* Added -Wextra and -Wshorten-64-to-32 and fixed resulting errors.
* Disable -Wshorten-32-to-64 since Kokoro is missing Clang.
* Fixed -Wextra warnings for gcc.
* Reordered UPB_UNUSED() to come after declarations.
* Added another -pedantic fix and log CC version.
* Fix compile error and conditionally run use_bazel.sh.
* Moved set -e after use_bazel.sh.
* Fixed typo in conditional.
* WIP.
* WIP.
* Tests are passing.
* Recover some perf: LIKELY doesn't propagate through functions. :(
* Added some more benchmarks.
* Simplify & optimize upb_arena_realloc().
* Only add owned blocks to the freelist.
* More optimization/simplification.
* Re-fixed the bug.
* Revert unintentional changes to parser.rl.
* Revert Lua changes for now.
* Revert the arena fuse changes for now.
* Added last_size to the arena representation.
* Re-applied Lua changes.
* Implemented upb_arena_fuse().
* Fix the compile by re-ordering statements.
* Improve comments.
* WIP.
* WIP.
* Tests are passing.
* Recover some perf: LIKELY doesn't propagate through functions. :(
* Added some more benchmarks.
* Simplify & optimize upb_arena_realloc().
* Only add owned blocks to the freelist.
* More optimization/simplification.
* Re-fixed the bug.
* Revert unintentional changes to parser.rl.
* Revert Lua changes for now.
* Revert the arena fuse changes for now.
* Added last_size to the arena representation.
* Fixed compile errors.
* Fixed compile error and changed benchmarks to do one allocation.
New code is smaller (in both source size and compiled size) and faster.
# Speed
The decoder speeds up on all machines I tested, though the amount of speedup varies. I was only able to test Intel CPUs.
### Linux Desktop
```
CPU: Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz
OS: Linux
name old time/op new time/op delta
CreateArena 4.72ns ± 0% 4.93ns ± 0% +4.47% (p=0.000 n=11+11)
ParseDescriptor 12.4µs ± 1% 9.1µs ± 1% -26.65% (p=0.000 n=11+11)
```
### Mac Laptop
```
CPU: Intel(R) Core(TM) i7-8850H CPU @ 2.60GHz
OS: macOS
name old time/op new time/op delta
CreateArena 5.33ns ± 3% 5.58ns ± 2% +4.69% (p=0.000 n=12+12)
ParseDescriptor 15.0µs ± 2% 11.9µs ± 2% -20.20% (p=0.000 n=12+12)
```
### Linux Workstation
```
CPU: Intel(R) Xeon(R) Gold 6154 CPU @ 3.00GHz
OS: Linux
name old time/op new time/op delta
CreateArena 5.29ns ± 0% 5.52ns ± 0% +4.37% (p=0.000 n=10+12)
ParseDescriptor 18.6µs ± 0% 16.4µs ± 0% -11.54% (p=0.000 n=12+12)
```
# Size
A few source files grow marginally because of some arena functionality moved inline. But `upb/decode.c` shrinks by 30% on Linux:
```
VM SIZE
--------------
+2.1% +283 upb/json_decode.c
+24% +205 upb/msg.c
+8.4% +115 upb/upb.c
+0.9% +28 upb/reflection.c
[ = ] 0 upb/def.c
[ = ] 0 upb/encode.c
[ = ] 0 upb/json_encode.c
[ = ] 0 upb/table.c
-30.3% -1.51Ki upb/decode.c
-0.7% -738 TOTAL
```
* Split upb::Arena/upb::Allocator from upb::Environment.
This will allow arenas and allocators to be used
independently of environments, which will be important
for an upcoming change (a message representation).
Overall this design feels cleaner that the previous
Environment/SeededAllocator design.
As part of this change, moved all allocations in upb
to use a global allocator instead of hard-coding
malloc/free. This will allow injecting OOM faults
for more robust testing.
One place that doesn't use the global allocator is
the tracked ref code. Instead of its previous approach
of CHECK_OOM() after every malloc() or table insert, it
simply uses an allocator that does this automatically.
I moved Allocator/Arena/Environment into upb.h.
This seems principled since these are the only types
in upb whose size is directly exposed to users, since
they form the basis of memory allocation strategy.
* Cleaned up some header includes and fixed more malloc -> upb_gmalloc().
* Changes from PR review.
* Don't use UINTPTR_MAX or UINT64_MAX.
* Punt on adding line/file for now.
* We actually can't store (uint64_t)-1, update comment and test.
A large part of this change contains surface-level
porting, like moving variable declarations to the
top of the block.
However there are a few more substantial things too:
- moved internal-only struct definitions to a separate
file (structdefs.int.h), for greater encapsulation
and ABI compatibility.
- removed the UPB_UPCAST macro, since it requires access
to the internal-only struct definitions. Replaced uses
with calls to inline, type-safe casting functions.
- removed the UPB_DEFINE_CLASS/UPB_DEFINE_STRUCT macros.
Class and struct definitions are now more explicit -- you
get to see the actual class/struct keywords in the source.
The casting convenience functions have been moved into
UPB_DECLARE_DERIVED_TYPE() and UPB_DECLARE_DERIVED_TYPE2().
- the new way that we duplicate base methods in derived types
is also more convenient and requires less duplication.
It is also less greppable, but hopefully that is not
too big a problem.
Compiler flags (-std=c89 -pedantic) should help to rigorously
enforce that the code is free of C99-isms.
A few functions are not available in C89 (strtoll). There
are temporary, hacky solutions in place.
- rewritten decoder; interpreted decoder is bytecode-based,
JIT decoder no longer falls back to the interpreter.
- C++ improvements: C++11-compatible iterators, upb::reffed_ptr
for RAII refcounting, better upcast/downcast support.
- removed the gross upb_value abstraction from public upb.h.
Many things have changed and been simplified.
The memory-management story for upb_def and upb_handlers
is much more robust; upb_def and upb_handlers should be
fairly stable interfaces now. There is still much work
to do for the runtime component (upb_sink).
Many improvements, too many to mention. One significant
perf regression warrants investigation:
omitfp.parsetoproto2_googlemessage1.upb_jit: 343 -> 252 (-26.53)
plain.parsetoproto2_googlemessage1.upb_jit: 334 -> 251 (-24.85)
25% regression for this benchmark is bad, but since I don't think
there's any fundamental design issue that caused it I'm going to
go ahead with the commit anyway. Can investigate and fix later.
Other benchmarks were neutral or showed slight improvement.
Added a upb_byteregion that tracks a region of
the input buffer; decoders use this instead of
using a upb_bytesrc directly. upb_byteregion
is also used as the way of passing a string to
a upb_handlers callback. This symmetry makes
decoders compose better; if you want to take
a parsed string and decode it as something else,
you can take the string directly from the callback
and feed it as input to another parser.
A commented-out version of a pinning interface
is present; I decline to actually implement it
(and accept its extra complexity) until/unless
it is clear that it is actually a win. But it
is included as a proof-of-concept, to show that
it fits well with the existing interface.
Includes are now via upb/foo.h.
Files specific to the protobuf format are
now in upb/pb (the core library is concerned
with message definitions, handlers, and
byte streams, but knows nothing about any
particular serializationf format).
This doesn't reflect any material change in
how I will be working on upb, and I have no
problem making this change. It's still open
source under the BSD license, and I'll still
be working on it well beyond the hours that
constitute a normal job.
This is a significant change to the upb_stream
protocol, and should hopefully be the last
significant change.
All callbacks are now registered ahead-of-time
instead of having delegated callbacks registered
at runtime, which makes it much easier to
aggressively optimize ahead-of-time (like with a
JIT).
Other impacts of this change:
- You no longer need to have loaded descriptor.proto
as a upb_def to load other descriptors! This means
the special-case code we used for bootstrapping is
no longer necessary, and we no longer need to link
the descriptor for descriptor.proto into upb.
- A client can now register any upb_value as what
will be delivered to their value callback, not
just a upb_fielddef*. This should allow for other
clients to get more bang out of the streaming
decoder.
This change unfortunately causes a bit of a performance
regression -- I think largely due to highly
suboptimal code that GCC generates when structs
are returned by value. See:
http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/
On the other hand, once we have a JIT this should
no longer matter.
Performance numbers:
plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88)
plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11)
plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37)
plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12)
plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47)
plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42)
omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07)
omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87)
omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74)
omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10)
omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40)
omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
Joshua Haberman
Josh Haberman
Joshua Haberman