Decompose RandenPoolSeedSeq from NonsecureURBGBase.
Adjust how the RandenPoolSeedSeq detects contiguous buffers passed to the generate function. Previously it made incorrect assumptions regarding the contiguous concept, which have been replaced with some type-based tests for a small number of known contiguous random access iterator types, including raw pointers.
PiperOrigin-RevId: 452564114
Change-Id: Idab1df9dd078d8e5c565c7fa7ccb9c0d3d392ad2
An external project may depend on an Abseil `TESTONLY` library like
`absl::random_mocking_bit_gen`. Such a library should not be excluded
from build when Abseil's own tests are not built.
fixes#997
--
a0847bf19789c97689f1a8b0133a53b8af5e5caa by Samuel Benzaquen <sbenza@google.com>:
Add support for absl::(u)int128 to random distributions and generators.
PiperOrigin-RevId: 411565479
GitOrigin-RevId: a0847bf19789c97689f1a8b0133a53b8af5e5caa
Change-Id: Ide434bdd93fcab8e90f791796498de14833b667c
As of this change, you can use `-DABSL_USE_EXTERNAL_GOOGLETEST=ON -DABSL_FIND_GOOGLETEST=ON` to have Abseil use the standard CMake find_package(GTest) mechanism.
--
e2de21d54c02b6419c57c0f4e2a16b608deca260 by Evan Brown <ezb@google.com>:
Remove the InsertEnd benchmark.
This benchmark has significantly different possible behaviors that can result in misleading metrics. Specifically, we can have a case where we are deallocating the last node in the b-tree in the erase and then allocating a new node in the insert call repeatedly, whereas normally, we end up just inserting/erasing a value from the last node. Also, the name of the benchmark is misleading because it involves an erase and an insert, but the name only mentions the insert.
PiperOrigin-RevId: 360930639
--
51f6bb97b9cbdb809c31b77e93ce080ca3cba9ea by Benjamin Barenblat <bbaren@google.com>:
Stop testing with double-double random variables
On POWER, long double is often represented as a pair of doubles added
together (double-double arithmetic). We’ve already special-cased
double-double arithmetic in a number of tests, but compiler
bugs [1, 2, 3] have now triggered both false positives and false
negatives, which suggests testing with double doubles is unlikely to
yield useful signal. Remove the special casing and detect if we’re on a
double-double system; if so, just don’t test long doubles.
[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99048
[2] https://bugs.llvm.org/show_bug.cgi?id=49131
[3] https://bugs.llvm.org/show_bug.cgi?id=49132
PiperOrigin-RevId: 360793161
--
07fb4d7932c2f5d711c480f759dacb0be60f975e by Abseil Team <absl-team@google.com>:
internal change
PiperOrigin-RevId: 360712825
GitOrigin-RevId: e2de21d54c02b6419c57c0f4e2a16b608deca260
Change-Id: I98389b5a8789dcc8f35abc00c767e909181665f0
--
756156bf03da050e8b27539a8247d9af7e44c6a2 by Abseil Team <absl-team@google.com>:
Fix a typo in cord.h: "accomodate" => "accommodate"
PiperOrigin-RevId: 356168875
--
638befdb342b608ec28910ee931ee200fdbe1fef by Samuel Benzaquen <sbenza@google.com>:
Fix float conversion for PPC.
In PPC `long double` is a double-double representation which behaves weirdly
wrt numeric_limits. Don't take `long double` into account when we are not
handling `long double` natively anyway.
Fix the convert test to always run the conversion even if we are not going to
compare against libc's printf result. This allows exercising the code itself to
make sure we don't trigger assertions or UB found by sanitizers.
PiperOrigin-RevId: 355857729
--
ff5f893319fa76b273c7785b76ef6c95b1791076 by Abseil Team <absl-team@google.com>:
Example usage tweak
PiperOrigin-RevId: 355695750
--
0efc454f90023fa651b226e5e3ba7395a3b60c6d by Benjamin Barenblat <bbaren@google.com>:
Remove endian-sensitivity from Abseil’s RNG
Ensure that the Abseil random number generator produces identical output
on both big- and little-endian platforms by byte-swapping appropriately
on big-endian systems.
PiperOrigin-RevId: 355635051
GitOrigin-RevId: 756156bf03da050e8b27539a8247d9af7e44c6a2
Change-Id: Iaaa69767b8e85d626742b9ba56fefb75f07c69ee
--
dab5caab05d89d03066ef92584660688595a3aaf by Mark Barolak <mbar@google.com>:
Add absl::Status and absl::StatusOr to absl/README.md
Import of https://github.com/abseil/abseil-cpp/pull/863
PiperOrigin-RevId: 347857368
--
1ca3c7a96417cd6e6d62f4dc36fd5ddaa61cfa20 by Chris Kennelly <ckennelly@google.com>:
Leverage integer power-of-2 functions and bit counting library in Abseil.
PiperOrigin-RevId: 347816486
--
e5cbe05879fd65dce7875e2e0105331a1615d89b by Chris Kennelly <ckennelly@google.com>:
Mitigate narrowing warning on MSVC.
If sizeof(x) <= sizeof(uint32_t), no truncation occurs when casting to
uint32_t, but the compiler cannot always determine this.
PiperOrigin-RevId: 347696526
--
079dff64cb175d282d9e22dfb4a522199ffdae2e by Benjamin Barenblat <bbaren@google.com>:
Avoid libgcc -NaN narrowing bug
When testing -NaN parsing, avoid narrowing -NaN from double to float.
This avoids a bug in libgcc
(https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98251).
PiperOrigin-RevId: 347654751
--
2e78a7634865aeef6765e1f447e96cf8d9985059 by Chris Kennelly <ckennelly@google.com>:
Mark popcount helpers as inline.
These are conditionally constexpr, so we need to add inline to cover the
non-constexpr builds to avoid ODR violations.
PiperOrigin-RevId: 347620138
--
437fbb363aea1654179f102dcdd607ec33c1af1e by Chris Kennelly <ckennelly@google.com>:
Use explicit narrowing cast.
This is never invoked in practice, but compilers with -Wimplicit-int-conversion
may trigger when sizeof(T) > sizeof(uint16_t) prior to determining this never
runs.
PiperOrigin-RevId: 347609857
GitOrigin-RevId: dab5caab05d89d03066ef92584660688595a3aaf
Change-Id: I6296ddffe7ec646f8ce121138f21e1e85a2cff4b
--
0bfa836596a9c787a2f0bdc283011dd1f6810c6e by Benjamin Barenblat <bbaren@google.com>:
Ignore missing CPU frequency on more architectures
Linux on MIPS, PA-RISC, RISC-V, and SystemZ doesn’t expose the nominal
CPU frequency via /sys, so don’t worry if `NominalCPUFrequency` returns
1.0 on those platforms.
Some POWER machines expose the CPU frequency; others do not. Since we
can’t predict which type of machine the tests will run on, simply
disable testing for `NominalCPUFrequency` on POWER.
PiperOrigin-RevId: 347079873
--
492b6834ed4a07cbc3abccd846f7e37d8c556ee5 by Benjamin Barenblat <bbaren@google.com>:
Use ABSL_HAVE_THREAD_LOCAL macro instead of copying code
Reduce code duplication by checking the ABSL_HAVE_THREAD_LOCAL macro
instead of copying code from base/config.h.
PiperOrigin-RevId: 347079561
--
8d656efce4da9cb032094377e58493d98427a536 by Abseil Team <absl-team@google.com>:
Rollback
PiperOrigin-RevId: 347078779
--
221bc69ec6dd7e2777ffcff6942584f979ef6382 by Abseil Team <absl-team@google.com>:
Add flag for 'shallow subcord' feature for experimental ring buffer rollout
There is a potential trade-off of CPU cost vs over-sharing cord data for subcord of large cords. This flag allows making subcords shallow for ringbuffers (with a potential larger waste of referenced source cords), which allows us to make subcord fast for this apps that do no persist (unmodified / plain copied) sub cords.
This change also introduces constants for the default settings, intended to keep the internal cord settings concistent with external flags.
PiperOrigin-RevId: 347053271
--
00a56c24293566734009f6bf2169a83fb37a35ba by Abseil Team <absl-team@google.com>:
Revert the usage of variant<> in Cord iterator and reader.
The introduction of the variant may lead to some missed compiler optimizations.
PiperOrigin-RevId: 347053041
--
c7b7b5ed7e3ab46b1e75b80f1a7de0bda26c8f70 by Chris Kennelly <ckennelly@google.com>:
Release library for integer power-of-2 functions and bit counting.
PiperOrigin-RevId: 347035065
--
5a035c0d9840b251967f9e7039fc6a4e01dd52f3 by Abseil Team <absl-team@google.com>:
Restructure Cord::ChunkIterator for future ring buffer support.
PiperOrigin-RevId: 346890054
GitOrigin-RevId: 0bfa836596a9c787a2f0bdc283011dd1f6810c6e
Change-Id: I3a58e2a44cb4c6f2116c43e2a4ccbc319d3ccecf
--
790f9061df340cd900e8da70e66c363f7af3c2eb by Abseil Team <absl-team@google.com>:
Add support for rvalue reference to function types.
PiperOrigin-RevId: 324508531
--
51fe201dbb41a3ebc3d49ff65250b5f464279d43 by Abseil Team <absl-team@google.com>:
Cleaning up function comment style; no substantive change.
PiperOrigin-RevId: 324497401
--
da8595d5266577d0c170528d12f6de17b8affcc2 by Abseil Team <absl-team@google.com>:
Add support for demangling GNU vector types.
PiperOrigin-RevId: 324494559
--
0cb0acf88c1750f6963c9cb85249f9b4f0bd5104 by Abseil Team <absl-team@google.com>:
Add support for thread-local types.
PiperOrigin-RevId: 324491183
--
c676bc8380560599cd26f7f231e04e6be532e904 by Abseil Team <absl-team@google.com>:
Add support for demangling "Du" (char8_t).
PiperOrigin-RevId: 324441607
--
b218bf6467bc62b327214782c881e8224ad91509 by Abseil Team <absl-team@google.com>:
Update doc comments in header of `any.h` to reflect that `absl::variant` has been released.
PiperOrigin-RevId: 324431690
--
e5b579f3f1aa598c1f62e71dba7103b98811de59 by Laramie Leavitt <lar@google.com>:
Bugfix: Fix bounds in absl::Uniform where one of the bounds is min/max.
When absl::Uniform(rng, tag, a, b) is called, the tag is used in conjunction with the type to determine whether or not to manipulate the bounds to make them inclusive or exclusive through the uniform_*_bound functions. Unfortunately, at limits of the interval the function was not well behaved.
The previous implementation used wrapping arithmetic. This causes incorrect bounds computation at the extremes (numeric_limits::min / numeric_limits::max) the bound would wrap.
Improve this situation by:
1/ Changing the uniform_*_bound functions to use saturating arithmetic instead of wrapping, thus in the unsigned case, the upper_bound of IntervalOpenOpen for 0 is now 0, rather than numeric_limits::max, likewise for the lower bound.
2/ Adjusting the hi/lo checks in the distributions. When the interval is empty, such as for absl::Uniform(absl::IntervalOpenOpen, gen, 1, 0), the return value is somewhat nonsensical. Now absl::Uniform more consistently returns the low input rather than any adjusted input. In the above case, that means that 1 is returned rather than 2.
NOTE: Calls to absl::Uniform where the resolved upper bound is < the lower bound are still ill-formed and should be avoided.
3/ Adding better tests.
The underlying uniform_*_distribution classes are not affected.
PiperOrigin-RevId: 324240873
GitOrigin-RevId: 790f9061df340cd900e8da70e66c363f7af3c2eb
Change-Id: I2a2208650ea3135c575e200b868ce1d275069fc8
--
e21e960918678629abf89ad1b694b7d4a456b434 by Greg Falcon <gfalcon@google.com>:
Roll back invoke() change due to large increases in compiler memory usage.
PiperOrigin-RevId: 315919455
--
f95872e1e1d7afdefbac94f42ea228d42d80eb6e by Greg Falcon <gfalcon@google.com>:
Rollback of invoke() changes due to compiler memory usage growth
PiperOrigin-RevId: 315911585
--
6c6c6ba6892016a2ce4703042800254fb9b15727 by Laramie Leavitt <lar@google.com>:
Move some of the common mocking code into MockHelpers.
Use MockHelpers to do mock signature detection and improve the dispatch mechansim.
PiperOrigin-RevId: 315825988
--
5e9380367d280c7fa6dbd4d0f48c31ade7f1d419 by Greg Falcon <gfalcon@google.com>:
Rename the internal implementation details Invoke and InvokeT to `invoke` and `invoke_result_t`, since these are re-implementations of C++17 library entites of the same names.
PiperOrigin-RevId: 315790467
GitOrigin-RevId: e21e960918678629abf89ad1b694b7d4a456b434
Change-Id: Ia75011f94cb033c1c9a4cb64cf14d283b91426ac
--
9e8b4a286d70df9487bff080816bd07ae38af5f8 by Evan Brown <ezb@google.com>:
Add btree_node::transfer_n/transfer_n_backward and replace usage of uninitialized_move_n and value_destroy_n.
PiperOrigin-RevId: 314600027
--
6c452aa1ee7e46ab941ba7d1fa636da8ea3d7370 by Laramie Leavitt <lar@google.com>:
Remove the MockingBitGenBase base class in favor of type-erasure in BitGenRef.
In Abseil random, mocking was split across two different classes,
MockingBitGenBase and MockingBitGen. This split existed because Google Mock is a
test-only library that we don't link into production, so MockingBitGenBase
provided a low-overhead scaffold used to lookup mocks when in test code, but
which is unused in production code.
That has been replaced by type-erasure which looks for a method named
CallImpl with the correct signature.
Weaken the coupling between MockingBitGen, DistributionCaller, and MockOverloadSet.
Rename CallImpl to InvokeMock()
Previously, the implementation of DistributionCaller was also split across different files using explicit instantiation of the DistributionCaller struct and some details in the Mocking classes. Now Distribution caller uses the presence of the InvokeMock() method to choose whether to use the mockable call path or the default call path.
PiperOrigin-RevId: 314584095
--
07853c47dc98698d67d65a3b9b662a65ab9def0a by Abseil Team <absl-team@google.com>:
Add PC / backtrace / symbolization support for Apple platforms.
Full backtrace support requires iOS 9+
PiperOrigin-RevId: 314415072
--
43889f17a132b31f6558c6482721cbbc776128fd by Gennadiy Rozental <rogeeff@google.com>:
Consolidate all reflection interface in the new module 'reflection' and expose interface to locate reflection handle by name.
PiperOrigin-RevId: 314390358
GitOrigin-RevId: 9e8b4a286d70df9487bff080816bd07ae38af5f8
Change-Id: I8e0910437740cf9ea9da5000adddfcef127e1158
--
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
--
f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>:
Slight restructuring of absl/random/internal randen implementation.
Convert round-keys.inc into randen_round_keys.cc file.
Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts.
Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed.
PiperOrigin-RevId: 312190313
--
dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>:
Internal change.
PiperOrigin-RevId: 312167304
--
f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>:
Cleanup StrFormat extensions a little.
PiperOrigin-RevId: 312166336
--
9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 312105213
--
9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>:
Complete IGNORE_TYPE macro renaming.
PiperOrigin-RevId: 311999699
--
64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>:
Switch to fixed bytes specific default value.
This fixes the Abseil Flags for big endian platforms.
PiperOrigin-RevId: 311844448
--
bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>:
Change many distribution tests to use pcg_engine as a deterministic source of entropy.
It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests.
Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values.
This changelist does the latter.
PiperOrigin-RevId: 311840096
GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266
Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
Russell Johnston
Laramie Leavitt
Andrei Polushin
Abseil Team
Florin Crișan