fba70f0e7 by Abseil Team <absl-team@google.com>:
Fix a typo in the comments
PiperOrigin-RevId: 415088461
--
41aae8322e913b82710153c22b97c611fdb6e1fb by Abseil Team <absl-team@google.com>:
Switch from `connect` to `rt_sigreturn` -- the latter is much less problematic
for system call sandboxes.
PiperOrigin-RevId: 415073965
--
870c5e3388b6a35611bff538626fe7a1c8c87171 by Abseil Team <absl-team@google.com>:
Add ABSL_HAVE_HWADDRESS_SANITIZER and ABSL_HAVE_LEAK_SANITIZER
PiperOrigin-RevId: 414871189
--
f213ed60a66b58da7ac40555adfb1d529ff0a4db by Derek Mauro <dmauro@google.com>:
Remove reference to __SANITIZE_MEMORY__, which does not exist
It appears to have been copied by pattern matching from the ASAN/TSAN
code blocks.
https://github.com/gcc-mirror/gcc/blob/f47662204de27f7685699eeef89aa173ccf32d85/gcc/cppbuiltin.c#L79-L126
PiperOrigin-RevId: 414806587
--
b152891e73ab515f397ceb53f66c8ee2f33863ea by Abseil Team <absl-team@google.com>:
Rollback previous commit: SYS_open is not defined in certain environments.
PiperOrigin-RevId: 414521820
--
5a1cbb282331023902e1374dd0d920c4effbe47f by Abseil Team <absl-team@google.com>:
Use syscall(SYS_open, ...) instead of open() to avoid possible symbol
interposition.
Also add some warning notes.
PiperOrigin-RevId: 414508186
--
1824d6593612710aafdc599a89b0adced7d787f6 by Abseil Team <absl-team@google.com>:
Correct aarch64 macro check
The macro is __aarch64__, not __arch64__.
PiperOrigin-RevId: 414446225
--
a1536a57b64dfd53945d33a01cfc08b18c99c97b by Abseil Team <absl-team@google.com>:
Fix backwards comment in the last commit.
PiperOrigin-RevId: 414281214
--
11ac021ba779513667a31cf2563ddafc57d6d913 by Abseil Team <absl-team@google.com>:
AddressIsReadable() didn't work correctly on ARM when the given pointer was
misaligned at the end of the page.
Fix that by aligning the pointer on an 8-byte boundary before checking it.
PiperOrigin-RevId: 414203863
GitOrigin-RevId: 07240ca7822d007cdcc79f2c40bd58b2c2010348
Change-Id: If5f129194d59f5c9e5d84efd8cd9e17a70e072ab
3 years ago
|
|
|
cord_internal::SmallMemmove<true>(data_.as_chars(), data, n);
|
|
|
|
set_inline_size(n);
|
|
|
|
}
|
|
|
|
|
|
|
|
inline char* Cord::InlineRep::set_data(size_t n) {
|
|
|
|
assert(n <= kMaxInline);
|
|
|
|
ResetToEmpty();
|
|
|
|
set_inline_size(n);
|
|
|
|
return data_.as_chars();
|
|
|
|
}
|
|
|
|
|
|
|
|
inline void Cord::InlineRep::reduce_size(size_t n) {
|
|
|
|
size_t tag = inline_size();
|
|
|
|
assert(tag <= kMaxInline);
|
|
|
|
assert(tag >= n);
|
|
|
|
tag -= n;
|
|
|
|
memset(data_.as_chars() + tag, 0, n);
|
|
|
|
set_inline_size(tag);
|
|
|
|
}
|
|
|
|
|
|
|
|
inline void Cord::InlineRep::remove_prefix(size_t n) {
|
|
|
|
cord_internal::SmallMemmove(data_.as_chars(), data_.as_chars() + n,
|
|
|
|
inline_size() - n);
|
|
|
|
reduce_size(n);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Returns `rep` converted into a CordRepBtree.
|
|
|
|
// Directly returns `rep` if `rep` is already a CordRepBtree.
|
|
|
|
static CordRepBtree* ForceBtree(CordRep* rep) {
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
return rep->IsBtree()
|
|
|
|
? rep->btree()
|
|
|
|
: CordRepBtree::Create(cord_internal::RemoveCrcNode(rep));
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::InlineRep::AppendTreeToInlined(CordRep* tree,
|
|
|
|
MethodIdentifier method) {
|
|
|
|
assert(!is_tree());
|
|
|
|
if (!data_.is_empty()) {
|
|
|
|
CordRepFlat* flat = MakeFlatWithExtraCapacity(0);
|
|
|
|
tree = CordRepBtree::Append(CordRepBtree::Create(flat), tree);
|
|
|
|
}
|
|
|
|
EmplaceTree(tree, method);
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::InlineRep::AppendTreeToTree(CordRep* tree, MethodIdentifier method) {
|
|
|
|
assert(is_tree());
|
|
|
|
const CordzUpdateScope scope(data_.cordz_info(), method);
|
|
|
|
tree = CordRepBtree::Append(ForceBtree(data_.as_tree()), tree);
|
|
|
|
SetTree(tree, scope);
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::InlineRep::AppendTree(CordRep* tree, MethodIdentifier method) {
|
Export of internal Abseil changes
--
07240ca7822d007cdcc79f2c40bd58b2c2010348 by Abseil Team <absl-team@google.com>:
Correct the comment from "AlphaNum" to "Arg".
PiperOrigin-RevId: 416139192
--
adcba4a6b3763626e1db7b1e8c108b3114903557 by Martijn Vels <mvels@google.com>:
Fix NewExternalRep() to require data being non-empty, and remove nullptr return.
PiperOrigin-RevId: 416135865
--
c0d14cd918fb16f15d1d84de9284b5c5ecc1f8f2 by Abseil Team <absl-team@google.com>:
Fix doc comment for absl::ascii_isprint().
The comment was incorrectly saying that it includes all whitespace.
It doesn't; the only whitespace char it includes is ' '.
PiperOrigin-RevId: 416112524
--
d83327800159c07002b6865e21232a12463e02dd by Abseil Team <absl-team@google.com>:
Internal change
PiperOrigin-RevId: 416099978
--
baf11e9ca42ca9140cdbf8075f971db8d65b1195 by Ilya Tokar <tokarip@google.com>:
Prevent compiler from optimizing Group_Match* benchmarks away.
Currently we benchmark single store of precomputed value.
Not all affected benchmarks show performance changes:
BM_Group_Match 0.53ns ± 1% 0.53ns ± 0% -0.42% (p=0.038 n=10+10)
BM_Group_MatchEmpty 0.26ns ± 1% 0.26ns ± 1% ~ (p=1.000 n=10+10)
BM_Group_MatchEmptyOrDeleted 0.26ns ± 1% 0.26ns ± 1% ~ (p=0.121 n=10+10)
BM_Group_CountLeadingEmptyOrDeleted 0.26ns ± 1% 0.45ns ± 0% +70.05% (p=0.000 n=10+8)
BM_Group_MatchFirstEmptyOrDeleted 0.26ns ± 0% 0.44ns ± 1% +65.91% (p=0.000 n=8+9)
But inspecting the generated code shows the difference,
e. g. BM_Group_MatchFirstEmptyOrDeleted
Before:
add $0xffffffffffffffff,%rbx
jne 30
After:
pcmpeqd %xmm0,%xmm0
pcmpgtb -0x30(%rbp),%xmm0
pmovmskb %xmm0,%eax
add: 0x23$0xffffffffffffffff,%rbx
jne 40
PiperOrigin-RevId: 416083515
--
122fbff893dc4571b3e75e4b241eb4495b925610 by Abseil Team <absl-team@google.com>:
Put namespace guard in ABSL_DECLARE_FLAG to make declaring a flag in a namespace a compiler error instead of a linker error.
PiperOrigin-RevId: 416036072
--
020fd8a20f5fa319e948846e003391fcb9e03868 by Ilya Tokar <tokarip@google.com>:
Make Cord::InlineRep::set_data unconditionally zero out memory.
Currently there is a single case where we don't zero out memory
as an optimization. Unconditional zeroing doesn't show any changes
in benchmarks, except for the unrelated improvement:
BM_CordPartialCopyToCord/1M/1 12.6ns ± 4% 12.6ns ± 4% ~ (p=0.857 n=16+19)
BM_CordPartialCopyToCord/1M/128 44.9ns ± 7% 45.0ns ± 3% ~ (p=0.468 n=18+17)
BM_CordPartialCopyToCord/1M/1k 64.5ns ± 4% 61.4ns ± 4% -4.82% (p=0.000 n=19+17)
BM_CordPartialCopyToCord/1M/8k 139ns ± 3% 128ns ±15% -7.76% (p=0.009 n=17+20)
BM_CordPartialCopyToCord/1M/16k 193ns ± 6% 168ns ± 6% -13.17% (p=0.000 n=17+17)
BM_CordPartialCopyToCord/4M/16k 199ns ± 4% 177ns ± 4% -11.36% (p=0.000 n=17+18)
BM_CordPartialCopyToCord/4M/32k 275ns ± 3% 250ns ± 4% -9.00% (p=0.000 n=18+18)
BM_CordPartialCopyToCord/4M/64k 291ns ± 4% 266ns ± 5% -8.53% (p=0.000 n=18+16)
BM_CordPartialCopyToCord/4M/128k 322ns ± 5% 291ns ± 4% -9.43% (p=0.000 n=20+18)
BM_CordPartialCopyToCord/8M/32k 281ns ± 5% 251ns ± 4% -10.38% (p=0.000 n=20+16)
BM_CordPartialCopyToCord/8M/64k 293ns ± 6% 267ns ± 4% -8.87% (p=0.000 n=16+19)
BM_CordPartialCopyToCord/8M/128k 334ns ± 3% 305ns ± 2% -8.56% (p=0.000 n=17+16)
This is clearly an alignmnet effect since number of the executed instructions is the same:
M_CordPartialCopyToCord/1M/1 155 ± 0% 155 ± 0% ~ (all samples are equal)
BM_CordPartialCopyToCord/1M/128 446 ± 0% 446 ± 0% ~ (p=0.332 n=36+39)
BM_CordPartialCopyToCord/1M/1k 473 ± 0% 473 ± 0% ~ (p=0.969 n=40+40)
BM_CordPartialCopyToCord/1M/8k 808 ± 0% 808 ± 0% ~ (p=0.127 n=40+39)
BM_CordPartialCopyToCord/1M/16k 957 ± 0% 957 ± 0% ~ (p=0.532 n=40+40)
BM_CordPartialCopyToCord/4M/16k 952 ± 0% 952 ± 0% ~ (p=0.686 n=39+39)
BM_CordPartialCopyToCord/4M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.690 n=40+40)
BM_CordPartialCopyToCord/4M/64k 1.23k ± 0% 1.23k ± 0% ~ (p=0.182 n=40+39)
BM_CordPartialCopyToCord/4M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.711 n=40+40)
BM_CordPartialCopyToCord/8M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.697 n=40+40)
BM_CordPartialCopyToCord/8M/64k 1.23k ± 0% 1.23k ± 0% +0.00% (p=0.049 n=40+40)
BM_CordPartialCopyToCord/8M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.507 n=40+40)
This makes code simpler and doesn't regress performance.
PiperOrigin-RevId: 415560574
--
37305b2690b31682088749e4d62f40d7095bdc54 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 415558737
--
86aaed569b9e743c1eb813a5f48def978a793db3 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415515201
--
6cdb8786cdcb4fa0b8a4b72fc98940877d1fdeff by Abseil Team <absl-team@google.com>:
Update SubmitMutexProfileData to accept wait_cycles instead of wait_timestamp
PiperOrigin-RevId: 415360871
--
9f979d307aa16ad09f214e04876cbe84395c0901 by Abseil Team <absl-team@google.com>:
absl::flat_hash_set compiles with -Wconversion -Wsign-compare
PiperOrigin-RevId: 415357498
--
9eceb14174708f15e61259d449b214a8a4c7f9e7 by Abseil Team <absl-team@google.com>:
Fix AddressIsReadable for the corner case of (aligned) addr == NULL.
PiperOrigin-RevId: 415307792
--
1a39ffe55898375e2d7f88c17c99db5a1b95b313 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415162872
--
64378549b110d5f5762185a5906c520fba70f0e7 by Abseil Team <absl-team@google.com>:
Fix a typo in the comments
PiperOrigin-RevId: 415088461
--
41aae8322e913b82710153c22b97c611fdb6e1fb by Abseil Team <absl-team@google.com>:
Switch from `connect` to `rt_sigreturn` -- the latter is much less problematic
for system call sandboxes.
PiperOrigin-RevId: 415073965
--
870c5e3388b6a35611bff538626fe7a1c8c87171 by Abseil Team <absl-team@google.com>:
Add ABSL_HAVE_HWADDRESS_SANITIZER and ABSL_HAVE_LEAK_SANITIZER
PiperOrigin-RevId: 414871189
--
f213ed60a66b58da7ac40555adfb1d529ff0a4db by Derek Mauro <dmauro@google.com>:
Remove reference to __SANITIZE_MEMORY__, which does not exist
It appears to have been copied by pattern matching from the ASAN/TSAN
code blocks.
https://github.com/gcc-mirror/gcc/blob/f47662204de27f7685699eeef89aa173ccf32d85/gcc/cppbuiltin.c#L79-L126
PiperOrigin-RevId: 414806587
--
b152891e73ab515f397ceb53f66c8ee2f33863ea by Abseil Team <absl-team@google.com>:
Rollback previous commit: SYS_open is not defined in certain environments.
PiperOrigin-RevId: 414521820
--
5a1cbb282331023902e1374dd0d920c4effbe47f by Abseil Team <absl-team@google.com>:
Use syscall(SYS_open, ...) instead of open() to avoid possible symbol
interposition.
Also add some warning notes.
PiperOrigin-RevId: 414508186
--
1824d6593612710aafdc599a89b0adced7d787f6 by Abseil Team <absl-team@google.com>:
Correct aarch64 macro check
The macro is __aarch64__, not __arch64__.
PiperOrigin-RevId: 414446225
--
a1536a57b64dfd53945d33a01cfc08b18c99c97b by Abseil Team <absl-team@google.com>:
Fix backwards comment in the last commit.
PiperOrigin-RevId: 414281214
--
11ac021ba779513667a31cf2563ddafc57d6d913 by Abseil Team <absl-team@google.com>:
AddressIsReadable() didn't work correctly on ARM when the given pointer was
misaligned at the end of the page.
Fix that by aligning the pointer on an 8-byte boundary before checking it.
PiperOrigin-RevId: 414203863
GitOrigin-RevId: 07240ca7822d007cdcc79f2c40bd58b2c2010348
Change-Id: If5f129194d59f5c9e5d84efd8cd9e17a70e072ab
3 years ago
|
|
|
assert(tree != nullptr);
|
|
|
|
assert(tree->length != 0);
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
assert(!tree->IsCrc());
|
|
|
|
if (data_.is_tree()) {
|
|
|
|
AppendTreeToTree(tree, method);
|
|
|
|
} else {
|
|
|
|
AppendTreeToInlined(tree, method);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::InlineRep::PrependTreeToInlined(CordRep* tree,
|
|
|
|
MethodIdentifier method) {
|
|
|
|
assert(!is_tree());
|
|
|
|
if (!data_.is_empty()) {
|
|
|
|
CordRepFlat* flat = MakeFlatWithExtraCapacity(0);
|
|
|
|
tree = CordRepBtree::Prepend(CordRepBtree::Create(flat), tree);
|
|
|
|
}
|
|
|
|
EmplaceTree(tree, method);
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::InlineRep::PrependTreeToTree(CordRep* tree,
|
|
|
|
MethodIdentifier method) {
|
|
|
|
assert(is_tree());
|
|
|
|
const CordzUpdateScope scope(data_.cordz_info(), method);
|
|
|
|
tree = CordRepBtree::Prepend(ForceBtree(data_.as_tree()), tree);
|
|
|
|
SetTree(tree, scope);
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::InlineRep::PrependTree(CordRep* tree, MethodIdentifier method) {
|
|
|
|
assert(tree != nullptr);
|
Export of internal Abseil changes
--
07240ca7822d007cdcc79f2c40bd58b2c2010348 by Abseil Team <absl-team@google.com>:
Correct the comment from "AlphaNum" to "Arg".
PiperOrigin-RevId: 416139192
--
adcba4a6b3763626e1db7b1e8c108b3114903557 by Martijn Vels <mvels@google.com>:
Fix NewExternalRep() to require data being non-empty, and remove nullptr return.
PiperOrigin-RevId: 416135865
--
c0d14cd918fb16f15d1d84de9284b5c5ecc1f8f2 by Abseil Team <absl-team@google.com>:
Fix doc comment for absl::ascii_isprint().
The comment was incorrectly saying that it includes all whitespace.
It doesn't; the only whitespace char it includes is ' '.
PiperOrigin-RevId: 416112524
--
d83327800159c07002b6865e21232a12463e02dd by Abseil Team <absl-team@google.com>:
Internal change
PiperOrigin-RevId: 416099978
--
baf11e9ca42ca9140cdbf8075f971db8d65b1195 by Ilya Tokar <tokarip@google.com>:
Prevent compiler from optimizing Group_Match* benchmarks away.
Currently we benchmark single store of precomputed value.
Not all affected benchmarks show performance changes:
BM_Group_Match 0.53ns ± 1% 0.53ns ± 0% -0.42% (p=0.038 n=10+10)
BM_Group_MatchEmpty 0.26ns ± 1% 0.26ns ± 1% ~ (p=1.000 n=10+10)
BM_Group_MatchEmptyOrDeleted 0.26ns ± 1% 0.26ns ± 1% ~ (p=0.121 n=10+10)
BM_Group_CountLeadingEmptyOrDeleted 0.26ns ± 1% 0.45ns ± 0% +70.05% (p=0.000 n=10+8)
BM_Group_MatchFirstEmptyOrDeleted 0.26ns ± 0% 0.44ns ± 1% +65.91% (p=0.000 n=8+9)
But inspecting the generated code shows the difference,
e. g. BM_Group_MatchFirstEmptyOrDeleted
Before:
add $0xffffffffffffffff,%rbx
jne 30
After:
pcmpeqd %xmm0,%xmm0
pcmpgtb -0x30(%rbp),%xmm0
pmovmskb %xmm0,%eax
add: 0x23$0xffffffffffffffff,%rbx
jne 40
PiperOrigin-RevId: 416083515
--
122fbff893dc4571b3e75e4b241eb4495b925610 by Abseil Team <absl-team@google.com>:
Put namespace guard in ABSL_DECLARE_FLAG to make declaring a flag in a namespace a compiler error instead of a linker error.
PiperOrigin-RevId: 416036072
--
020fd8a20f5fa319e948846e003391fcb9e03868 by Ilya Tokar <tokarip@google.com>:
Make Cord::InlineRep::set_data unconditionally zero out memory.
Currently there is a single case where we don't zero out memory
as an optimization. Unconditional zeroing doesn't show any changes
in benchmarks, except for the unrelated improvement:
BM_CordPartialCopyToCord/1M/1 12.6ns ± 4% 12.6ns ± 4% ~ (p=0.857 n=16+19)
BM_CordPartialCopyToCord/1M/128 44.9ns ± 7% 45.0ns ± 3% ~ (p=0.468 n=18+17)
BM_CordPartialCopyToCord/1M/1k 64.5ns ± 4% 61.4ns ± 4% -4.82% (p=0.000 n=19+17)
BM_CordPartialCopyToCord/1M/8k 139ns ± 3% 128ns ±15% -7.76% (p=0.009 n=17+20)
BM_CordPartialCopyToCord/1M/16k 193ns ± 6% 168ns ± 6% -13.17% (p=0.000 n=17+17)
BM_CordPartialCopyToCord/4M/16k 199ns ± 4% 177ns ± 4% -11.36% (p=0.000 n=17+18)
BM_CordPartialCopyToCord/4M/32k 275ns ± 3% 250ns ± 4% -9.00% (p=0.000 n=18+18)
BM_CordPartialCopyToCord/4M/64k 291ns ± 4% 266ns ± 5% -8.53% (p=0.000 n=18+16)
BM_CordPartialCopyToCord/4M/128k 322ns ± 5% 291ns ± 4% -9.43% (p=0.000 n=20+18)
BM_CordPartialCopyToCord/8M/32k 281ns ± 5% 251ns ± 4% -10.38% (p=0.000 n=20+16)
BM_CordPartialCopyToCord/8M/64k 293ns ± 6% 267ns ± 4% -8.87% (p=0.000 n=16+19)
BM_CordPartialCopyToCord/8M/128k 334ns ± 3% 305ns ± 2% -8.56% (p=0.000 n=17+16)
This is clearly an alignmnet effect since number of the executed instructions is the same:
M_CordPartialCopyToCord/1M/1 155 ± 0% 155 ± 0% ~ (all samples are equal)
BM_CordPartialCopyToCord/1M/128 446 ± 0% 446 ± 0% ~ (p=0.332 n=36+39)
BM_CordPartialCopyToCord/1M/1k 473 ± 0% 473 ± 0% ~ (p=0.969 n=40+40)
BM_CordPartialCopyToCord/1M/8k 808 ± 0% 808 ± 0% ~ (p=0.127 n=40+39)
BM_CordPartialCopyToCord/1M/16k 957 ± 0% 957 ± 0% ~ (p=0.532 n=40+40)
BM_CordPartialCopyToCord/4M/16k 952 ± 0% 952 ± 0% ~ (p=0.686 n=39+39)
BM_CordPartialCopyToCord/4M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.690 n=40+40)
BM_CordPartialCopyToCord/4M/64k 1.23k ± 0% 1.23k ± 0% ~ (p=0.182 n=40+39)
BM_CordPartialCopyToCord/4M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.711 n=40+40)
BM_CordPartialCopyToCord/8M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.697 n=40+40)
BM_CordPartialCopyToCord/8M/64k 1.23k ± 0% 1.23k ± 0% +0.00% (p=0.049 n=40+40)
BM_CordPartialCopyToCord/8M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.507 n=40+40)
This makes code simpler and doesn't regress performance.
PiperOrigin-RevId: 415560574
--
37305b2690b31682088749e4d62f40d7095bdc54 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 415558737
--
86aaed569b9e743c1eb813a5f48def978a793db3 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415515201
--
6cdb8786cdcb4fa0b8a4b72fc98940877d1fdeff by Abseil Team <absl-team@google.com>:
Update SubmitMutexProfileData to accept wait_cycles instead of wait_timestamp
PiperOrigin-RevId: 415360871
--
9f979d307aa16ad09f214e04876cbe84395c0901 by Abseil Team <absl-team@google.com>:
absl::flat_hash_set compiles with -Wconversion -Wsign-compare
PiperOrigin-RevId: 415357498
--
9eceb14174708f15e61259d449b214a8a4c7f9e7 by Abseil Team <absl-team@google.com>:
Fix AddressIsReadable for the corner case of (aligned) addr == NULL.
PiperOrigin-RevId: 415307792
--
1a39ffe55898375e2d7f88c17c99db5a1b95b313 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415162872
--
64378549b110d5f5762185a5906c520fba70f0e7 by Abseil Team <absl-team@google.com>:
Fix a typo in the comments
PiperOrigin-RevId: 415088461
--
41aae8322e913b82710153c22b97c611fdb6e1fb by Abseil Team <absl-team@google.com>:
Switch from `connect` to `rt_sigreturn` -- the latter is much less problematic
for system call sandboxes.
PiperOrigin-RevId: 415073965
--
870c5e3388b6a35611bff538626fe7a1c8c87171 by Abseil Team <absl-team@google.com>:
Add ABSL_HAVE_HWADDRESS_SANITIZER and ABSL_HAVE_LEAK_SANITIZER
PiperOrigin-RevId: 414871189
--
f213ed60a66b58da7ac40555adfb1d529ff0a4db by Derek Mauro <dmauro@google.com>:
Remove reference to __SANITIZE_MEMORY__, which does not exist
It appears to have been copied by pattern matching from the ASAN/TSAN
code blocks.
https://github.com/gcc-mirror/gcc/blob/f47662204de27f7685699eeef89aa173ccf32d85/gcc/cppbuiltin.c#L79-L126
PiperOrigin-RevId: 414806587
--
b152891e73ab515f397ceb53f66c8ee2f33863ea by Abseil Team <absl-team@google.com>:
Rollback previous commit: SYS_open is not defined in certain environments.
PiperOrigin-RevId: 414521820
--
5a1cbb282331023902e1374dd0d920c4effbe47f by Abseil Team <absl-team@google.com>:
Use syscall(SYS_open, ...) instead of open() to avoid possible symbol
interposition.
Also add some warning notes.
PiperOrigin-RevId: 414508186
--
1824d6593612710aafdc599a89b0adced7d787f6 by Abseil Team <absl-team@google.com>:
Correct aarch64 macro check
The macro is __aarch64__, not __arch64__.
PiperOrigin-RevId: 414446225
--
a1536a57b64dfd53945d33a01cfc08b18c99c97b by Abseil Team <absl-team@google.com>:
Fix backwards comment in the last commit.
PiperOrigin-RevId: 414281214
--
11ac021ba779513667a31cf2563ddafc57d6d913 by Abseil Team <absl-team@google.com>:
AddressIsReadable() didn't work correctly on ARM when the given pointer was
misaligned at the end of the page.
Fix that by aligning the pointer on an 8-byte boundary before checking it.
PiperOrigin-RevId: 414203863
GitOrigin-RevId: 07240ca7822d007cdcc79f2c40bd58b2c2010348
Change-Id: If5f129194d59f5c9e5d84efd8cd9e17a70e072ab
3 years ago
|
|
|
assert(tree->length != 0);
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
assert(!tree->IsCrc());
|
|
|
|
if (data_.is_tree()) {
|
|
|
|
PrependTreeToTree(tree, method);
|
|
|
|
} else {
|
|
|
|
PrependTreeToInlined(tree, method);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Searches for a non-full flat node at the rightmost leaf of the tree. If a
|
|
|
|
// suitable leaf is found, the function will update the length field for all
|
|
|
|
// nodes to account for the size increase. The append region address will be
|
|
|
|
// written to region and the actual size increase will be written to size.
|
|
|
|
static inline bool PrepareAppendRegion(CordRep* root, char** region,
|
|
|
|
size_t* size, size_t max_length) {
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
if (root->IsBtree() && root->refcount.IsOne()) {
|
|
|
|
Span<char> span = root->btree()->GetAppendBuffer(max_length);
|
|
|
|
if (!span.empty()) {
|
|
|
|
*region = span.data();
|
|
|
|
*size = span.size();
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
CordRep* dst = root;
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
if (!dst->IsFlat() || !dst->refcount.IsOne()) {
|
|
|
|
*region = nullptr;
|
|
|
|
*size = 0;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
const size_t in_use = dst->length;
|
|
|
|
const size_t capacity = dst->flat()->Capacity();
|
|
|
|
if (in_use == capacity) {
|
|
|
|
*region = nullptr;
|
|
|
|
*size = 0;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
const size_t size_increase = std::min(capacity - in_use, max_length);
|
|
|
|
dst->length += size_increase;
|
|
|
|
|
|
|
|
*region = dst->flat()->Data() + in_use;
|
|
|
|
*size = size_increase;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::InlineRep::AssignSlow(const Cord::InlineRep& src) {
|
|
|
|
assert(&src != this);
|
|
|
|
assert(is_tree() || src.is_tree());
|
|
|
|
auto constexpr method = CordzUpdateTracker::kAssignCord;
|
|
|
|
if (ABSL_PREDICT_TRUE(!is_tree())) {
|
|
|
|
EmplaceTree(CordRep::Ref(src.as_tree()), src.data_, method);
|
|
|
|
return;
|
|
|
|
}
|
Export of internal Abseil changes
--
5f3c139695d5c497ca030e95a607537a7be7caa7 by Benjamin Barenblat <bbaren@google.com>:
Don’t examine irrelevant destination buckets in DiscreteDistributionTest
Abseil generates discrete distributions using Walker’s aliasing
algorithm. This creates uniformly distributed buckets, each with a
probability of sending traffic to a different bucket. Abseil represents
a bucket as a pair
(probability of retaining traffic ×
alternate bucket if traffic is passed)
and a distribution as a vector of such pairs. For example, {(0.3, 1),
(1.0, 1)} represents a distribution with two buckets, the zeroth of
which passes 70% of its traffic to bucket 1 and the first of which holds
on to all its traffic.
This representation is not unique: When a bucket retains traffic with
probability 1, the alternate bucket is irrelevant. Continuing the
example above, {(0.3, 1), (1.0, 0)} _also_ represents a two-bucket
distribution where the zeroth bucket passes 70% of its traffic to the
first and the first hangs on to all traffic. Exactly what representation
Abseil generates for a given input is related to how much precision is
used in intermediate floating-point operations, which is an
architectural implementation detail. Remove sensitivity to that detail
by not examining the alternate bucket when the retention probability is
1.0.
PiperOrigin-RevId: 372993410
--
062ac80699f748831c09a061538abffec2cdea5c by Martijn Vels <mvels@google.com>:
Avoid alredy sampled cord remaining sampled if not picked or source is sampled
PiperOrigin-RevId: 372985990
--
a9f3537e1110b7bb6450fd72a03f0c5dc6b8c89b by Evan Brown <ezb@google.com>:
Add tests for function pointer comparators, comparators that have SFINAE-visible comparison operators that are unimplemented, and for implicit construction from unadapted comparators.
PiperOrigin-RevId: 372927616
GitOrigin-RevId: 5f3c139695d5c497ca030e95a607537a7be7caa7
Change-Id: I996a8452e7bd88f9dd2e59633b01bbc09f42620d
4 years ago
|
|
|
|
|
|
|
CordRep* tree = as_tree();
|
|
|
|
if (CordRep* src_tree = src.tree()) {
|
Export of internal Abseil changes
--
9fc37c11b9e46287acef00ee06ed9adcba54dd13 by Greg Falcon <gfalcon@google.com>:
Rename absl::hash_internal::HashState to absl::hash_internal::MixingHashState.
Before this change, we had two classes named HashState: absl::HashState, the public API used for type erasure, and absl::hash_internal::HashState, the internal concrete implementation ordinarily used.
The internal class used to be named `CityHashState`, but we renamed it to `HashState` it when we changed underlying hash implementation to wyhash. This inadvertent naming conflict made the code much harder to read, and this change intends to undo that.
PiperOrigin-RevId: 373481959
--
4aec55ffddebd085c239352a2e20721091f719a1 by Greg Falcon <gfalcon@google.com>:
Introduce absl::HashOf(), a convenience wrapper around absl::Hash that calculates hashes from the values of its arguments.
PiperOrigin-RevId: 373461406
--
86b5fd8db50bbc8bd0aa9258523527381fe0445d by Abseil Team <absl-team@google.com>:
Improve speed of BlockingCounter by making its most common path lock free.
With the new implementation, the fast path of BlockingCounter::DecrementCount()
is only a fetch_sub operation. This is most times much more efficient than the
previous implementation (full mutex lock/unlock). As a matter of fact, in most
actual usecases in practice, the waiter thread is already waiting on the
Wait() call when DecrementCount() is called, which makes Mutex::Unlock() take
the slow path as there's a waiter thread that it might need to wake up.
PiperOrigin-RevId: 373394164
--
65c876be5eac0cd32583ff8535ede4109d39cf3f by Martijn Vels <mvels@google.com>:
Move the 'sample copied cord' logic into MaybeTrackCord(),
This changes move the logic for selecting if a cord should remain being sampled from Cord to CordzInfo::MaybeTrackCord, and updates the documentation for the latter method.
PiperOrigin-RevId: 373363168
--
e84410bd0aada293a81dfb82656c952e209e21fb by Martijn Vels <mvels@google.com>:
Add check for the first call to cordz_should_profile() for each thread.
This prevents the first cord of a newly created thread to be always sampled, which is a 'bad' kind of determinism for sampling.
PiperOrigin-RevId: 373229768
--
bf09c589dc099ac8f4af780bf7e609c53c27574c by Samuel Benzaquen <sbenza@google.com>:
Refactor the Flags structure into an enum.
This gives us more control over the representation and allows for easier
merging during parsing.
PiperOrigin-RevId: 373163038
--
b947b0c51083b7b6508284b5d31819596c91729e by Derek Mauro <dmauro@google.com>:
Fixes warnings about shadowed variables
Fixes #956
PiperOrigin-RevId: 373158133
GitOrigin-RevId: 9fc37c11b9e46287acef00ee06ed9adcba54dd13
Change-Id: I91f35699f9bf439d1a870c6493946a310afe088c
4 years ago
|
|
|
// Leave any existing `cordz_info` in place, and let MaybeTrackCord()
|
|
|
|
// decide if this cord should be (or remains to be) sampled or not.
|
|
|
|
data_.set_tree(CordRep::Ref(src_tree));
|
|
|
|
CordzInfo::MaybeTrackCord(data_, src.data_, method);
|
|
|
|
} else {
|
|
|
|
CordzInfo::MaybeUntrackCord(data_.cordz_info());
|
|
|
|
data_ = src.data_;
|
|
|
|
}
|
|
|
|
CordRep::Unref(tree);
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::InlineRep::UnrefTree() {
|
|
|
|
if (is_tree()) {
|
|
|
|
CordzInfo::MaybeUntrackCord(data_.cordz_info());
|
|
|
|
CordRep::Unref(tree());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// --------------------------------------------------------------------
|
|
|
|
// Constructors and destructors
|
|
|
|
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
Cord::Cord(absl::string_view src, MethodIdentifier method)
|
|
|
|
: contents_(InlineData::kDefaultInit) {
|
|
|
|
const size_t n = src.size();
|
|
|
|
if (n <= InlineRep::kMaxInline) {
|
Export of internal Abseil changes
--
07240ca7822d007cdcc79f2c40bd58b2c2010348 by Abseil Team <absl-team@google.com>:
Correct the comment from "AlphaNum" to "Arg".
PiperOrigin-RevId: 416139192
--
adcba4a6b3763626e1db7b1e8c108b3114903557 by Martijn Vels <mvels@google.com>:
Fix NewExternalRep() to require data being non-empty, and remove nullptr return.
PiperOrigin-RevId: 416135865
--
c0d14cd918fb16f15d1d84de9284b5c5ecc1f8f2 by Abseil Team <absl-team@google.com>:
Fix doc comment for absl::ascii_isprint().
The comment was incorrectly saying that it includes all whitespace.
It doesn't; the only whitespace char it includes is ' '.
PiperOrigin-RevId: 416112524
--
d83327800159c07002b6865e21232a12463e02dd by Abseil Team <absl-team@google.com>:
Internal change
PiperOrigin-RevId: 416099978
--
baf11e9ca42ca9140cdbf8075f971db8d65b1195 by Ilya Tokar <tokarip@google.com>:
Prevent compiler from optimizing Group_Match* benchmarks away.
Currently we benchmark single store of precomputed value.
Not all affected benchmarks show performance changes:
BM_Group_Match 0.53ns ± 1% 0.53ns ± 0% -0.42% (p=0.038 n=10+10)
BM_Group_MatchEmpty 0.26ns ± 1% 0.26ns ± 1% ~ (p=1.000 n=10+10)
BM_Group_MatchEmptyOrDeleted 0.26ns ± 1% 0.26ns ± 1% ~ (p=0.121 n=10+10)
BM_Group_CountLeadingEmptyOrDeleted 0.26ns ± 1% 0.45ns ± 0% +70.05% (p=0.000 n=10+8)
BM_Group_MatchFirstEmptyOrDeleted 0.26ns ± 0% 0.44ns ± 1% +65.91% (p=0.000 n=8+9)
But inspecting the generated code shows the difference,
e. g. BM_Group_MatchFirstEmptyOrDeleted
Before:
add $0xffffffffffffffff,%rbx
jne 30
After:
pcmpeqd %xmm0,%xmm0
pcmpgtb -0x30(%rbp),%xmm0
pmovmskb %xmm0,%eax
add: 0x23$0xffffffffffffffff,%rbx
jne 40
PiperOrigin-RevId: 416083515
--
122fbff893dc4571b3e75e4b241eb4495b925610 by Abseil Team <absl-team@google.com>:
Put namespace guard in ABSL_DECLARE_FLAG to make declaring a flag in a namespace a compiler error instead of a linker error.
PiperOrigin-RevId: 416036072
--
020fd8a20f5fa319e948846e003391fcb9e03868 by Ilya Tokar <tokarip@google.com>:
Make Cord::InlineRep::set_data unconditionally zero out memory.
Currently there is a single case where we don't zero out memory
as an optimization. Unconditional zeroing doesn't show any changes
in benchmarks, except for the unrelated improvement:
BM_CordPartialCopyToCord/1M/1 12.6ns ± 4% 12.6ns ± 4% ~ (p=0.857 n=16+19)
BM_CordPartialCopyToCord/1M/128 44.9ns ± 7% 45.0ns ± 3% ~ (p=0.468 n=18+17)
BM_CordPartialCopyToCord/1M/1k 64.5ns ± 4% 61.4ns ± 4% -4.82% (p=0.000 n=19+17)
BM_CordPartialCopyToCord/1M/8k 139ns ± 3% 128ns ±15% -7.76% (p=0.009 n=17+20)
BM_CordPartialCopyToCord/1M/16k 193ns ± 6% 168ns ± 6% -13.17% (p=0.000 n=17+17)
BM_CordPartialCopyToCord/4M/16k 199ns ± 4% 177ns ± 4% -11.36% (p=0.000 n=17+18)
BM_CordPartialCopyToCord/4M/32k 275ns ± 3% 250ns ± 4% -9.00% (p=0.000 n=18+18)
BM_CordPartialCopyToCord/4M/64k 291ns ± 4% 266ns ± 5% -8.53% (p=0.000 n=18+16)
BM_CordPartialCopyToCord/4M/128k 322ns ± 5% 291ns ± 4% -9.43% (p=0.000 n=20+18)
BM_CordPartialCopyToCord/8M/32k 281ns ± 5% 251ns ± 4% -10.38% (p=0.000 n=20+16)
BM_CordPartialCopyToCord/8M/64k 293ns ± 6% 267ns ± 4% -8.87% (p=0.000 n=16+19)
BM_CordPartialCopyToCord/8M/128k 334ns ± 3% 305ns ± 2% -8.56% (p=0.000 n=17+16)
This is clearly an alignmnet effect since number of the executed instructions is the same:
M_CordPartialCopyToCord/1M/1 155 ± 0% 155 ± 0% ~ (all samples are equal)
BM_CordPartialCopyToCord/1M/128 446 ± 0% 446 ± 0% ~ (p=0.332 n=36+39)
BM_CordPartialCopyToCord/1M/1k 473 ± 0% 473 ± 0% ~ (p=0.969 n=40+40)
BM_CordPartialCopyToCord/1M/8k 808 ± 0% 808 ± 0% ~ (p=0.127 n=40+39)
BM_CordPartialCopyToCord/1M/16k 957 ± 0% 957 ± 0% ~ (p=0.532 n=40+40)
BM_CordPartialCopyToCord/4M/16k 952 ± 0% 952 ± 0% ~ (p=0.686 n=39+39)
BM_CordPartialCopyToCord/4M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.690 n=40+40)
BM_CordPartialCopyToCord/4M/64k 1.23k ± 0% 1.23k ± 0% ~ (p=0.182 n=40+39)
BM_CordPartialCopyToCord/4M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.711 n=40+40)
BM_CordPartialCopyToCord/8M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.697 n=40+40)
BM_CordPartialCopyToCord/8M/64k 1.23k ± 0% 1.23k ± 0% +0.00% (p=0.049 n=40+40)
BM_CordPartialCopyToCord/8M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.507 n=40+40)
This makes code simpler and doesn't regress performance.
PiperOrigin-RevId: 415560574
--
37305b2690b31682088749e4d62f40d7095bdc54 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 415558737
--
86aaed569b9e743c1eb813a5f48def978a793db3 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415515201
--
6cdb8786cdcb4fa0b8a4b72fc98940877d1fdeff by Abseil Team <absl-team@google.com>:
Update SubmitMutexProfileData to accept wait_cycles instead of wait_timestamp
PiperOrigin-RevId: 415360871
--
9f979d307aa16ad09f214e04876cbe84395c0901 by Abseil Team <absl-team@google.com>:
absl::flat_hash_set compiles with -Wconversion -Wsign-compare
PiperOrigin-RevId: 415357498
--
9eceb14174708f15e61259d449b214a8a4c7f9e7 by Abseil Team <absl-team@google.com>:
Fix AddressIsReadable for the corner case of (aligned) addr == NULL.
PiperOrigin-RevId: 415307792
--
1a39ffe55898375e2d7f88c17c99db5a1b95b313 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415162872
--
64378549b110d5f5762185a5906c520fba70f0e7 by Abseil Team <absl-team@google.com>:
Fix a typo in the comments
PiperOrigin-RevId: 415088461
--
41aae8322e913b82710153c22b97c611fdb6e1fb by Abseil Team <absl-team@google.com>:
Switch from `connect` to `rt_sigreturn` -- the latter is much less problematic
for system call sandboxes.
PiperOrigin-RevId: 415073965
--
870c5e3388b6a35611bff538626fe7a1c8c87171 by Abseil Team <absl-team@google.com>:
Add ABSL_HAVE_HWADDRESS_SANITIZER and ABSL_HAVE_LEAK_SANITIZER
PiperOrigin-RevId: 414871189
--
f213ed60a66b58da7ac40555adfb1d529ff0a4db by Derek Mauro <dmauro@google.com>:
Remove reference to __SANITIZE_MEMORY__, which does not exist
It appears to have been copied by pattern matching from the ASAN/TSAN
code blocks.
https://github.com/gcc-mirror/gcc/blob/f47662204de27f7685699eeef89aa173ccf32d85/gcc/cppbuiltin.c#L79-L126
PiperOrigin-RevId: 414806587
--
b152891e73ab515f397ceb53f66c8ee2f33863ea by Abseil Team <absl-team@google.com>:
Rollback previous commit: SYS_open is not defined in certain environments.
PiperOrigin-RevId: 414521820
--
5a1cbb282331023902e1374dd0d920c4effbe47f by Abseil Team <absl-team@google.com>:
Use syscall(SYS_open, ...) instead of open() to avoid possible symbol
interposition.
Also add some warning notes.
PiperOrigin-RevId: 414508186
--
1824d6593612710aafdc599a89b0adced7d787f6 by Abseil Team <absl-team@google.com>:
Correct aarch64 macro check
The macro is __aarch64__, not __arch64__.
PiperOrigin-RevId: 414446225
--
a1536a57b64dfd53945d33a01cfc08b18c99c97b by Abseil Team <absl-team@google.com>:
Fix backwards comment in the last commit.
PiperOrigin-RevId: 414281214
--
11ac021ba779513667a31cf2563ddafc57d6d913 by Abseil Team <absl-team@google.com>:
AddressIsReadable() didn't work correctly on ARM when the given pointer was
misaligned at the end of the page.
Fix that by aligning the pointer on an 8-byte boundary before checking it.
PiperOrigin-RevId: 414203863
GitOrigin-RevId: 07240ca7822d007cdcc79f2c40bd58b2c2010348
Change-Id: If5f129194d59f5c9e5d84efd8cd9e17a70e072ab
3 years ago
|
|
|
contents_.set_data(src.data(), n);
|
|
|
|
} else {
|
|
|
|
CordRep* rep = NewTree(src.data(), n, 0);
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
contents_.EmplaceTree(rep, method);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Export of internal Abseil changes
--
7d0468a6610ed85586d5c87fd65de8dac5118923 by Derek Mauro <dmauro@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 313226473
--
1131ef6d116f5ce7d46537a82f300ea06dcaaa53 by Gennadiy Rozental <rogeeff@google.com>:
Migrate internal interface to use mutable references.
PiperOrigin-RevId: 312931131
--
96225212a9f5fbd0b38c71fe65539164992c7c3b by Laramie Leavitt <lar@google.com>:
Remove random/internal/distributions.h
This file was something of an historical artifact. All of the related
code has either been removed or migraged, and so the only remaining type
belongs with uniform_helper.h, as it is used to infer the return type
of the absl::Uniform method in a few cases.
PiperOrigin-RevId: 312878173
--
6dcbd5be58ad425e08740ff64088373ee7fe4a72 by Mark Barolak <mbar@google.com>:
Release the StrFormat test case for Cords to open source.
PiperOrigin-RevId: 312707974
--
34484d18dfb63a0a7ad6e2aaeb570e33592968be by Abseil Team <absl-team@google.com>:
Let Cord::Cord(string&&), Cord::operator=(string&&),
Cord::Append(string&&), and Cord::Prepend(string&&) steal string data
and embed it into the Cord as a single external chunk, instead of
copying it into flat chunks (at most 4083-byte each).
Stealing string data is faster, but it creates a long chunk, which leads
to a higher more memory usage if its subcords are created and outlive
the whole Cord.
These functions revert to copying the data if any of the following
conditions holds:
- string size is at most kMaxBytesToCopy (511), to avoid the overhead
of an external chunk for short strings;
- less than half of string capacity is used, to avoid pinning to much
unused memory.
PiperOrigin-RevId: 312683785
GitOrigin-RevId: 7d0468a6610ed85586d5c87fd65de8dac5118923
Change-Id: If79b5a1dfe6d53a8ddddbc7da84338f11fc4cfa3
5 years ago
|
|
|
template <typename T, Cord::EnableIfString<T>>
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
Cord::Cord(T&& src) : contents_(InlineData::kDefaultInit) {
|
|
|
|
if (src.size() <= InlineRep::kMaxInline) {
|
Export of internal Abseil changes
--
07240ca7822d007cdcc79f2c40bd58b2c2010348 by Abseil Team <absl-team@google.com>:
Correct the comment from "AlphaNum" to "Arg".
PiperOrigin-RevId: 416139192
--
adcba4a6b3763626e1db7b1e8c108b3114903557 by Martijn Vels <mvels@google.com>:
Fix NewExternalRep() to require data being non-empty, and remove nullptr return.
PiperOrigin-RevId: 416135865
--
c0d14cd918fb16f15d1d84de9284b5c5ecc1f8f2 by Abseil Team <absl-team@google.com>:
Fix doc comment for absl::ascii_isprint().
The comment was incorrectly saying that it includes all whitespace.
It doesn't; the only whitespace char it includes is ' '.
PiperOrigin-RevId: 416112524
--
d83327800159c07002b6865e21232a12463e02dd by Abseil Team <absl-team@google.com>:
Internal change
PiperOrigin-RevId: 416099978
--
baf11e9ca42ca9140cdbf8075f971db8d65b1195 by Ilya Tokar <tokarip@google.com>:
Prevent compiler from optimizing Group_Match* benchmarks away.
Currently we benchmark single store of precomputed value.
Not all affected benchmarks show performance changes:
BM_Group_Match 0.53ns ± 1% 0.53ns ± 0% -0.42% (p=0.038 n=10+10)
BM_Group_MatchEmpty 0.26ns ± 1% 0.26ns ± 1% ~ (p=1.000 n=10+10)
BM_Group_MatchEmptyOrDeleted 0.26ns ± 1% 0.26ns ± 1% ~ (p=0.121 n=10+10)
BM_Group_CountLeadingEmptyOrDeleted 0.26ns ± 1% 0.45ns ± 0% +70.05% (p=0.000 n=10+8)
BM_Group_MatchFirstEmptyOrDeleted 0.26ns ± 0% 0.44ns ± 1% +65.91% (p=0.000 n=8+9)
But inspecting the generated code shows the difference,
e. g. BM_Group_MatchFirstEmptyOrDeleted
Before:
add $0xffffffffffffffff,%rbx
jne 30
After:
pcmpeqd %xmm0,%xmm0
pcmpgtb -0x30(%rbp),%xmm0
pmovmskb %xmm0,%eax
add: 0x23$0xffffffffffffffff,%rbx
jne 40
PiperOrigin-RevId: 416083515
--
122fbff893dc4571b3e75e4b241eb4495b925610 by Abseil Team <absl-team@google.com>:
Put namespace guard in ABSL_DECLARE_FLAG to make declaring a flag in a namespace a compiler error instead of a linker error.
PiperOrigin-RevId: 416036072
--
020fd8a20f5fa319e948846e003391fcb9e03868 by Ilya Tokar <tokarip@google.com>:
Make Cord::InlineRep::set_data unconditionally zero out memory.
Currently there is a single case where we don't zero out memory
as an optimization. Unconditional zeroing doesn't show any changes
in benchmarks, except for the unrelated improvement:
BM_CordPartialCopyToCord/1M/1 12.6ns ± 4% 12.6ns ± 4% ~ (p=0.857 n=16+19)
BM_CordPartialCopyToCord/1M/128 44.9ns ± 7% 45.0ns ± 3% ~ (p=0.468 n=18+17)
BM_CordPartialCopyToCord/1M/1k 64.5ns ± 4% 61.4ns ± 4% -4.82% (p=0.000 n=19+17)
BM_CordPartialCopyToCord/1M/8k 139ns ± 3% 128ns ±15% -7.76% (p=0.009 n=17+20)
BM_CordPartialCopyToCord/1M/16k 193ns ± 6% 168ns ± 6% -13.17% (p=0.000 n=17+17)
BM_CordPartialCopyToCord/4M/16k 199ns ± 4% 177ns ± 4% -11.36% (p=0.000 n=17+18)
BM_CordPartialCopyToCord/4M/32k 275ns ± 3% 250ns ± 4% -9.00% (p=0.000 n=18+18)
BM_CordPartialCopyToCord/4M/64k 291ns ± 4% 266ns ± 5% -8.53% (p=0.000 n=18+16)
BM_CordPartialCopyToCord/4M/128k 322ns ± 5% 291ns ± 4% -9.43% (p=0.000 n=20+18)
BM_CordPartialCopyToCord/8M/32k 281ns ± 5% 251ns ± 4% -10.38% (p=0.000 n=20+16)
BM_CordPartialCopyToCord/8M/64k 293ns ± 6% 267ns ± 4% -8.87% (p=0.000 n=16+19)
BM_CordPartialCopyToCord/8M/128k 334ns ± 3% 305ns ± 2% -8.56% (p=0.000 n=17+16)
This is clearly an alignmnet effect since number of the executed instructions is the same:
M_CordPartialCopyToCord/1M/1 155 ± 0% 155 ± 0% ~ (all samples are equal)
BM_CordPartialCopyToCord/1M/128 446 ± 0% 446 ± 0% ~ (p=0.332 n=36+39)
BM_CordPartialCopyToCord/1M/1k 473 ± 0% 473 ± 0% ~ (p=0.969 n=40+40)
BM_CordPartialCopyToCord/1M/8k 808 ± 0% 808 ± 0% ~ (p=0.127 n=40+39)
BM_CordPartialCopyToCord/1M/16k 957 ± 0% 957 ± 0% ~ (p=0.532 n=40+40)
BM_CordPartialCopyToCord/4M/16k 952 ± 0% 952 ± 0% ~ (p=0.686 n=39+39)
BM_CordPartialCopyToCord/4M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.690 n=40+40)
BM_CordPartialCopyToCord/4M/64k 1.23k ± 0% 1.23k ± 0% ~ (p=0.182 n=40+39)
BM_CordPartialCopyToCord/4M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.711 n=40+40)
BM_CordPartialCopyToCord/8M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.697 n=40+40)
BM_CordPartialCopyToCord/8M/64k 1.23k ± 0% 1.23k ± 0% +0.00% (p=0.049 n=40+40)
BM_CordPartialCopyToCord/8M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.507 n=40+40)
This makes code simpler and doesn't regress performance.
PiperOrigin-RevId: 415560574
--
37305b2690b31682088749e4d62f40d7095bdc54 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 415558737
--
86aaed569b9e743c1eb813a5f48def978a793db3 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415515201
--
6cdb8786cdcb4fa0b8a4b72fc98940877d1fdeff by Abseil Team <absl-team@google.com>:
Update SubmitMutexProfileData to accept wait_cycles instead of wait_timestamp
PiperOrigin-RevId: 415360871
--
9f979d307aa16ad09f214e04876cbe84395c0901 by Abseil Team <absl-team@google.com>:
absl::flat_hash_set compiles with -Wconversion -Wsign-compare
PiperOrigin-RevId: 415357498
--
9eceb14174708f15e61259d449b214a8a4c7f9e7 by Abseil Team <absl-team@google.com>:
Fix AddressIsReadable for the corner case of (aligned) addr == NULL.
PiperOrigin-RevId: 415307792
--
1a39ffe55898375e2d7f88c17c99db5a1b95b313 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415162872
--
64378549b110d5f5762185a5906c520fba70f0e7 by Abseil Team <absl-team@google.com>:
Fix a typo in the comments
PiperOrigin-RevId: 415088461
--
41aae8322e913b82710153c22b97c611fdb6e1fb by Abseil Team <absl-team@google.com>:
Switch from `connect` to `rt_sigreturn` -- the latter is much less problematic
for system call sandboxes.
PiperOrigin-RevId: 415073965
--
870c5e3388b6a35611bff538626fe7a1c8c87171 by Abseil Team <absl-team@google.com>:
Add ABSL_HAVE_HWADDRESS_SANITIZER and ABSL_HAVE_LEAK_SANITIZER
PiperOrigin-RevId: 414871189
--
f213ed60a66b58da7ac40555adfb1d529ff0a4db by Derek Mauro <dmauro@google.com>:
Remove reference to __SANITIZE_MEMORY__, which does not exist
It appears to have been copied by pattern matching from the ASAN/TSAN
code blocks.
https://github.com/gcc-mirror/gcc/blob/f47662204de27f7685699eeef89aa173ccf32d85/gcc/cppbuiltin.c#L79-L126
PiperOrigin-RevId: 414806587
--
b152891e73ab515f397ceb53f66c8ee2f33863ea by Abseil Team <absl-team@google.com>:
Rollback previous commit: SYS_open is not defined in certain environments.
PiperOrigin-RevId: 414521820
--
5a1cbb282331023902e1374dd0d920c4effbe47f by Abseil Team <absl-team@google.com>:
Use syscall(SYS_open, ...) instead of open() to avoid possible symbol
interposition.
Also add some warning notes.
PiperOrigin-RevId: 414508186
--
1824d6593612710aafdc599a89b0adced7d787f6 by Abseil Team <absl-team@google.com>:
Correct aarch64 macro check
The macro is __aarch64__, not __arch64__.
PiperOrigin-RevId: 414446225
--
a1536a57b64dfd53945d33a01cfc08b18c99c97b by Abseil Team <absl-team@google.com>:
Fix backwards comment in the last commit.
PiperOrigin-RevId: 414281214
--
11ac021ba779513667a31cf2563ddafc57d6d913 by Abseil Team <absl-team@google.com>:
AddressIsReadable() didn't work correctly on ARM when the given pointer was
misaligned at the end of the page.
Fix that by aligning the pointer on an 8-byte boundary before checking it.
PiperOrigin-RevId: 414203863
GitOrigin-RevId: 07240ca7822d007cdcc79f2c40bd58b2c2010348
Change-Id: If5f129194d59f5c9e5d84efd8cd9e17a70e072ab
3 years ago
|
|
|
contents_.set_data(src.data(), src.size());
|
Export of internal Abseil changes
--
7d0468a6610ed85586d5c87fd65de8dac5118923 by Derek Mauro <dmauro@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 313226473
--
1131ef6d116f5ce7d46537a82f300ea06dcaaa53 by Gennadiy Rozental <rogeeff@google.com>:
Migrate internal interface to use mutable references.
PiperOrigin-RevId: 312931131
--
96225212a9f5fbd0b38c71fe65539164992c7c3b by Laramie Leavitt <lar@google.com>:
Remove random/internal/distributions.h
This file was something of an historical artifact. All of the related
code has either been removed or migraged, and so the only remaining type
belongs with uniform_helper.h, as it is used to infer the return type
of the absl::Uniform method in a few cases.
PiperOrigin-RevId: 312878173
--
6dcbd5be58ad425e08740ff64088373ee7fe4a72 by Mark Barolak <mbar@google.com>:
Release the StrFormat test case for Cords to open source.
PiperOrigin-RevId: 312707974
--
34484d18dfb63a0a7ad6e2aaeb570e33592968be by Abseil Team <absl-team@google.com>:
Let Cord::Cord(string&&), Cord::operator=(string&&),
Cord::Append(string&&), and Cord::Prepend(string&&) steal string data
and embed it into the Cord as a single external chunk, instead of
copying it into flat chunks (at most 4083-byte each).
Stealing string data is faster, but it creates a long chunk, which leads
to a higher more memory usage if its subcords are created and outlive
the whole Cord.
These functions revert to copying the data if any of the following
conditions holds:
- string size is at most kMaxBytesToCopy (511), to avoid the overhead
of an external chunk for short strings;
- less than half of string capacity is used, to avoid pinning to much
unused memory.
PiperOrigin-RevId: 312683785
GitOrigin-RevId: 7d0468a6610ed85586d5c87fd65de8dac5118923
Change-Id: If79b5a1dfe6d53a8ddddbc7da84338f11fc4cfa3
5 years ago
|
|
|
} else {
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
CordRep* rep = CordRepFromString(std::forward<T>(src));
|
|
|
|
contents_.EmplaceTree(rep, CordzUpdateTracker::kConstructorString);
|
Export of internal Abseil changes
--
7d0468a6610ed85586d5c87fd65de8dac5118923 by Derek Mauro <dmauro@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 313226473
--
1131ef6d116f5ce7d46537a82f300ea06dcaaa53 by Gennadiy Rozental <rogeeff@google.com>:
Migrate internal interface to use mutable references.
PiperOrigin-RevId: 312931131
--
96225212a9f5fbd0b38c71fe65539164992c7c3b by Laramie Leavitt <lar@google.com>:
Remove random/internal/distributions.h
This file was something of an historical artifact. All of the related
code has either been removed or migraged, and so the only remaining type
belongs with uniform_helper.h, as it is used to infer the return type
of the absl::Uniform method in a few cases.
PiperOrigin-RevId: 312878173
--
6dcbd5be58ad425e08740ff64088373ee7fe4a72 by Mark Barolak <mbar@google.com>:
Release the StrFormat test case for Cords to open source.
PiperOrigin-RevId: 312707974
--
34484d18dfb63a0a7ad6e2aaeb570e33592968be by Abseil Team <absl-team@google.com>:
Let Cord::Cord(string&&), Cord::operator=(string&&),
Cord::Append(string&&), and Cord::Prepend(string&&) steal string data
and embed it into the Cord as a single external chunk, instead of
copying it into flat chunks (at most 4083-byte each).
Stealing string data is faster, but it creates a long chunk, which leads
to a higher more memory usage if its subcords are created and outlive
the whole Cord.
These functions revert to copying the data if any of the following
conditions holds:
- string size is at most kMaxBytesToCopy (511), to avoid the overhead
of an external chunk for short strings;
- less than half of string capacity is used, to avoid pinning to much
unused memory.
PiperOrigin-RevId: 312683785
GitOrigin-RevId: 7d0468a6610ed85586d5c87fd65de8dac5118923
Change-Id: If79b5a1dfe6d53a8ddddbc7da84338f11fc4cfa3
5 years ago
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
template Cord::Cord(std::string&& src);
|
|
|
|
|
|
|
|
// The destruction code is separate so that the compiler can determine
|
|
|
|
// that it does not need to call the destructor on a moved-from Cord.
|
|
|
|
void Cord::DestroyCordSlow() {
|
|
|
|
assert(contents_.is_tree());
|
|
|
|
CordzInfo::MaybeUntrackCord(contents_.cordz_info());
|
|
|
|
CordRep::Unref(VerifyTree(contents_.as_tree()));
|
|
|
|
}
|
|
|
|
|
|
|
|
// --------------------------------------------------------------------
|
|
|
|
// Mutators
|
|
|
|
|
|
|
|
void Cord::Clear() {
|
|
|
|
if (CordRep* tree = contents_.clear()) {
|
|
|
|
CordRep::Unref(tree);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
Cord& Cord::AssignLargeString(std::string&& src) {
|
|
|
|
auto constexpr method = CordzUpdateTracker::kAssignString;
|
|
|
|
assert(src.size() > kMaxBytesToCopy);
|
|
|
|
CordRep* rep = CordRepFromString(std::move(src));
|
|
|
|
if (CordRep* tree = contents_.tree()) {
|
|
|
|
CordzUpdateScope scope(contents_.cordz_info(), method);
|
|
|
|
contents_.SetTree(rep, scope);
|
|
|
|
CordRep::Unref(tree);
|
|
|
|
} else {
|
|
|
|
contents_.EmplaceTree(rep, method);
|
|
|
|
}
|
|
|
|
return *this;
|
|
|
|
}
|
|
|
|
|
|
|
|
Cord& Cord::operator=(absl::string_view src) {
|
|
|
|
auto constexpr method = CordzUpdateTracker::kAssignString;
|
|
|
|
const char* data = src.data();
|
|
|
|
size_t length = src.size();
|
|
|
|
CordRep* tree = contents_.tree();
|
|
|
|
if (length <= InlineRep::kMaxInline) {
|
|
|
|
// Embed into this->contents_, which is somewhat subtle:
|
|
|
|
// - MaybeUntrackCord must be called before Unref(tree).
|
|
|
|
// - MaybeUntrackCord must be called before set_data() clobbers cordz_info.
|
|
|
|
// - set_data() must be called before Unref(tree) as it may reference tree.
|
|
|
|
if (tree != nullptr) CordzInfo::MaybeUntrackCord(contents_.cordz_info());
|
Export of internal Abseil changes
--
07240ca7822d007cdcc79f2c40bd58b2c2010348 by Abseil Team <absl-team@google.com>:
Correct the comment from "AlphaNum" to "Arg".
PiperOrigin-RevId: 416139192
--
adcba4a6b3763626e1db7b1e8c108b3114903557 by Martijn Vels <mvels@google.com>:
Fix NewExternalRep() to require data being non-empty, and remove nullptr return.
PiperOrigin-RevId: 416135865
--
c0d14cd918fb16f15d1d84de9284b5c5ecc1f8f2 by Abseil Team <absl-team@google.com>:
Fix doc comment for absl::ascii_isprint().
The comment was incorrectly saying that it includes all whitespace.
It doesn't; the only whitespace char it includes is ' '.
PiperOrigin-RevId: 416112524
--
d83327800159c07002b6865e21232a12463e02dd by Abseil Team <absl-team@google.com>:
Internal change
PiperOrigin-RevId: 416099978
--
baf11e9ca42ca9140cdbf8075f971db8d65b1195 by Ilya Tokar <tokarip@google.com>:
Prevent compiler from optimizing Group_Match* benchmarks away.
Currently we benchmark single store of precomputed value.
Not all affected benchmarks show performance changes:
BM_Group_Match 0.53ns ± 1% 0.53ns ± 0% -0.42% (p=0.038 n=10+10)
BM_Group_MatchEmpty 0.26ns ± 1% 0.26ns ± 1% ~ (p=1.000 n=10+10)
BM_Group_MatchEmptyOrDeleted 0.26ns ± 1% 0.26ns ± 1% ~ (p=0.121 n=10+10)
BM_Group_CountLeadingEmptyOrDeleted 0.26ns ± 1% 0.45ns ± 0% +70.05% (p=0.000 n=10+8)
BM_Group_MatchFirstEmptyOrDeleted 0.26ns ± 0% 0.44ns ± 1% +65.91% (p=0.000 n=8+9)
But inspecting the generated code shows the difference,
e. g. BM_Group_MatchFirstEmptyOrDeleted
Before:
add $0xffffffffffffffff,%rbx
jne 30
After:
pcmpeqd %xmm0,%xmm0
pcmpgtb -0x30(%rbp),%xmm0
pmovmskb %xmm0,%eax
add: 0x23$0xffffffffffffffff,%rbx
jne 40
PiperOrigin-RevId: 416083515
--
122fbff893dc4571b3e75e4b241eb4495b925610 by Abseil Team <absl-team@google.com>:
Put namespace guard in ABSL_DECLARE_FLAG to make declaring a flag in a namespace a compiler error instead of a linker error.
PiperOrigin-RevId: 416036072
--
020fd8a20f5fa319e948846e003391fcb9e03868 by Ilya Tokar <tokarip@google.com>:
Make Cord::InlineRep::set_data unconditionally zero out memory.
Currently there is a single case where we don't zero out memory
as an optimization. Unconditional zeroing doesn't show any changes
in benchmarks, except for the unrelated improvement:
BM_CordPartialCopyToCord/1M/1 12.6ns ± 4% 12.6ns ± 4% ~ (p=0.857 n=16+19)
BM_CordPartialCopyToCord/1M/128 44.9ns ± 7% 45.0ns ± 3% ~ (p=0.468 n=18+17)
BM_CordPartialCopyToCord/1M/1k 64.5ns ± 4% 61.4ns ± 4% -4.82% (p=0.000 n=19+17)
BM_CordPartialCopyToCord/1M/8k 139ns ± 3% 128ns ±15% -7.76% (p=0.009 n=17+20)
BM_CordPartialCopyToCord/1M/16k 193ns ± 6% 168ns ± 6% -13.17% (p=0.000 n=17+17)
BM_CordPartialCopyToCord/4M/16k 199ns ± 4% 177ns ± 4% -11.36% (p=0.000 n=17+18)
BM_CordPartialCopyToCord/4M/32k 275ns ± 3% 250ns ± 4% -9.00% (p=0.000 n=18+18)
BM_CordPartialCopyToCord/4M/64k 291ns ± 4% 266ns ± 5% -8.53% (p=0.000 n=18+16)
BM_CordPartialCopyToCord/4M/128k 322ns ± 5% 291ns ± 4% -9.43% (p=0.000 n=20+18)
BM_CordPartialCopyToCord/8M/32k 281ns ± 5% 251ns ± 4% -10.38% (p=0.000 n=20+16)
BM_CordPartialCopyToCord/8M/64k 293ns ± 6% 267ns ± 4% -8.87% (p=0.000 n=16+19)
BM_CordPartialCopyToCord/8M/128k 334ns ± 3% 305ns ± 2% -8.56% (p=0.000 n=17+16)
This is clearly an alignmnet effect since number of the executed instructions is the same:
M_CordPartialCopyToCord/1M/1 155 ± 0% 155 ± 0% ~ (all samples are equal)
BM_CordPartialCopyToCord/1M/128 446 ± 0% 446 ± 0% ~ (p=0.332 n=36+39)
BM_CordPartialCopyToCord/1M/1k 473 ± 0% 473 ± 0% ~ (p=0.969 n=40+40)
BM_CordPartialCopyToCord/1M/8k 808 ± 0% 808 ± 0% ~ (p=0.127 n=40+39)
BM_CordPartialCopyToCord/1M/16k 957 ± 0% 957 ± 0% ~ (p=0.532 n=40+40)
BM_CordPartialCopyToCord/4M/16k 952 ± 0% 952 ± 0% ~ (p=0.686 n=39+39)
BM_CordPartialCopyToCord/4M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.690 n=40+40)
BM_CordPartialCopyToCord/4M/64k 1.23k ± 0% 1.23k ± 0% ~ (p=0.182 n=40+39)
BM_CordPartialCopyToCord/4M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.711 n=40+40)
BM_CordPartialCopyToCord/8M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.697 n=40+40)
BM_CordPartialCopyToCord/8M/64k 1.23k ± 0% 1.23k ± 0% +0.00% (p=0.049 n=40+40)
BM_CordPartialCopyToCord/8M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.507 n=40+40)
This makes code simpler and doesn't regress performance.
PiperOrigin-RevId: 415560574
--
37305b2690b31682088749e4d62f40d7095bdc54 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 415558737
--
86aaed569b9e743c1eb813a5f48def978a793db3 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415515201
--
6cdb8786cdcb4fa0b8a4b72fc98940877d1fdeff by Abseil Team <absl-team@google.com>:
Update SubmitMutexProfileData to accept wait_cycles instead of wait_timestamp
PiperOrigin-RevId: 415360871
--
9f979d307aa16ad09f214e04876cbe84395c0901 by Abseil Team <absl-team@google.com>:
absl::flat_hash_set compiles with -Wconversion -Wsign-compare
PiperOrigin-RevId: 415357498
--
9eceb14174708f15e61259d449b214a8a4c7f9e7 by Abseil Team <absl-team@google.com>:
Fix AddressIsReadable for the corner case of (aligned) addr == NULL.
PiperOrigin-RevId: 415307792
--
1a39ffe55898375e2d7f88c17c99db5a1b95b313 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415162872
--
64378549b110d5f5762185a5906c520fba70f0e7 by Abseil Team <absl-team@google.com>:
Fix a typo in the comments
PiperOrigin-RevId: 415088461
--
41aae8322e913b82710153c22b97c611fdb6e1fb by Abseil Team <absl-team@google.com>:
Switch from `connect` to `rt_sigreturn` -- the latter is much less problematic
for system call sandboxes.
PiperOrigin-RevId: 415073965
--
870c5e3388b6a35611bff538626fe7a1c8c87171 by Abseil Team <absl-team@google.com>:
Add ABSL_HAVE_HWADDRESS_SANITIZER and ABSL_HAVE_LEAK_SANITIZER
PiperOrigin-RevId: 414871189
--
f213ed60a66b58da7ac40555adfb1d529ff0a4db by Derek Mauro <dmauro@google.com>:
Remove reference to __SANITIZE_MEMORY__, which does not exist
It appears to have been copied by pattern matching from the ASAN/TSAN
code blocks.
https://github.com/gcc-mirror/gcc/blob/f47662204de27f7685699eeef89aa173ccf32d85/gcc/cppbuiltin.c#L79-L126
PiperOrigin-RevId: 414806587
--
b152891e73ab515f397ceb53f66c8ee2f33863ea by Abseil Team <absl-team@google.com>:
Rollback previous commit: SYS_open is not defined in certain environments.
PiperOrigin-RevId: 414521820
--
5a1cbb282331023902e1374dd0d920c4effbe47f by Abseil Team <absl-team@google.com>:
Use syscall(SYS_open, ...) instead of open() to avoid possible symbol
interposition.
Also add some warning notes.
PiperOrigin-RevId: 414508186
--
1824d6593612710aafdc599a89b0adced7d787f6 by Abseil Team <absl-team@google.com>:
Correct aarch64 macro check
The macro is __aarch64__, not __arch64__.
PiperOrigin-RevId: 414446225
--
a1536a57b64dfd53945d33a01cfc08b18c99c97b by Abseil Team <absl-team@google.com>:
Fix backwards comment in the last commit.
PiperOrigin-RevId: 414281214
--
11ac021ba779513667a31cf2563ddafc57d6d913 by Abseil Team <absl-team@google.com>:
AddressIsReadable() didn't work correctly on ARM when the given pointer was
misaligned at the end of the page.
Fix that by aligning the pointer on an 8-byte boundary before checking it.
PiperOrigin-RevId: 414203863
GitOrigin-RevId: 07240ca7822d007cdcc79f2c40bd58b2c2010348
Change-Id: If5f129194d59f5c9e5d84efd8cd9e17a70e072ab
3 years ago
|
|
|
contents_.set_data(data, length);
|
|
|
|
if (tree != nullptr) CordRep::Unref(tree);
|
|
|
|
return *this;
|
|
|
|
}
|
|
|
|
if (tree != nullptr) {
|
|
|
|
CordzUpdateScope scope(contents_.cordz_info(), method);
|
|
|
|
if (tree->IsFlat() && tree->flat()->Capacity() >= length &&
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
tree->refcount.IsOne()) {
|
|
|
|
// Copy in place if the existing FLAT node is reusable.
|
|
|
|
memmove(tree->flat()->Data(), data, length);
|
|
|
|
tree->length = length;
|
|
|
|
VerifyTree(tree);
|
|
|
|
return *this;
|
|
|
|
}
|
|
|
|
contents_.SetTree(NewTree(data, length, 0), scope);
|
|
|
|
CordRep::Unref(tree);
|
|
|
|
} else {
|
|
|
|
contents_.EmplaceTree(NewTree(data, length, 0), method);
|
|
|
|
}
|
|
|
|
return *this;
|
|
|
|
}
|
|
|
|
|
|
|
|
// TODO(sanjay): Move to Cord::InlineRep section of file. For now,
|
|
|
|
// we keep it here to make diffs easier.
|
|
|
|
void Cord::InlineRep::AppendArray(absl::string_view src,
|
|
|
|
MethodIdentifier method) {
|
|
|
|
if (src.empty()) return; // memcpy(_, nullptr, 0) is undefined.
|
|
|
|
|
|
|
|
size_t appended = 0;
|
|
|
|
CordRep* rep = tree();
|
|
|
|
const CordRep* const root = rep;
|
|
|
|
CordzUpdateScope scope(root ? cordz_info() : nullptr, method);
|
|
|
|
if (root != nullptr) {
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
rep = cord_internal::RemoveCrcNode(rep);
|
|
|
|
char* region;
|
|
|
|
if (PrepareAppendRegion(rep, ®ion, &appended, src.size())) {
|
|
|
|
memcpy(region, src.data(), appended);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
// Try to fit in the inline buffer if possible.
|
|
|
|
size_t inline_length = inline_size();
|
|
|
|
if (src.size() <= kMaxInline - inline_length) {
|
|
|
|
// Append new data to embedded array
|
|
|
|
memcpy(data_.as_chars() + inline_length, src.data(), src.size());
|
|
|
|
set_inline_size(inline_length + src.size());
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Allocate flat to be a perfect fit on first append exceeding inlined size.
|
|
|
|
// Subsequent growth will use amortized growth until we reach maximum flat
|
|
|
|
// size.
|
|
|
|
rep = CordRepFlat::New(inline_length + src.size());
|
|
|
|
appended = std::min(src.size(), rep->flat()->Capacity() - inline_length);
|
|
|
|
memcpy(rep->flat()->Data(), data_.as_chars(), inline_length);
|
|
|
|
memcpy(rep->flat()->Data() + inline_length, src.data(), appended);
|
|
|
|
rep->length = inline_length + appended;
|
|
|
|
}
|
|
|
|
|
|
|
|
src.remove_prefix(appended);
|
|
|
|
if (src.empty()) {
|
|
|
|
CommitTree(root, rep, scope, method);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// TODO(b/192061034): keep legacy 10% growth rate: consider other rates.
|
|
|
|
rep = ForceBtree(rep);
|
|
|
|
const size_t min_growth = std::max<size_t>(rep->length / 10, src.size());
|
|
|
|
rep = CordRepBtree::Append(rep->btree(), src, min_growth - src.size());
|
|
|
|
|
|
|
|
CommitTree(root, rep, scope, method);
|
|
|
|
}
|
|
|
|
|
|
|
|
inline CordRep* Cord::TakeRep() const& {
|
|
|
|
return CordRep::Ref(contents_.tree());
|
|
|
|
}
|
|
|
|
|
|
|
|
inline CordRep* Cord::TakeRep() && {
|
|
|
|
CordRep* rep = contents_.tree();
|
|
|
|
contents_.clear();
|
|
|
|
return rep;
|
|
|
|
}
|
|
|
|
|
|
|
|
template <typename C>
|
|
|
|
inline void Cord::AppendImpl(C&& src) {
|
|
|
|
auto constexpr method = CordzUpdateTracker::kAppendCord;
|
|
|
|
if (empty()) {
|
|
|
|
// Since destination is empty, we can avoid allocating a node,
|
|
|
|
if (src.contents_.is_tree()) {
|
|
|
|
// by taking the tree directly
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
CordRep* rep =
|
|
|
|
cord_internal::RemoveCrcNode(std::forward<C>(src).TakeRep());
|
|
|
|
contents_.EmplaceTree(rep, method);
|
|
|
|
} else {
|
|
|
|
// or copying over inline data
|
|
|
|
contents_.data_ = src.contents_.data_;
|
|
|
|
}
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// For short cords, it is faster to copy data if there is room in dst.
|
|
|
|
const size_t src_size = src.contents_.size();
|
|
|
|
if (src_size <= kMaxBytesToCopy) {
|
|
|
|
CordRep* src_tree = src.contents_.tree();
|
|
|
|
if (src_tree == nullptr) {
|
|
|
|
// src has embedded data.
|
|
|
|
contents_.AppendArray({src.contents_.data(), src_size}, method);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
if (src_tree->IsFlat()) {
|
|
|
|
// src tree just has one flat node.
|
|
|
|
contents_.AppendArray({src_tree->flat()->Data(), src_size}, method);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
if (&src == this) {
|
|
|
|
// ChunkIterator below assumes that src is not modified during traversal.
|
|
|
|
Append(Cord(src));
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
// TODO(mec): Should we only do this if "dst" has space?
|
|
|
|
for (absl::string_view chunk : src.Chunks()) {
|
|
|
|
Append(chunk);
|
|
|
|
}
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Guaranteed to be a tree (kMaxBytesToCopy > kInlinedSize)
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
CordRep* rep = cord_internal::RemoveCrcNode(std::forward<C>(src).TakeRep());
|
|
|
|
contents_.AppendTree(rep, CordzUpdateTracker::kAppendCord);
|
|
|
|
}
|
|
|
|
|
|
|
|
static CordRep::ExtractResult ExtractAppendBuffer(CordRep* rep,
|
|
|
|
size_t min_capacity) {
|
|
|
|
switch (rep->tag) {
|
|
|
|
case cord_internal::BTREE:
|
|
|
|
return CordRepBtree::ExtractAppendBuffer(rep->btree(), min_capacity);
|
|
|
|
default:
|
|
|
|
if (rep->IsFlat() && rep->refcount.IsOne() &&
|
|
|
|
rep->flat()->Capacity() - rep->length >= min_capacity) {
|
|
|
|
return {nullptr, rep};
|
|
|
|
}
|
|
|
|
return {rep, nullptr};
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static CordBuffer CreateAppendBuffer(InlineData& data, size_t block_size,
|
|
|
|
size_t capacity) {
|
|
|
|
// Watch out for overflow, people can ask for size_t::max().
|
|
|
|
const size_t size = data.inline_size();
|
|
|
|
const size_t max_capacity = std::numeric_limits<size_t>::max() - size;
|
|
|
|
capacity = (std::min)(max_capacity, capacity) + size;
|
|
|
|
CordBuffer buffer =
|
|
|
|
block_size ? CordBuffer::CreateWithCustomLimit(block_size, capacity)
|
|
|
|
: CordBuffer::CreateWithDefaultLimit(capacity);
|
|
|
|
cord_internal::SmallMemmove(buffer.data(), data.as_chars(), size);
|
|
|
|
buffer.SetLength(size);
|
|
|
|
data = {};
|
|
|
|
return buffer;
|
|
|
|
}
|
|
|
|
|
|
|
|
CordBuffer Cord::GetAppendBufferSlowPath(size_t block_size, size_t capacity,
|
|
|
|
size_t min_capacity) {
|
|
|
|
auto constexpr method = CordzUpdateTracker::kGetAppendBuffer;
|
|
|
|
CordRep* tree = contents_.tree();
|
|
|
|
if (tree != nullptr) {
|
|
|
|
CordzUpdateScope scope(contents_.cordz_info(), method);
|
|
|
|
CordRep::ExtractResult result = ExtractAppendBuffer(tree, min_capacity);
|
|
|
|
if (result.extracted != nullptr) {
|
|
|
|
contents_.SetTreeOrEmpty(result.tree, scope);
|
|
|
|
return CordBuffer(result.extracted->flat());
|
|
|
|
}
|
|
|
|
return block_size ? CordBuffer::CreateWithCustomLimit(block_size, capacity)
|
|
|
|
: CordBuffer::CreateWithDefaultLimit(capacity);
|
|
|
|
}
|
|
|
|
return CreateAppendBuffer(contents_.data_, block_size, capacity);
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::Append(const Cord& src) {
|
|
|
|
AppendImpl(src);
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::Append(Cord&& src) {
|
|
|
|
AppendImpl(std::move(src));
|
|
|
|
}
|
|
|
|
|
Export of internal Abseil changes
--
7d0468a6610ed85586d5c87fd65de8dac5118923 by Derek Mauro <dmauro@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 313226473
--
1131ef6d116f5ce7d46537a82f300ea06dcaaa53 by Gennadiy Rozental <rogeeff@google.com>:
Migrate internal interface to use mutable references.
PiperOrigin-RevId: 312931131
--
96225212a9f5fbd0b38c71fe65539164992c7c3b by Laramie Leavitt <lar@google.com>:
Remove random/internal/distributions.h
This file was something of an historical artifact. All of the related
code has either been removed or migraged, and so the only remaining type
belongs with uniform_helper.h, as it is used to infer the return type
of the absl::Uniform method in a few cases.
PiperOrigin-RevId: 312878173
--
6dcbd5be58ad425e08740ff64088373ee7fe4a72 by Mark Barolak <mbar@google.com>:
Release the StrFormat test case for Cords to open source.
PiperOrigin-RevId: 312707974
--
34484d18dfb63a0a7ad6e2aaeb570e33592968be by Abseil Team <absl-team@google.com>:
Let Cord::Cord(string&&), Cord::operator=(string&&),
Cord::Append(string&&), and Cord::Prepend(string&&) steal string data
and embed it into the Cord as a single external chunk, instead of
copying it into flat chunks (at most 4083-byte each).
Stealing string data is faster, but it creates a long chunk, which leads
to a higher more memory usage if its subcords are created and outlive
the whole Cord.
These functions revert to copying the data if any of the following
conditions holds:
- string size is at most kMaxBytesToCopy (511), to avoid the overhead
of an external chunk for short strings;
- less than half of string capacity is used, to avoid pinning to much
unused memory.
PiperOrigin-RevId: 312683785
GitOrigin-RevId: 7d0468a6610ed85586d5c87fd65de8dac5118923
Change-Id: If79b5a1dfe6d53a8ddddbc7da84338f11fc4cfa3
5 years ago
|
|
|
template <typename T, Cord::EnableIfString<T>>
|
|
|
|
void Cord::Append(T&& src) {
|
|
|
|
if (src.size() <= kMaxBytesToCopy) {
|
|
|
|
Append(absl::string_view(src));
|
|
|
|
} else {
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
CordRep* rep = CordRepFromString(std::forward<T>(src));
|
|
|
|
contents_.AppendTree(rep, CordzUpdateTracker::kAppendString);
|
Export of internal Abseil changes
--
7d0468a6610ed85586d5c87fd65de8dac5118923 by Derek Mauro <dmauro@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 313226473
--
1131ef6d116f5ce7d46537a82f300ea06dcaaa53 by Gennadiy Rozental <rogeeff@google.com>:
Migrate internal interface to use mutable references.
PiperOrigin-RevId: 312931131
--
96225212a9f5fbd0b38c71fe65539164992c7c3b by Laramie Leavitt <lar@google.com>:
Remove random/internal/distributions.h
This file was something of an historical artifact. All of the related
code has either been removed or migraged, and so the only remaining type
belongs with uniform_helper.h, as it is used to infer the return type
of the absl::Uniform method in a few cases.
PiperOrigin-RevId: 312878173
--
6dcbd5be58ad425e08740ff64088373ee7fe4a72 by Mark Barolak <mbar@google.com>:
Release the StrFormat test case for Cords to open source.
PiperOrigin-RevId: 312707974
--
34484d18dfb63a0a7ad6e2aaeb570e33592968be by Abseil Team <absl-team@google.com>:
Let Cord::Cord(string&&), Cord::operator=(string&&),
Cord::Append(string&&), and Cord::Prepend(string&&) steal string data
and embed it into the Cord as a single external chunk, instead of
copying it into flat chunks (at most 4083-byte each).
Stealing string data is faster, but it creates a long chunk, which leads
to a higher more memory usage if its subcords are created and outlive
the whole Cord.
These functions revert to copying the data if any of the following
conditions holds:
- string size is at most kMaxBytesToCopy (511), to avoid the overhead
of an external chunk for short strings;
- less than half of string capacity is used, to avoid pinning to much
unused memory.
PiperOrigin-RevId: 312683785
GitOrigin-RevId: 7d0468a6610ed85586d5c87fd65de8dac5118923
Change-Id: If79b5a1dfe6d53a8ddddbc7da84338f11fc4cfa3
5 years ago
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
template void Cord::Append(std::string&& src);
|
|
|
|
|
|
|
|
void Cord::Prepend(const Cord& src) {
|
|
|
|
CordRep* src_tree = src.contents_.tree();
|
|
|
|
if (src_tree != nullptr) {
|
|
|
|
CordRep::Ref(src_tree);
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
contents_.PrependTree(cord_internal::RemoveCrcNode(src_tree),
|
|
|
|
CordzUpdateTracker::kPrependCord);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// `src` cord is inlined.
|
|
|
|
absl::string_view src_contents(src.contents_.data(), src.contents_.size());
|
|
|
|
return Prepend(src_contents);
|
|
|
|
}
|
|
|
|
|
Export of internal Abseil changes
--
f49e405201d2ffd5955503fa8ad0f08ec0cdfb2b by Martijn Vels <mvels@google.com>:
Add common [container.requirements] type definitions to `CharRange` and `ChunkRange`
The presence of these allow these range classes to be used in various utility functions which require some minimum type of container. For example, this change allows tests to use `EXPECT_THAT(cord.Chunks(), ElementsAre(...))`
PiperOrigin-RevId: 406941278
--
0c195f073632e21d9a4bce158047b2ba8551c2d1 by Evan Brown <ezb@google.com>:
Use explicit exponential growth in SubstituteAndAppendArray.
PiperOrigin-RevId: 406931952
--
afb043bccd809a55cab78abadb7548a057d9eda0 by Jorg Brown <jorg@google.com>:
Use longer var names in macro to avoid clang-tidy warning
PiperOrigin-RevId: 406930978
--
80397e2604e6b3d929a34742c3a32581b34d3ac4 by Martijn Vels <mvels@google.com>:
Add future kAppendBuffer and kPrependBuffer API trackers for Cordz sampling
PiperOrigin-RevId: 406912759
--
e910ce919ef83933f08a690e8b7325c7cc5b6d5d by Martijn Vels <mvels@google.com>:
Implement Prepend(string_view) in terms of PrependArray(string_view, MethodIdentifier).
PiperOrigin-RevId: 406891665
--
c9cff43d4c0568ed01f2fca0f6ef038ae03112b5 by Martijn Vels <mvels@google.com>:
Add 'Rebuild' logic to CordRepBtree
There are btree hostile scenarios where an application could perform repeated split/insert/merge operations on a cord leading to a tree exceeding the maximum height. While this should be rare in practice, this change adds a Rebuild() method that will rebuild a tree with a 100% fill factor, and we will invoke this rebuild when a tree exceeds the maximum height. This basically follows the similar 'balance' logic in Concat trees (although the latter is common in Concat uses)
PiperOrigin-RevId: 406875739
--
5b2b8fb88f1ebfdc1c670088152da2cb2ea4c376 by Martijn Vels <mvels@google.com>:
Add 'in place' enabled RemoveSuffix
An in-place RemoveSuffix is more efficient than SubTree() as it can directly modify privately owned nodes and flats allowing easy re-use of free capacity in right-most flats that may turn into Substring edges when using SubTree.
PiperOrigin-RevId: 406431230
--
f09903c0a3d7344f59aaf1380a16ea10829217d4 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 406430373
--
9957af575c33bb18dc170572a4ee8cc5901df6b2 by Greg Falcon <gfalcon@google.com>:
Initial groundwork to allow storing checksum data inside CordRep instances.
This uses a RefcountAndFlags bit that was reserved for this purpose, and will be leveraged in a follow-up change to allow attaching checksums to a Cord's value.
This change splits RefcountAndFlags::IsOne() into two distinct operations:
* IsOne(): This returns true when the associated CordRep is not shared with other threads. This is useful for functions that consume CordRep instances; for example, code that consumes an unshared CordRep can assume ownership of its children without modifying those refcounts.
* IsMutable(): This returns true when the associated CordRep reference is not shared with other threads, *and* does not store an associated checksum value. This is useful for functions that modify a CordRep's contents: code may modify the bytes of a mutable-unshared CordRep without fear of races with other threads, or of invalidating a stored checksum.
The tricky part of this CL is ensuring that the correct choice between IsMutable() and IsOne() was made at each point. An incorrect application of IsOne() could lead to correctness bugs in the future. Code conditioned on IsOne() may delete the CordRep in question, or assume ownership of its children, but must not modify the CordRep's data without explicitly adjusting the CRC.
PiperOrigin-RevId: 406191103
--
686544814079e5ab6d4593cca0c068b510be400a by Martijn Vels <mvels@google.com>:
Reduce the size in the LargeString test when running with Sanitizers
PiperOrigin-RevId: 406186945
--
735b4490bdb695c35731f06ce4b8de14ce2be6ed by Alex Strelnikov <strel@google.com>:
Release absl::SimpleHexAtoi.
PiperOrigin-RevId: 406143188
GitOrigin-RevId: f49e405201d2ffd5955503fa8ad0f08ec0cdfb2b
Change-Id: Ic6527ac40fa03ea02ca813e8bb7868a219544de4
3 years ago
|
|
|
void Cord::PrependArray(absl::string_view src, MethodIdentifier method) {
|
|
|
|
if (src.empty()) return; // memcpy(_, nullptr, 0) is undefined.
|
|
|
|
if (!contents_.is_tree()) {
|
|
|
|
size_t cur_size = contents_.inline_size();
|
|
|
|
if (cur_size + src.size() <= InlineRep::kMaxInline) {
|
|
|
|
// Use embedded storage.
|
|
|
|
char data[InlineRep::kMaxInline + 1] = {0};
|
|
|
|
memcpy(data, src.data(), src.size());
|
|
|
|
memcpy(data + src.size(), contents_.data(), cur_size);
|
|
|
|
memcpy(contents_.data_.as_chars(), data, InlineRep::kMaxInline + 1);
|
|
|
|
contents_.set_inline_size(cur_size + src.size());
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
CordRep* rep = NewTree(src.data(), src.size(), 0);
|
Export of internal Abseil changes
--
f49e405201d2ffd5955503fa8ad0f08ec0cdfb2b by Martijn Vels <mvels@google.com>:
Add common [container.requirements] type definitions to `CharRange` and `ChunkRange`
The presence of these allow these range classes to be used in various utility functions which require some minimum type of container. For example, this change allows tests to use `EXPECT_THAT(cord.Chunks(), ElementsAre(...))`
PiperOrigin-RevId: 406941278
--
0c195f073632e21d9a4bce158047b2ba8551c2d1 by Evan Brown <ezb@google.com>:
Use explicit exponential growth in SubstituteAndAppendArray.
PiperOrigin-RevId: 406931952
--
afb043bccd809a55cab78abadb7548a057d9eda0 by Jorg Brown <jorg@google.com>:
Use longer var names in macro to avoid clang-tidy warning
PiperOrigin-RevId: 406930978
--
80397e2604e6b3d929a34742c3a32581b34d3ac4 by Martijn Vels <mvels@google.com>:
Add future kAppendBuffer and kPrependBuffer API trackers for Cordz sampling
PiperOrigin-RevId: 406912759
--
e910ce919ef83933f08a690e8b7325c7cc5b6d5d by Martijn Vels <mvels@google.com>:
Implement Prepend(string_view) in terms of PrependArray(string_view, MethodIdentifier).
PiperOrigin-RevId: 406891665
--
c9cff43d4c0568ed01f2fca0f6ef038ae03112b5 by Martijn Vels <mvels@google.com>:
Add 'Rebuild' logic to CordRepBtree
There are btree hostile scenarios where an application could perform repeated split/insert/merge operations on a cord leading to a tree exceeding the maximum height. While this should be rare in practice, this change adds a Rebuild() method that will rebuild a tree with a 100% fill factor, and we will invoke this rebuild when a tree exceeds the maximum height. This basically follows the similar 'balance' logic in Concat trees (although the latter is common in Concat uses)
PiperOrigin-RevId: 406875739
--
5b2b8fb88f1ebfdc1c670088152da2cb2ea4c376 by Martijn Vels <mvels@google.com>:
Add 'in place' enabled RemoveSuffix
An in-place RemoveSuffix is more efficient than SubTree() as it can directly modify privately owned nodes and flats allowing easy re-use of free capacity in right-most flats that may turn into Substring edges when using SubTree.
PiperOrigin-RevId: 406431230
--
f09903c0a3d7344f59aaf1380a16ea10829217d4 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 406430373
--
9957af575c33bb18dc170572a4ee8cc5901df6b2 by Greg Falcon <gfalcon@google.com>:
Initial groundwork to allow storing checksum data inside CordRep instances.
This uses a RefcountAndFlags bit that was reserved for this purpose, and will be leveraged in a follow-up change to allow attaching checksums to a Cord's value.
This change splits RefcountAndFlags::IsOne() into two distinct operations:
* IsOne(): This returns true when the associated CordRep is not shared with other threads. This is useful for functions that consume CordRep instances; for example, code that consumes an unshared CordRep can assume ownership of its children without modifying those refcounts.
* IsMutable(): This returns true when the associated CordRep reference is not shared with other threads, *and* does not store an associated checksum value. This is useful for functions that modify a CordRep's contents: code may modify the bytes of a mutable-unshared CordRep without fear of races with other threads, or of invalidating a stored checksum.
The tricky part of this CL is ensuring that the correct choice between IsMutable() and IsOne() was made at each point. An incorrect application of IsOne() could lead to correctness bugs in the future. Code conditioned on IsOne() may delete the CordRep in question, or assume ownership of its children, but must not modify the CordRep's data without explicitly adjusting the CRC.
PiperOrigin-RevId: 406191103
--
686544814079e5ab6d4593cca0c068b510be400a by Martijn Vels <mvels@google.com>:
Reduce the size in the LargeString test when running with Sanitizers
PiperOrigin-RevId: 406186945
--
735b4490bdb695c35731f06ce4b8de14ce2be6ed by Alex Strelnikov <strel@google.com>:
Release absl::SimpleHexAtoi.
PiperOrigin-RevId: 406143188
GitOrigin-RevId: f49e405201d2ffd5955503fa8ad0f08ec0cdfb2b
Change-Id: Ic6527ac40fa03ea02ca813e8bb7868a219544de4
3 years ago
|
|
|
contents_.PrependTree(rep, method);
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::AppendPrecise(absl::string_view src, MethodIdentifier method) {
|
|
|
|
assert(!src.empty());
|
|
|
|
assert(src.size() <= cord_internal::kMaxFlatLength);
|
|
|
|
if (contents_.remaining_inline_capacity() >= src.size()) {
|
|
|
|
const size_t inline_length = contents_.inline_size();
|
|
|
|
memcpy(contents_.data_.as_chars() + inline_length, src.data(), src.size());
|
|
|
|
contents_.set_inline_size(inline_length + src.size());
|
|
|
|
} else {
|
|
|
|
contents_.AppendTree(CordRepFlat::Create(src), method);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::PrependPrecise(absl::string_view src, MethodIdentifier method) {
|
|
|
|
assert(!src.empty());
|
|
|
|
assert(src.size() <= cord_internal::kMaxFlatLength);
|
|
|
|
if (contents_.remaining_inline_capacity() >= src.size()) {
|
|
|
|
const size_t inline_length = contents_.inline_size();
|
|
|
|
char data[InlineRep::kMaxInline + 1] = {0};
|
|
|
|
memcpy(data, src.data(), src.size());
|
|
|
|
memcpy(data + src.size(), contents_.data(), inline_length);
|
|
|
|
memcpy(contents_.data_.as_chars(), data, InlineRep::kMaxInline + 1);
|
|
|
|
contents_.set_inline_size(inline_length + src.size());
|
|
|
|
} else {
|
|
|
|
contents_.PrependTree(CordRepFlat::Create(src), method);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Export of internal Abseil changes
--
7d0468a6610ed85586d5c87fd65de8dac5118923 by Derek Mauro <dmauro@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 313226473
--
1131ef6d116f5ce7d46537a82f300ea06dcaaa53 by Gennadiy Rozental <rogeeff@google.com>:
Migrate internal interface to use mutable references.
PiperOrigin-RevId: 312931131
--
96225212a9f5fbd0b38c71fe65539164992c7c3b by Laramie Leavitt <lar@google.com>:
Remove random/internal/distributions.h
This file was something of an historical artifact. All of the related
code has either been removed or migraged, and so the only remaining type
belongs with uniform_helper.h, as it is used to infer the return type
of the absl::Uniform method in a few cases.
PiperOrigin-RevId: 312878173
--
6dcbd5be58ad425e08740ff64088373ee7fe4a72 by Mark Barolak <mbar@google.com>:
Release the StrFormat test case for Cords to open source.
PiperOrigin-RevId: 312707974
--
34484d18dfb63a0a7ad6e2aaeb570e33592968be by Abseil Team <absl-team@google.com>:
Let Cord::Cord(string&&), Cord::operator=(string&&),
Cord::Append(string&&), and Cord::Prepend(string&&) steal string data
and embed it into the Cord as a single external chunk, instead of
copying it into flat chunks (at most 4083-byte each).
Stealing string data is faster, but it creates a long chunk, which leads
to a higher more memory usage if its subcords are created and outlive
the whole Cord.
These functions revert to copying the data if any of the following
conditions holds:
- string size is at most kMaxBytesToCopy (511), to avoid the overhead
of an external chunk for short strings;
- less than half of string capacity is used, to avoid pinning to much
unused memory.
PiperOrigin-RevId: 312683785
GitOrigin-RevId: 7d0468a6610ed85586d5c87fd65de8dac5118923
Change-Id: If79b5a1dfe6d53a8ddddbc7da84338f11fc4cfa3
5 years ago
|
|
|
template <typename T, Cord::EnableIfString<T>>
|
|
|
|
inline void Cord::Prepend(T&& src) {
|
|
|
|
if (src.size() <= kMaxBytesToCopy) {
|
|
|
|
Prepend(absl::string_view(src));
|
|
|
|
} else {
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
CordRep* rep = CordRepFromString(std::forward<T>(src));
|
|
|
|
contents_.PrependTree(rep, CordzUpdateTracker::kPrependString);
|
Export of internal Abseil changes
--
7d0468a6610ed85586d5c87fd65de8dac5118923 by Derek Mauro <dmauro@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 313226473
--
1131ef6d116f5ce7d46537a82f300ea06dcaaa53 by Gennadiy Rozental <rogeeff@google.com>:
Migrate internal interface to use mutable references.
PiperOrigin-RevId: 312931131
--
96225212a9f5fbd0b38c71fe65539164992c7c3b by Laramie Leavitt <lar@google.com>:
Remove random/internal/distributions.h
This file was something of an historical artifact. All of the related
code has either been removed or migraged, and so the only remaining type
belongs with uniform_helper.h, as it is used to infer the return type
of the absl::Uniform method in a few cases.
PiperOrigin-RevId: 312878173
--
6dcbd5be58ad425e08740ff64088373ee7fe4a72 by Mark Barolak <mbar@google.com>:
Release the StrFormat test case for Cords to open source.
PiperOrigin-RevId: 312707974
--
34484d18dfb63a0a7ad6e2aaeb570e33592968be by Abseil Team <absl-team@google.com>:
Let Cord::Cord(string&&), Cord::operator=(string&&),
Cord::Append(string&&), and Cord::Prepend(string&&) steal string data
and embed it into the Cord as a single external chunk, instead of
copying it into flat chunks (at most 4083-byte each).
Stealing string data is faster, but it creates a long chunk, which leads
to a higher more memory usage if its subcords are created and outlive
the whole Cord.
These functions revert to copying the data if any of the following
conditions holds:
- string size is at most kMaxBytesToCopy (511), to avoid the overhead
of an external chunk for short strings;
- less than half of string capacity is used, to avoid pinning to much
unused memory.
PiperOrigin-RevId: 312683785
GitOrigin-RevId: 7d0468a6610ed85586d5c87fd65de8dac5118923
Change-Id: If79b5a1dfe6d53a8ddddbc7da84338f11fc4cfa3
5 years ago
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
template void Cord::Prepend(std::string&& src);
|
|
|
|
|
|
|
|
void Cord::RemovePrefix(size_t n) {
|
|
|
|
ABSL_INTERNAL_CHECK(n <= size(),
|
|
|
|
absl::StrCat("Requested prefix size ", n,
|
|
|
|
" exceeds Cord's size ", size()));
|
|
|
|
CordRep* tree = contents_.tree();
|
|
|
|
if (tree == nullptr) {
|
|
|
|
contents_.remove_prefix(n);
|
|
|
|
} else {
|
|
|
|
auto constexpr method = CordzUpdateTracker::kRemovePrefix;
|
|
|
|
CordzUpdateScope scope(contents_.cordz_info(), method);
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
tree = cord_internal::RemoveCrcNode(tree);
|
|
|
|
if (n >= tree->length) {
|
|
|
|
CordRep::Unref(tree);
|
|
|
|
tree = nullptr;
|
|
|
|
} else if (tree->IsBtree()) {
|
|
|
|
CordRep* old = tree;
|
|
|
|
tree = tree->btree()->SubTree(n, tree->length - n);
|
|
|
|
CordRep::Unref(old);
|
|
|
|
} else if (tree->IsSubstring() && tree->refcount.IsOne()) {
|
|
|
|
tree->substring()->start += n;
|
|
|
|
tree->length -= n;
|
|
|
|
} else {
|
|
|
|
CordRep* rep = CordRepSubstring::Substring(tree, n, tree->length - n);
|
|
|
|
CordRep::Unref(tree);
|
|
|
|
tree = rep;
|
|
|
|
}
|
|
|
|
contents_.SetTreeOrEmpty(tree, scope);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::RemoveSuffix(size_t n) {
|
|
|
|
ABSL_INTERNAL_CHECK(n <= size(),
|
|
|
|
absl::StrCat("Requested suffix size ", n,
|
|
|
|
" exceeds Cord's size ", size()));
|
|
|
|
CordRep* tree = contents_.tree();
|
|
|
|
if (tree == nullptr) {
|
|
|
|
contents_.reduce_size(n);
|
|
|
|
} else {
|
|
|
|
auto constexpr method = CordzUpdateTracker::kRemoveSuffix;
|
|
|
|
CordzUpdateScope scope(contents_.cordz_info(), method);
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
tree = cord_internal::RemoveCrcNode(tree);
|
|
|
|
if (n >= tree->length) {
|
|
|
|
CordRep::Unref(tree);
|
|
|
|
tree = nullptr;
|
|
|
|
} else if (tree->IsBtree()) {
|
Export of internal Abseil changes
--
f49e405201d2ffd5955503fa8ad0f08ec0cdfb2b by Martijn Vels <mvels@google.com>:
Add common [container.requirements] type definitions to `CharRange` and `ChunkRange`
The presence of these allow these range classes to be used in various utility functions which require some minimum type of container. For example, this change allows tests to use `EXPECT_THAT(cord.Chunks(), ElementsAre(...))`
PiperOrigin-RevId: 406941278
--
0c195f073632e21d9a4bce158047b2ba8551c2d1 by Evan Brown <ezb@google.com>:
Use explicit exponential growth in SubstituteAndAppendArray.
PiperOrigin-RevId: 406931952
--
afb043bccd809a55cab78abadb7548a057d9eda0 by Jorg Brown <jorg@google.com>:
Use longer var names in macro to avoid clang-tidy warning
PiperOrigin-RevId: 406930978
--
80397e2604e6b3d929a34742c3a32581b34d3ac4 by Martijn Vels <mvels@google.com>:
Add future kAppendBuffer and kPrependBuffer API trackers for Cordz sampling
PiperOrigin-RevId: 406912759
--
e910ce919ef83933f08a690e8b7325c7cc5b6d5d by Martijn Vels <mvels@google.com>:
Implement Prepend(string_view) in terms of PrependArray(string_view, MethodIdentifier).
PiperOrigin-RevId: 406891665
--
c9cff43d4c0568ed01f2fca0f6ef038ae03112b5 by Martijn Vels <mvels@google.com>:
Add 'Rebuild' logic to CordRepBtree
There are btree hostile scenarios where an application could perform repeated split/insert/merge operations on a cord leading to a tree exceeding the maximum height. While this should be rare in practice, this change adds a Rebuild() method that will rebuild a tree with a 100% fill factor, and we will invoke this rebuild when a tree exceeds the maximum height. This basically follows the similar 'balance' logic in Concat trees (although the latter is common in Concat uses)
PiperOrigin-RevId: 406875739
--
5b2b8fb88f1ebfdc1c670088152da2cb2ea4c376 by Martijn Vels <mvels@google.com>:
Add 'in place' enabled RemoveSuffix
An in-place RemoveSuffix is more efficient than SubTree() as it can directly modify privately owned nodes and flats allowing easy re-use of free capacity in right-most flats that may turn into Substring edges when using SubTree.
PiperOrigin-RevId: 406431230
--
f09903c0a3d7344f59aaf1380a16ea10829217d4 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 406430373
--
9957af575c33bb18dc170572a4ee8cc5901df6b2 by Greg Falcon <gfalcon@google.com>:
Initial groundwork to allow storing checksum data inside CordRep instances.
This uses a RefcountAndFlags bit that was reserved for this purpose, and will be leveraged in a follow-up change to allow attaching checksums to a Cord's value.
This change splits RefcountAndFlags::IsOne() into two distinct operations:
* IsOne(): This returns true when the associated CordRep is not shared with other threads. This is useful for functions that consume CordRep instances; for example, code that consumes an unshared CordRep can assume ownership of its children without modifying those refcounts.
* IsMutable(): This returns true when the associated CordRep reference is not shared with other threads, *and* does not store an associated checksum value. This is useful for functions that modify a CordRep's contents: code may modify the bytes of a mutable-unshared CordRep without fear of races with other threads, or of invalidating a stored checksum.
The tricky part of this CL is ensuring that the correct choice between IsMutable() and IsOne() was made at each point. An incorrect application of IsOne() could lead to correctness bugs in the future. Code conditioned on IsOne() may delete the CordRep in question, or assume ownership of its children, but must not modify the CordRep's data without explicitly adjusting the CRC.
PiperOrigin-RevId: 406191103
--
686544814079e5ab6d4593cca0c068b510be400a by Martijn Vels <mvels@google.com>:
Reduce the size in the LargeString test when running with Sanitizers
PiperOrigin-RevId: 406186945
--
735b4490bdb695c35731f06ce4b8de14ce2be6ed by Alex Strelnikov <strel@google.com>:
Release absl::SimpleHexAtoi.
PiperOrigin-RevId: 406143188
GitOrigin-RevId: f49e405201d2ffd5955503fa8ad0f08ec0cdfb2b
Change-Id: Ic6527ac40fa03ea02ca813e8bb7868a219544de4
3 years ago
|
|
|
tree = CordRepBtree::RemoveSuffix(tree->btree(), n);
|
|
|
|
} else if (!tree->IsExternal() && tree->refcount.IsOne()) {
|
|
|
|
assert(tree->IsFlat() || tree->IsSubstring());
|
|
|
|
tree->length -= n;
|
|
|
|
} else {
|
|
|
|
CordRep* rep = CordRepSubstring::Substring(tree, 0, tree->length - n);
|
|
|
|
CordRep::Unref(tree);
|
|
|
|
tree = rep;
|
|
|
|
}
|
|
|
|
contents_.SetTreeOrEmpty(tree, scope);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
Cord Cord::Subcord(size_t pos, size_t new_size) const {
|
|
|
|
Cord sub_cord;
|
|
|
|
size_t length = size();
|
|
|
|
if (pos > length) pos = length;
|
|
|
|
if (new_size > length - pos) new_size = length - pos;
|
|
|
|
if (new_size == 0) return sub_cord;
|
|
|
|
|
|
|
|
CordRep* tree = contents_.tree();
|
|
|
|
if (tree == nullptr) {
|
Export of internal Abseil changes
--
07240ca7822d007cdcc79f2c40bd58b2c2010348 by Abseil Team <absl-team@google.com>:
Correct the comment from "AlphaNum" to "Arg".
PiperOrigin-RevId: 416139192
--
adcba4a6b3763626e1db7b1e8c108b3114903557 by Martijn Vels <mvels@google.com>:
Fix NewExternalRep() to require data being non-empty, and remove nullptr return.
PiperOrigin-RevId: 416135865
--
c0d14cd918fb16f15d1d84de9284b5c5ecc1f8f2 by Abseil Team <absl-team@google.com>:
Fix doc comment for absl::ascii_isprint().
The comment was incorrectly saying that it includes all whitespace.
It doesn't; the only whitespace char it includes is ' '.
PiperOrigin-RevId: 416112524
--
d83327800159c07002b6865e21232a12463e02dd by Abseil Team <absl-team@google.com>:
Internal change
PiperOrigin-RevId: 416099978
--
baf11e9ca42ca9140cdbf8075f971db8d65b1195 by Ilya Tokar <tokarip@google.com>:
Prevent compiler from optimizing Group_Match* benchmarks away.
Currently we benchmark single store of precomputed value.
Not all affected benchmarks show performance changes:
BM_Group_Match 0.53ns ± 1% 0.53ns ± 0% -0.42% (p=0.038 n=10+10)
BM_Group_MatchEmpty 0.26ns ± 1% 0.26ns ± 1% ~ (p=1.000 n=10+10)
BM_Group_MatchEmptyOrDeleted 0.26ns ± 1% 0.26ns ± 1% ~ (p=0.121 n=10+10)
BM_Group_CountLeadingEmptyOrDeleted 0.26ns ± 1% 0.45ns ± 0% +70.05% (p=0.000 n=10+8)
BM_Group_MatchFirstEmptyOrDeleted 0.26ns ± 0% 0.44ns ± 1% +65.91% (p=0.000 n=8+9)
But inspecting the generated code shows the difference,
e. g. BM_Group_MatchFirstEmptyOrDeleted
Before:
add $0xffffffffffffffff,%rbx
jne 30
After:
pcmpeqd %xmm0,%xmm0
pcmpgtb -0x30(%rbp),%xmm0
pmovmskb %xmm0,%eax
add: 0x23$0xffffffffffffffff,%rbx
jne 40
PiperOrigin-RevId: 416083515
--
122fbff893dc4571b3e75e4b241eb4495b925610 by Abseil Team <absl-team@google.com>:
Put namespace guard in ABSL_DECLARE_FLAG to make declaring a flag in a namespace a compiler error instead of a linker error.
PiperOrigin-RevId: 416036072
--
020fd8a20f5fa319e948846e003391fcb9e03868 by Ilya Tokar <tokarip@google.com>:
Make Cord::InlineRep::set_data unconditionally zero out memory.
Currently there is a single case where we don't zero out memory
as an optimization. Unconditional zeroing doesn't show any changes
in benchmarks, except for the unrelated improvement:
BM_CordPartialCopyToCord/1M/1 12.6ns ± 4% 12.6ns ± 4% ~ (p=0.857 n=16+19)
BM_CordPartialCopyToCord/1M/128 44.9ns ± 7% 45.0ns ± 3% ~ (p=0.468 n=18+17)
BM_CordPartialCopyToCord/1M/1k 64.5ns ± 4% 61.4ns ± 4% -4.82% (p=0.000 n=19+17)
BM_CordPartialCopyToCord/1M/8k 139ns ± 3% 128ns ±15% -7.76% (p=0.009 n=17+20)
BM_CordPartialCopyToCord/1M/16k 193ns ± 6% 168ns ± 6% -13.17% (p=0.000 n=17+17)
BM_CordPartialCopyToCord/4M/16k 199ns ± 4% 177ns ± 4% -11.36% (p=0.000 n=17+18)
BM_CordPartialCopyToCord/4M/32k 275ns ± 3% 250ns ± 4% -9.00% (p=0.000 n=18+18)
BM_CordPartialCopyToCord/4M/64k 291ns ± 4% 266ns ± 5% -8.53% (p=0.000 n=18+16)
BM_CordPartialCopyToCord/4M/128k 322ns ± 5% 291ns ± 4% -9.43% (p=0.000 n=20+18)
BM_CordPartialCopyToCord/8M/32k 281ns ± 5% 251ns ± 4% -10.38% (p=0.000 n=20+16)
BM_CordPartialCopyToCord/8M/64k 293ns ± 6% 267ns ± 4% -8.87% (p=0.000 n=16+19)
BM_CordPartialCopyToCord/8M/128k 334ns ± 3% 305ns ± 2% -8.56% (p=0.000 n=17+16)
This is clearly an alignmnet effect since number of the executed instructions is the same:
M_CordPartialCopyToCord/1M/1 155 ± 0% 155 ± 0% ~ (all samples are equal)
BM_CordPartialCopyToCord/1M/128 446 ± 0% 446 ± 0% ~ (p=0.332 n=36+39)
BM_CordPartialCopyToCord/1M/1k 473 ± 0% 473 ± 0% ~ (p=0.969 n=40+40)
BM_CordPartialCopyToCord/1M/8k 808 ± 0% 808 ± 0% ~ (p=0.127 n=40+39)
BM_CordPartialCopyToCord/1M/16k 957 ± 0% 957 ± 0% ~ (p=0.532 n=40+40)
BM_CordPartialCopyToCord/4M/16k 952 ± 0% 952 ± 0% ~ (p=0.686 n=39+39)
BM_CordPartialCopyToCord/4M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.690 n=40+40)
BM_CordPartialCopyToCord/4M/64k 1.23k ± 0% 1.23k ± 0% ~ (p=0.182 n=40+39)
BM_CordPartialCopyToCord/4M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.711 n=40+40)
BM_CordPartialCopyToCord/8M/32k 1.12k ± 0% 1.12k ± 0% ~ (p=0.697 n=40+40)
BM_CordPartialCopyToCord/8M/64k 1.23k ± 0% 1.23k ± 0% +0.00% (p=0.049 n=40+40)
BM_CordPartialCopyToCord/8M/128k 1.44k ± 0% 1.44k ± 0% ~ (p=0.507 n=40+40)
This makes code simpler and doesn't regress performance.
PiperOrigin-RevId: 415560574
--
37305b2690b31682088749e4d62f40d7095bdc54 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 415558737
--
86aaed569b9e743c1eb813a5f48def978a793db3 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415515201
--
6cdb8786cdcb4fa0b8a4b72fc98940877d1fdeff by Abseil Team <absl-team@google.com>:
Update SubmitMutexProfileData to accept wait_cycles instead of wait_timestamp
PiperOrigin-RevId: 415360871
--
9f979d307aa16ad09f214e04876cbe84395c0901 by Abseil Team <absl-team@google.com>:
absl::flat_hash_set compiles with -Wconversion -Wsign-compare
PiperOrigin-RevId: 415357498
--
9eceb14174708f15e61259d449b214a8a4c7f9e7 by Abseil Team <absl-team@google.com>:
Fix AddressIsReadable for the corner case of (aligned) addr == NULL.
PiperOrigin-RevId: 415307792
--
1a39ffe55898375e2d7f88c17c99db5a1b95b313 by Martijn Vels <mvels@google.com>:
Internal change
PiperOrigin-RevId: 415162872
--
64378549b110d5f5762185a5906c520fba70f0e7 by Abseil Team <absl-team@google.com>:
Fix a typo in the comments
PiperOrigin-RevId: 415088461
--
41aae8322e913b82710153c22b97c611fdb6e1fb by Abseil Team <absl-team@google.com>:
Switch from `connect` to `rt_sigreturn` -- the latter is much less problematic
for system call sandboxes.
PiperOrigin-RevId: 415073965
--
870c5e3388b6a35611bff538626fe7a1c8c87171 by Abseil Team <absl-team@google.com>:
Add ABSL_HAVE_HWADDRESS_SANITIZER and ABSL_HAVE_LEAK_SANITIZER
PiperOrigin-RevId: 414871189
--
f213ed60a66b58da7ac40555adfb1d529ff0a4db by Derek Mauro <dmauro@google.com>:
Remove reference to __SANITIZE_MEMORY__, which does not exist
It appears to have been copied by pattern matching from the ASAN/TSAN
code blocks.
https://github.com/gcc-mirror/gcc/blob/f47662204de27f7685699eeef89aa173ccf32d85/gcc/cppbuiltin.c#L79-L126
PiperOrigin-RevId: 414806587
--
b152891e73ab515f397ceb53f66c8ee2f33863ea by Abseil Team <absl-team@google.com>:
Rollback previous commit: SYS_open is not defined in certain environments.
PiperOrigin-RevId: 414521820
--
5a1cbb282331023902e1374dd0d920c4effbe47f by Abseil Team <absl-team@google.com>:
Use syscall(SYS_open, ...) instead of open() to avoid possible symbol
interposition.
Also add some warning notes.
PiperOrigin-RevId: 414508186
--
1824d6593612710aafdc599a89b0adced7d787f6 by Abseil Team <absl-team@google.com>:
Correct aarch64 macro check
The macro is __aarch64__, not __arch64__.
PiperOrigin-RevId: 414446225
--
a1536a57b64dfd53945d33a01cfc08b18c99c97b by Abseil Team <absl-team@google.com>:
Fix backwards comment in the last commit.
PiperOrigin-RevId: 414281214
--
11ac021ba779513667a31cf2563ddafc57d6d913 by Abseil Team <absl-team@google.com>:
AddressIsReadable() didn't work correctly on ARM when the given pointer was
misaligned at the end of the page.
Fix that by aligning the pointer on an 8-byte boundary before checking it.
PiperOrigin-RevId: 414203863
GitOrigin-RevId: 07240ca7822d007cdcc79f2c40bd58b2c2010348
Change-Id: If5f129194d59f5c9e5d84efd8cd9e17a70e072ab
3 years ago
|
|
|
sub_cord.contents_.set_data(contents_.data() + pos, new_size);
|
|
|
|
return sub_cord;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (new_size <= InlineRep::kMaxInline) {
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
char* dest = sub_cord.contents_.data_.as_chars();
|
|
|
|
Cord::ChunkIterator it = chunk_begin();
|
|
|
|
it.AdvanceBytes(pos);
|
|
|
|
size_t remaining_size = new_size;
|
|
|
|
while (remaining_size > it->size()) {
|
|
|
|
cord_internal::SmallMemmove(dest, it->data(), it->size());
|
|
|
|
remaining_size -= it->size();
|
|
|
|
dest += it->size();
|
|
|
|
++it;
|
|
|
|
}
|
|
|
|
cord_internal::SmallMemmove(dest, it->data(), remaining_size);
|
|
|
|
sub_cord.contents_.set_inline_size(new_size);
|
|
|
|
return sub_cord;
|
|
|
|
}
|
|
|
|
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
tree = cord_internal::SkipCrcNode(tree);
|
|
|
|
if (tree->IsBtree()) {
|
|
|
|
tree = tree->btree()->SubTree(pos, new_size);
|
|
|
|
} else {
|
|
|
|
tree = CordRepSubstring::Substring(tree, pos, new_size);
|
|
|
|
}
|
|
|
|
sub_cord.contents_.EmplaceTree(tree, contents_.data_,
|
|
|
|
CordzUpdateTracker::kSubCord);
|
|
|
|
return sub_cord;
|
|
|
|
}
|
|
|
|
|
|
|
|
// --------------------------------------------------------------------
|
|
|
|
// Comparators
|
|
|
|
|
|
|
|
namespace {
|
|
|
|
|
|
|
|
int ClampResult(int memcmp_res) {
|
|
|
|
return static_cast<int>(memcmp_res > 0) - static_cast<int>(memcmp_res < 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
int CompareChunks(absl::string_view* lhs, absl::string_view* rhs,
|
|
|
|
size_t* size_to_compare) {
|
|
|
|
size_t compared_size = std::min(lhs->size(), rhs->size());
|
|
|
|
assert(*size_to_compare >= compared_size);
|
|
|
|
*size_to_compare -= compared_size;
|
|
|
|
|
|
|
|
int memcmp_res = ::memcmp(lhs->data(), rhs->data(), compared_size);
|
|
|
|
if (memcmp_res != 0) return memcmp_res;
|
|
|
|
|
|
|
|
lhs->remove_prefix(compared_size);
|
|
|
|
rhs->remove_prefix(compared_size);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
// This overload set computes comparison results from memcmp result. This
|
|
|
|
// interface is used inside GenericCompare below. Differet implementations
|
|
|
|
// are specialized for int and bool. For int we clamp result to {-1, 0, 1}
|
|
|
|
// set. For bool we just interested in "value == 0".
|
|
|
|
template <typename ResultType>
|
|
|
|
ResultType ComputeCompareResult(int memcmp_res) {
|
|
|
|
return ClampResult(memcmp_res);
|
|
|
|
}
|
|
|
|
template <>
|
|
|
|
bool ComputeCompareResult<bool>(int memcmp_res) {
|
|
|
|
return memcmp_res == 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
} // namespace
|
|
|
|
|
|
|
|
// Helper routine. Locates the first flat or external chunk of the Cord without
|
|
|
|
// initializing the iterator, and returns a string_view referencing the data.
|
|
|
|
inline absl::string_view Cord::InlineRep::FindFlatStartPiece() const {
|
|
|
|
if (!is_tree()) {
|
|
|
|
return absl::string_view(data_.as_chars(), data_.inline_size());
|
|
|
|
}
|
|
|
|
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
CordRep* node = cord_internal::SkipCrcNode(tree());
|
|
|
|
if (node->IsFlat()) {
|
|
|
|
return absl::string_view(node->flat()->Data(), node->length);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (node->IsExternal()) {
|
|
|
|
return absl::string_view(node->external()->base, node->length);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (node->IsBtree()) {
|
|
|
|
CordRepBtree* tree = node->btree();
|
|
|
|
int height = tree->height();
|
|
|
|
while (--height >= 0) {
|
|
|
|
tree = tree->Edge(CordRepBtree::kFront)->btree();
|
|
|
|
}
|
|
|
|
return tree->Data(tree->begin());
|
|
|
|
}
|
|
|
|
|
|
|
|
// Get the child node if we encounter a SUBSTRING.
|
|
|
|
size_t offset = 0;
|
|
|
|
size_t length = node->length;
|
|
|
|
assert(length != 0);
|
|
|
|
|
|
|
|
if (node->IsSubstring()) {
|
|
|
|
offset = node->substring()->start;
|
|
|
|
node = node->substring()->child;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (node->IsFlat()) {
|
|
|
|
return absl::string_view(node->flat()->Data() + offset, length);
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(node->IsExternal() && "Expect FLAT or EXTERNAL node here");
|
|
|
|
|
|
|
|
return absl::string_view(node->external()->base + offset, length);
|
|
|
|
}
|
|
|
|
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
void Cord::SetExpectedChecksum(uint32_t crc) {
|
|
|
|
auto constexpr method = CordzUpdateTracker::kSetExpectedChecksum;
|
|
|
|
if (empty()) return;
|
|
|
|
|
|
|
|
if (!contents_.is_tree()) {
|
|
|
|
CordRep* rep = contents_.MakeFlatWithExtraCapacity(0);
|
|
|
|
rep = CordRepCrc::New(rep, crc);
|
|
|
|
contents_.EmplaceTree(rep, method);
|
|
|
|
} else {
|
|
|
|
const CordzUpdateScope scope(contents_.data_.cordz_info(), method);
|
|
|
|
CordRep* rep = CordRepCrc::New(contents_.data_.as_tree(), crc);
|
|
|
|
contents_.SetTree(rep, scope);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
absl::optional<uint32_t> Cord::ExpectedChecksum() const {
|
|
|
|
if (!contents_.is_tree() || !contents_.tree()->IsCrc()) {
|
|
|
|
return absl::nullopt;
|
|
|
|
}
|
|
|
|
return contents_.tree()->crc()->crc;
|
|
|
|
}
|
|
|
|
|
|
|
|
inline int Cord::CompareSlowPath(absl::string_view rhs, size_t compared_size,
|
|
|
|
size_t size_to_compare) const {
|
|
|
|
auto advance = [](Cord::ChunkIterator* it, absl::string_view* chunk) {
|
|
|
|
if (!chunk->empty()) return true;
|
|
|
|
++*it;
|
|
|
|
if (it->bytes_remaining_ == 0) return false;
|
|
|
|
*chunk = **it;
|
|
|
|
return true;
|
|
|
|
};
|
|
|
|
|
|
|
|
Cord::ChunkIterator lhs_it = chunk_begin();
|
|
|
|
|
|
|
|
// compared_size is inside first chunk.
|
|
|
|
absl::string_view lhs_chunk =
|
|
|
|
(lhs_it.bytes_remaining_ != 0) ? *lhs_it : absl::string_view();
|
|
|
|
assert(compared_size <= lhs_chunk.size());
|
|
|
|
assert(compared_size <= rhs.size());
|
|
|
|
lhs_chunk.remove_prefix(compared_size);
|
|
|
|
rhs.remove_prefix(compared_size);
|
|
|
|
size_to_compare -= compared_size; // skip already compared size.
|
|
|
|
|
|
|
|
while (advance(&lhs_it, &lhs_chunk) && !rhs.empty()) {
|
|
|
|
int comparison_result = CompareChunks(&lhs_chunk, &rhs, &size_to_compare);
|
|
|
|
if (comparison_result != 0) return comparison_result;
|
|
|
|
if (size_to_compare == 0) return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
return static_cast<int>(rhs.empty()) - static_cast<int>(lhs_chunk.empty());
|
|
|
|
}
|
|
|
|
|
|
|
|
inline int Cord::CompareSlowPath(const Cord& rhs, size_t compared_size,
|
|
|
|
size_t size_to_compare) const {
|
|
|
|
auto advance = [](Cord::ChunkIterator* it, absl::string_view* chunk) {
|
|
|
|
if (!chunk->empty()) return true;
|
|
|
|
++*it;
|
|
|
|
if (it->bytes_remaining_ == 0) return false;
|
|
|
|
*chunk = **it;
|
|
|
|
return true;
|
|
|
|
};
|
|
|
|
|
|
|
|
Cord::ChunkIterator lhs_it = chunk_begin();
|
|
|
|
Cord::ChunkIterator rhs_it = rhs.chunk_begin();
|
|
|
|
|
|
|
|
// compared_size is inside both first chunks.
|
|
|
|
absl::string_view lhs_chunk =
|
|
|
|
(lhs_it.bytes_remaining_ != 0) ? *lhs_it : absl::string_view();
|
|
|
|
absl::string_view rhs_chunk =
|
|
|
|
(rhs_it.bytes_remaining_ != 0) ? *rhs_it : absl::string_view();
|
|
|
|
assert(compared_size <= lhs_chunk.size());
|
|
|
|
assert(compared_size <= rhs_chunk.size());
|
|
|
|
lhs_chunk.remove_prefix(compared_size);
|
|
|
|
rhs_chunk.remove_prefix(compared_size);
|
|
|
|
size_to_compare -= compared_size; // skip already compared size.
|
|
|
|
|
|
|
|
while (advance(&lhs_it, &lhs_chunk) && advance(&rhs_it, &rhs_chunk)) {
|
|
|
|
int memcmp_res = CompareChunks(&lhs_chunk, &rhs_chunk, &size_to_compare);
|
|
|
|
if (memcmp_res != 0) return memcmp_res;
|
|
|
|
if (size_to_compare == 0) return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
return static_cast<int>(rhs_chunk.empty()) -
|
|
|
|
static_cast<int>(lhs_chunk.empty());
|
|
|
|
}
|
|
|
|
|
|
|
|
inline absl::string_view Cord::GetFirstChunk(const Cord& c) {
|
|
|
|
return c.contents_.FindFlatStartPiece();
|
|
|
|
}
|
|
|
|
inline absl::string_view Cord::GetFirstChunk(absl::string_view sv) {
|
|
|
|
return sv;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Compares up to 'size_to_compare' bytes of 'lhs' with 'rhs'. It is assumed
|
|
|
|
// that 'size_to_compare' is greater that size of smallest of first chunks.
|
|
|
|
template <typename ResultType, typename RHS>
|
|
|
|
ResultType GenericCompare(const Cord& lhs, const RHS& rhs,
|
|
|
|
size_t size_to_compare) {
|
|
|
|
absl::string_view lhs_chunk = Cord::GetFirstChunk(lhs);
|
|
|
|
absl::string_view rhs_chunk = Cord::GetFirstChunk(rhs);
|
|
|
|
|
|
|
|
size_t compared_size = std::min(lhs_chunk.size(), rhs_chunk.size());
|
|
|
|
assert(size_to_compare >= compared_size);
|
|
|
|
int memcmp_res = ::memcmp(lhs_chunk.data(), rhs_chunk.data(), compared_size);
|
|
|
|
if (compared_size == size_to_compare || memcmp_res != 0) {
|
|
|
|
return ComputeCompareResult<ResultType>(memcmp_res);
|
|
|
|
}
|
|
|
|
|
|
|
|
return ComputeCompareResult<ResultType>(
|
|
|
|
lhs.CompareSlowPath(rhs, compared_size, size_to_compare));
|
|
|
|
}
|
|
|
|
|
|
|
|
bool Cord::EqualsImpl(absl::string_view rhs, size_t size_to_compare) const {
|
|
|
|
return GenericCompare<bool>(*this, rhs, size_to_compare);
|
|
|
|
}
|
|
|
|
|
|
|
|
bool Cord::EqualsImpl(const Cord& rhs, size_t size_to_compare) const {
|
|
|
|
return GenericCompare<bool>(*this, rhs, size_to_compare);
|
|
|
|
}
|
|
|
|
|
|
|
|
template <typename RHS>
|
|
|
|
inline int SharedCompareImpl(const Cord& lhs, const RHS& rhs) {
|
|
|
|
size_t lhs_size = lhs.size();
|
|
|
|
size_t rhs_size = rhs.size();
|
|
|
|
if (lhs_size == rhs_size) {
|
|
|
|
return GenericCompare<int>(lhs, rhs, lhs_size);
|
|
|
|
}
|
|
|
|
if (lhs_size < rhs_size) {
|
|
|
|
auto data_comp_res = GenericCompare<int>(lhs, rhs, lhs_size);
|
|
|
|
return data_comp_res == 0 ? -1 : data_comp_res;
|
|
|
|
}
|
|
|
|
|
|
|
|
auto data_comp_res = GenericCompare<int>(lhs, rhs, rhs_size);
|
|
|
|
return data_comp_res == 0 ? +1 : data_comp_res;
|
|
|
|
}
|
|
|
|
|
|
|
|
int Cord::Compare(absl::string_view rhs) const {
|
|
|
|
return SharedCompareImpl(*this, rhs);
|
|
|
|
}
|
|
|
|
|
|
|
|
int Cord::CompareImpl(const Cord& rhs) const {
|
|
|
|
return SharedCompareImpl(*this, rhs);
|
|
|
|
}
|
|
|
|
|
|
|
|
bool Cord::EndsWith(absl::string_view rhs) const {
|
|
|
|
size_t my_size = size();
|
|
|
|
size_t rhs_size = rhs.size();
|
|
|
|
|
|
|
|
if (my_size < rhs_size) return false;
|
|
|
|
|
|
|
|
Cord tmp(*this);
|
|
|
|
tmp.RemovePrefix(my_size - rhs_size);
|
|
|
|
return tmp.EqualsImpl(rhs, rhs_size);
|
|
|
|
}
|
|
|
|
|
|
|
|
bool Cord::EndsWith(const Cord& rhs) const {
|
|
|
|
size_t my_size = size();
|
|
|
|
size_t rhs_size = rhs.size();
|
|
|
|
|
|
|
|
if (my_size < rhs_size) return false;
|
|
|
|
|
|
|
|
Cord tmp(*this);
|
|
|
|
tmp.RemovePrefix(my_size - rhs_size);
|
|
|
|
return tmp.EqualsImpl(rhs, rhs_size);
|
|
|
|
}
|
|
|
|
|
|
|
|
// --------------------------------------------------------------------
|
|
|
|
// Misc.
|
|
|
|
|
|
|
|
Cord::operator std::string() const {
|
|
|
|
std::string s;
|
|
|
|
absl::CopyCordToString(*this, &s);
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
void CopyCordToString(const Cord& src, std::string* dst) {
|
|
|
|
if (!src.contents_.is_tree()) {
|
|
|
|
src.contents_.CopyTo(dst);
|
|
|
|
} else {
|
|
|
|
absl::strings_internal::STLStringResizeUninitialized(dst, src.size());
|
|
|
|
src.CopyToArraySlowPath(&(*dst)[0]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void Cord::CopyToArraySlowPath(char* dst) const {
|
|
|
|
assert(contents_.is_tree());
|
|
|
|
absl::string_view fragment;
|
|
|
|
if (GetFlatAux(contents_.tree(), &fragment)) {
|
|
|
|
memcpy(dst, fragment.data(), fragment.size());
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
for (absl::string_view chunk : Chunks()) {
|
|
|
|
memcpy(dst, chunk.data(), chunk.size());
|
|
|
|
dst += chunk.size();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
Cord Cord::ChunkIterator::AdvanceAndReadBytes(size_t n) {
|
|
|
|
ABSL_HARDENING_ASSERT(bytes_remaining_ >= n &&
|
|
|
|
"Attempted to iterate past `end()`");
|
|
|
|
Cord subcord;
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
auto constexpr method = CordzUpdateTracker::kCordReader;
|
|
|
|
|
|
|
|
if (n <= InlineRep::kMaxInline) {
|
|
|
|
// Range to read fits in inline data. Flatten it.
|
|
|
|
char* data = subcord.contents_.set_data(n);
|
|
|
|
while (n > current_chunk_.size()) {
|
|
|
|
memcpy(data, current_chunk_.data(), current_chunk_.size());
|
|
|
|
data += current_chunk_.size();
|
|
|
|
n -= current_chunk_.size();
|
|
|
|
++*this;
|
|
|
|
}
|
|
|
|
memcpy(data, current_chunk_.data(), n);
|
|
|
|
if (n < current_chunk_.size()) {
|
|
|
|
RemoveChunkPrefix(n);
|
|
|
|
} else if (n > 0) {
|
|
|
|
++*this;
|
|
|
|
}
|
|
|
|
return subcord;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (btree_reader_) {
|
|
|
|
size_t chunk_size = current_chunk_.size();
|
|
|
|
if (n <= chunk_size && n <= kMaxBytesToCopy) {
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
subcord = Cord(current_chunk_.substr(0, n), method);
|
|
|
|
if (n < chunk_size) {
|
|
|
|
current_chunk_.remove_prefix(n);
|
|
|
|
} else {
|
|
|
|
current_chunk_ = btree_reader_.Next();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
CordRep* rep;
|
|
|
|
current_chunk_ = btree_reader_.Read(n, chunk_size, rep);
|
Export of internal Abseil changes
--
60b8e77be4bab1bbd3b4c3b70054879229634511 by Derek Mauro <dmauro@google.com>:
Use _MSVC_LANG for some C++ dialect checks since MSVC doesn't
set __cplusplus accurately by default.
https://devblogs.microsoft.com/cppblog/msvc-now-correctly-reports-__cplusplus/
See GitHub #722.
PiperOrigin-RevId: 371362181
--
5d736accdff04db0e722f377c0d79f2d3ed53263 by Martijn Vels <mvels@google.com>:
Fix the estimated memory size for CordRepExternal
PiperOrigin-RevId: 371350380
--
eaaa1d8a167aeca67a2aa3a098a2b61a9d72172f by Martijn Vels <mvels@google.com>:
Remove flakes by not enforcing re-allocated pointers do never match original
Tests that do multiple updates could end up with the original allocated pointer on a 2nd resize, so the 'EqIfPrivate' should not assume that if we do 'not' have the capacity that all following relocations will never match the original. We only care about 'pointer unchanged if private and there is capacity', trying to establish 'pointer changed at some point due to re-allocation; is pointless.
PiperOrigin-RevId: 371338965
--
d1837bee6bade1902b095c1cbf64231668bb84c5 by Martijn Vels <mvels@google.com>:
Undo inline of small data copy in cord
This leads to a performance regression as the code is not inlined (absent hard FDO inputs), and there are no suitable tail call options.
PiperOrigin-RevId: 371332332
--
06dc64b833069efc7d18b11df607c8c22be690da by Martijn Vels <mvels@google.com>:
Add final instrumentation for Cordz and remove 'old' cordz logic.
This change instruments the last cord function for cordz. It removes the 'old' functions: set_tree, replace_tree, UpdateCordzStatistics and RecordMetrics.
PiperOrigin-RevId: 371219909
--
a5e0be538579c603052feec03e6d9910c43ea787 by Martijn Vels <mvels@google.com>:
Extend the life of CordRep* if inside a snapshot
If a snapshot (potentially) includes the current CordzInfo, we need to extent the lifetime of the CordRep*, as the snapshot 'point in time' observation of the cord should ideally be preserved.
PiperOrigin-RevId: 371146151
--
74d77a89774cd6c8ecdeebee0193b294a39383d6 by Martijn Vels <mvels@google.com>:
Instrument std::string consuming methods: ctor, operator=, Append and Prepend
This change moves the 'steal into CordRep' logic into a separate function so we can use it directly in the ctor, operator assign and append and prepend, allowing Cordz instrumentation with the proper method attributes.
The assign operator is implemented in AssignLargeString leaving the dispatch inlined in cord.h (which as a side effects also allows clean tail calls in the AssignLargeString method)
PiperOrigin-RevId: 371094756
--
b39effc45266b7ce2e7f96caa3b16cb6e3acc2dd by Martijn Vels <mvels@google.com>:
Add Cordz instrumentation to CordReader
PiperOrigin-RevId: 370990181
GitOrigin-RevId: 60b8e77be4bab1bbd3b4c3b70054879229634511
Change-Id: I96af62e6f1a643e8b1228ae01e6c84e33706bb05
4 years ago
|
|
|
subcord.contents_.EmplaceTree(rep, method);
|
|
|
|
}
|
|
|
|
bytes_remaining_ -= n;
|
|
|
|
return subcord;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Short circuit if reading the entire data edge.
|
|
|
|
assert(current_leaf_ != nullptr);
|
|
|
|
if (n == current_leaf_->length) {
|
|
|
|
bytes_remaining_ = 0;
|
|
|
|
current_chunk_ = {};
|
|
|
|
CordRep* tree = CordRep::Ref(current_leaf_);
|
|
|
|
subcord.contents_.EmplaceTree(VerifyTree(tree), method);
|
|
|
|
return subcord;
|
|
|
|
}
|
|
|
|
|
|
|
|
// From this point on, we need a partial substring node.
|
|
|
|
// Get pointer to the underlying flat or external data payload and
|
|
|
|
// compute data pointer and offset into current flat or external.
|
|
|
|
CordRep* payload = current_leaf_->IsSubstring()
|
|
|
|
? current_leaf_->substring()->child
|
|
|
|
: current_leaf_;
|
|
|
|
const char* data = payload->IsExternal() ? payload->external()->base
|
|
|
|
: payload->flat()->Data();
|
|
|
|
const size_t offset = static_cast<size_t>(current_chunk_.data() - data);
|
|
|
|
|
|
|
|
auto* tree = CordRepSubstring::Substring(payload, offset, n);
|
|
|
|
subcord.contents_.EmplaceTree(VerifyTree(tree), method);
|
|
|
|
bytes_remaining_ -= n;
|
|
|
|
current_chunk_.remove_prefix(n);
|
|
|
|
return subcord;
|
|
|
|
}
|
|
|
|
|
|
|
|
char Cord::operator[](size_t i) const {
|
|
|
|
ABSL_HARDENING_ASSERT(i < size());
|
|
|
|
size_t offset = i;
|
|
|
|
const CordRep* rep = contents_.tree();
|
|
|
|
if (rep == nullptr) {
|
|
|
|
return contents_.data()[i];
|
|
|
|
}
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
rep = cord_internal::SkipCrcNode(rep);
|
|
|
|
while (true) {
|
|
|
|
assert(rep != nullptr);
|
|
|
|
assert(offset < rep->length);
|
|
|
|
if (rep->IsFlat()) {
|
|
|
|
// Get the "i"th character directly from the flat array.
|
|
|
|
return rep->flat()->Data()[offset];
|
|
|
|
} else if (rep->IsBtree()) {
|
|
|
|
return rep->btree()->GetCharacter(offset);
|
|
|
|
} else if (rep->IsExternal()) {
|
|
|
|
// Get the "i"th character from the external array.
|
|
|
|
return rep->external()->base[offset];
|
|
|
|
} else {
|
|
|
|
// This must be a substring a node, so bypass it to get to the child.
|
|
|
|
assert(rep->IsSubstring());
|
|
|
|
offset += rep->substring()->start;
|
|
|
|
rep = rep->substring()->child;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
absl::string_view Cord::FlattenSlowPath() {
|
|
|
|
assert(contents_.is_tree());
|
|
|
|
size_t total_size = size();
|
|
|
|
CordRep* new_rep;
|
|
|
|
char* new_buffer;
|
|
|
|
|
|
|
|
// Try to put the contents into a new flat rep. If they won't fit in the
|
|
|
|
// biggest possible flat node, use an external rep instead.
|
|
|
|
if (total_size <= kMaxFlatLength) {
|
|
|
|
new_rep = CordRepFlat::New(total_size);
|
|
|
|
new_rep->length = total_size;
|
|
|
|
new_buffer = new_rep->flat()->Data();
|
|
|
|
CopyToArraySlowPath(new_buffer);
|
|
|
|
} else {
|
|
|
|
new_buffer = std::allocator<char>().allocate(total_size);
|
|
|
|
CopyToArraySlowPath(new_buffer);
|
|
|
|
new_rep = absl::cord_internal::NewExternalRep(
|
|
|
|
absl::string_view(new_buffer, total_size), [](absl::string_view s) {
|
|
|
|
std::allocator<char>().deallocate(const_cast<char*>(s.data()),
|
|
|
|
s.size());
|
|
|
|
});
|
|
|
|
}
|
|
|
|
CordzUpdateScope scope(contents_.cordz_info(), CordzUpdateTracker::kFlatten);
|
|
|
|
CordRep::Unref(contents_.as_tree());
|
|
|
|
contents_.SetTree(new_rep, scope);
|
|
|
|
return absl::string_view(new_buffer, total_size);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* static */ bool Cord::GetFlatAux(CordRep* rep, absl::string_view* fragment) {
|
|
|
|
assert(rep != nullptr);
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
rep = cord_internal::SkipCrcNode(rep);
|
|
|
|
if (rep->IsFlat()) {
|
|
|
|
*fragment = absl::string_view(rep->flat()->Data(), rep->length);
|
|
|
|
return true;
|
|
|
|
} else if (rep->IsExternal()) {
|
|
|
|
*fragment = absl::string_view(rep->external()->base, rep->length);
|
|
|
|
return true;
|
|
|
|
} else if (rep->IsBtree()) {
|
|
|
|
return rep->btree()->IsFlat(fragment);
|
|
|
|
} else if (rep->IsSubstring()) {
|
|
|
|
CordRep* child = rep->substring()->child;
|
|
|
|
if (child->IsFlat()) {
|
|
|
|
*fragment = absl::string_view(
|
|
|
|
child->flat()->Data() + rep->substring()->start, rep->length);
|
|
|
|
return true;
|
|
|
|
} else if (child->IsExternal()) {
|
|
|
|
*fragment = absl::string_view(
|
|
|
|
child->external()->base + rep->substring()->start, rep->length);
|
|
|
|
return true;
|
|
|
|
} else if (child->IsBtree()) {
|
|
|
|
return child->btree()->IsFlat(rep->substring()->start, rep->length,
|
|
|
|
fragment);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* static */ void Cord::ForEachChunkAux(
|
|
|
|
absl::cord_internal::CordRep* rep,
|
|
|
|
absl::FunctionRef<void(absl::string_view)> callback) {
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
assert(rep != nullptr);
|
|
|
|
rep = cord_internal::SkipCrcNode(rep);
|
|
|
|
|
|
|
|
if (rep->IsBtree()) {
|
|
|
|
ChunkIterator it(rep), end;
|
|
|
|
while (it != end) {
|
|
|
|
callback(*it);
|
|
|
|
++it;
|
|
|
|
}
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// This is a leaf node, so invoke our callback.
|
Export of internal Abseil changes
--
83e4cdf03a4d702b30e69204060de09e462e23c6 by Greg Falcon <gfalcon@google.com>:
Revert the crc addition to RefcountAndFlags, and restore related comments to their original state.
In development, the implementation of SetExpectedCrc() changed, and there is no longer a need to track the CRC status in the refcount.
Since the distinction between IsOne() and IsMutable() is subtle *and unused*, removing it now can help avoid subtle bugs in the future. This distinction can always be added back later, if it proves necessary.
Keep the reserved bit for now; all it costs is one extra mask instruction in the refcount checks, and space for extra state in Cord is always hard to find.
PiperOrigin-RevId: 408647038
--
ee67585cf66954176615271f50f8b278119dd138 by Greg Falcon <gfalcon@google.com>:
Implement Cord::SetExpectedChecksum() and Cord::ExpectedChecksum().
SetExpectedChecksum() will store a uint32_t out-of-band alongside a Cord's data. This value persists through copies and assignments. Mutating operations on a Cord cause the value to be forgotten. ExpectedChecksum() retrieves the stored value, if present.
This API is intended for storing a CRC32C checksum alongside data, allowing checksums to be passed through dataflows and validated at the final step. However, this API is agnostic to the meaning of the stored value. No CRC32C validation is performed by these new APIs.
This implementation adds a new CordRep node, CordRepCrc. A CordRepCrc may (currently) only live at the top of a tree. This allows traversal logic to be agnostic to these nodes, instead putting the needed branches at the mutation level. This also implements the property requested from API review, that any mutation is guaranteed to permanently forget the stored CRC.
PiperOrigin-RevId: 408611221
--
a86f592402b37c854ebdc77d2b9b425451a7a675 by Martijn Vels <mvels@google.com>:
Move 'ExtractResult' into CordRep
The result of an extract operation is logically identical for any tree implementation, and having a single type makes 'tree independent' implementation in cord.cc more concise.
PiperOrigin-RevId: 408332408
--
baa7647e21db59a87f75af9cac62172ce38a0f71 by Abseil Team <absl-team@google.com>:
Replace usages of `assert` macros with `ABSL_HARDENING_ASSERT`.
PiperOrigin-RevId: 408272133
--
c7658133d8662c39fa5035fc93a364c7c3d327e0 by Martijn Vels <mvels@google.com>:
Add CordRepBtree::ExtractAppendBuffer
PiperOrigin-RevId: 407944179
--
5775100363b5890ebfe710fadebf040445eab991 by Martijn Vels <mvels@google.com>:
Add CordRepConcat::ExtractAppendBuffer
PiperOrigin-RevId: 407932968
--
9f520ba1600a93352c78f644a369c7c76195ee86 by Greg Falcon <gfalcon@google.com>:
Add cordz tracking for crc nodes.
This also adds a new kSetExpectedChecksum method to the list of tracked methods. This is presently unused but will be used soon.
PiperOrigin-RevId: 407884120
GitOrigin-RevId: 83e4cdf03a4d702b30e69204060de09e462e23c6
Change-Id: I134ace2d87215813eaa60a282996a33884676c06
3 years ago
|
|
|
absl::cord_internal::CordRep* current_node = cord_internal::SkipCrcNode(rep);
|
|
|
|
absl::string_view chunk;
|
|
|
|
bool success = GetFlatAux(current_node, &chunk);
|
|
|
|
assert(success);
|
|
|
|
if (success) {
|
|
|
|
callback(chunk);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void DumpNode(CordRep* rep, bool include_data, std::ostream* os,
|
|
|
|
int indent) {
|
|
|
|
const int kIndentStep = 1;
|
|
|
|
absl::InlinedVector<CordRep*, kInlinedVectorSize> stack;
|
|
|
|
absl::InlinedVector<int, kInlinedVectorSize> indents;
|
|
|
|
for (;;) {
|
|
|
|
*os << std::setw(3) << rep->refcount.Get();
|
|
|
|
*os << " " << std::setw(7) << rep->length;
|
|
|
|
*os << " [";
|
|
|
|
if (include_data) *os << static_cast<void*>(rep);
|
|
|
|
*os << "]";
|
|
|
|
*os << " " << std::setw(indent) << "";
|
|
|
|
if (rep->IsCrc()) {
|
|
|
|
*os << "CRC crc=" << rep->crc()->crc << "\n";
|
|
|
|
indent += kIndentStep;
|
|
|
|
rep = rep->crc()->child;
|
|
|
|
} else if (rep->IsSubstring()) {
|
|
|
|
*os << "SUBSTRING @ " << rep->substring()->start << "\n";
|
|
|
|
indent += kIndentStep;
|
|
|
|
rep = rep->substring()->child;
|
|
|
|
} else { // Leaf or ring
|
|
|
|
if (rep->IsExternal()) {
|
|
|
|
*os << "EXTERNAL [";
|
|
|
|
if (include_data)
|
|
|
|
*os << absl::CEscape(std::string(rep->external()->base, rep->length));
|
|
|
|
*os << "]\n";
|
|
|
|
} else if (rep->IsFlat()) {
|
|
|
|
*os << "FLAT cap=" << rep->flat()->Capacity() << " [";
|
|
|
|
if (include_data)
|
|
|
|
*os << absl::CEscape(std::string(rep->flat()->Data(), rep->length));
|
|
|
|
*os << "]\n";
|
|
|
|
} else {
|
|
|
|
CordRepBtree::Dump(rep, /*label=*/ "", include_data, *os);
|
|
|
|
}
|
|
|
|
if (stack.empty()) break;
|
|
|
|
rep = stack.back();
|
|
|
|
stack.pop_back();
|
|
|
|
indent = indents.back();
|
|
|
|
indents.pop_back();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
ABSL_INTERNAL_CHECK(indents.empty(), "");
|
|
|
|
}
|
|
|
|
|
|
|
|
static std::string ReportError(CordRep* root, CordRep* node) {
|
|
|
|
std::ostringstream buf;
|
|
|
|
buf << "Error at node " << node << " in:";
|
|
|
|
DumpNode(root, true, &buf);
|
|
|
|
return buf.str();
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool VerifyNode(CordRep* root, CordRep* start_node,
|
|
|
|
bool /* full_validation */) {
|
|
|
|
absl::InlinedVector<CordRep*, 2> worklist;
|
|
|
|
worklist.push_back(start_node);
|
|
|
|
do {
|
|
|
|
CordRep* node = worklist.back();
|
|
|
|
worklist.pop_back();
|
|
|
|
|
|
|
|
ABSL_INTERNAL_CHECK(node != nullptr, ReportError(root, node));
|
|
|
|
if (node != root) {
|
|
|
|
ABSL_INTERNAL_CHECK(node->length != 0, ReportError(root, node));
|
|
|
|
ABSL_INTERNAL_CHECK(!node->IsCrc(), ReportError(root, node));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (node->IsFlat()) {
|
|
|
|
ABSL_INTERNAL_CHECK(node->length <= node->flat()->Capacity(),
|
|
|
|
ReportError(root, node));
|
|
|
|
} else if (node->IsExternal()) {
|
|
|
|
ABSL_INTERNAL_CHECK(node->external()->base != nullptr,
|
|
|
|
ReportError(root, node));
|
|
|
|
} else if (node->IsSubstring()) {
|
|
|
|
ABSL_INTERNAL_CHECK(
|
|
|
|
node->substring()->start < node->substring()->child->length,
|
|
|
|
ReportError(root, node));
|
|
|
|
ABSL_INTERNAL_CHECK(node->substring()->start + node->length <=
|
|
|
|
node->substring()->child->length,
|
|
|
|
ReportError(root, node));
|
|
|
|
} else if (node->IsCrc()) {
|
|
|
|
ABSL_INTERNAL_CHECK(node->crc()->child != nullptr,
|
|
|
|
ReportError(root, node));
|
|
|
|
ABSL_INTERNAL_CHECK(node->crc()->length == node->crc()->child->length,
|
|
|
|
ReportError(root, node));
|
|
|
|
worklist.push_back(node->crc()->child);
|
|
|
|
}
|
|
|
|
} while (!worklist.empty());
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::ostream& operator<<(std::ostream& out, const Cord& cord) {
|
|
|
|
for (absl::string_view chunk : cord.Chunks()) {
|
|
|
|
out.write(chunk.data(), static_cast<std::streamsize>(chunk.size()));
|
|
|
|
}
|
|
|
|
return out;
|
|
|
|
}
|
|
|
|
|
|
|
|
namespace strings_internal {
|
|
|
|
size_t CordTestAccess::FlatOverhead() { return cord_internal::kFlatOverhead; }
|
|
|
|
size_t CordTestAccess::MaxFlatLength() { return cord_internal::kMaxFlatLength; }
|
|
|
|
size_t CordTestAccess::FlatTagToLength(uint8_t tag) {
|
|
|
|
return cord_internal::TagToLength(tag);
|
|
|
|
}
|
|
|
|
uint8_t CordTestAccess::LengthToTag(size_t s) {
|
|
|
|
ABSL_INTERNAL_CHECK(s <= kMaxFlatLength, absl::StrCat("Invalid length ", s));
|
|
|
|
return cord_internal::AllocatedSizeToTag(s + cord_internal::kFlatOverhead);
|
|
|
|
}
|
|
|
|
size_t CordTestAccess::SizeofCordRepExternal() {
|
|
|
|
return sizeof(CordRepExternal);
|
|
|
|
}
|
|
|
|
size_t CordTestAccess::SizeofCordRepSubstring() {
|
|
|
|
return sizeof(CordRepSubstring);
|
|
|
|
}
|
|
|
|
} // namespace strings_internal
|
|
|
|
ABSL_NAMESPACE_END
|
|
|
|
} // namespace absl
|