blob: 82003c79035a8b276cc0a4d1deda236fcf845e63 [file] [log] [blame]
Name: tcmalloc
Short Name: gperftools
URL: http://gperftools.googlecode.com/
Version: unknown
Revision: 144
Security Critical: yes
License: BSD
Description:
This contains Chromium's locally patched copy of tcmalloc.
Contents:
chromium/
The chromium patched sources, copied from the vendor/
subdirectory and containing our local modifications.
We only copy over the vendor/src/ subdirectory (the only piece
we need) but still leave it in a chromium/src/ subdirectory to
keep the directory structures in parallel.
vendor/
Vanilla sources from upstream:
http://gperftools.googlecode.com/svn/trunk
The current revision is:
Last Changed Rev: 144
Last Changed Date: 2012-02-04 00:10:11 +0000 (Sat, 04 Feb 2012)
HOWTOs:
Take a new version from upstream:
1) Grab the revision:
$ svn export [-r {tcmalloc-rev}] \
http://gperftools.googlecode.com/svn/trunk \
vendor-{tcmalloc-rev}
2) Check for added or deleted files:
$ diff -q -r -x .svn vendor vendor-{tcmalloc-rev}
3) Copy the new revision on top of the checked-in vendor branch:
$ cp -r vendor-{tcmalloc-rev}/* vendor
C:\> xcopy /e/y/i vendor-{tcmalloc-rev}\* vendor
4) Make all vendor files non-executable.
$ find . -executable -type f -exec chmod a-x {} \;
5) "svn add" or "svn rm" added or removed files (based on your
"diff -q -r" output from above)
6) Create the CL, upload, check it in:
$ gcl change CL
$ gcl upload CL
$ gcl commit CL
Note the revision number since you're going to want to merge
that to the local chromium branch.
Merge a new upstream version with our local patched copy:
1) Merge the local revision to chromium/src
$ svn merge -c {chrome-rev} svn://chrome-svn/chrome/trunk/src/third_party/tcmalloc/vendor/src chromium/src
2) Resolve any conflicts
3) Create the CL, upload, check in:
$ gcl change CL
$ gcl upload CL
$ gcl commit CL
Modifications:
- Converted to utf-8 with: vim +"argdo write ++enc=utf-8" *.h *.c
- Added support for android.
- Use NULL instead of static_cast<uintptr_t>(0) in stack_trace_table.cc,
for -std=c++11 compatibility.
- Added support for pseudo-stack heap profiling via a callback to retrieve a
simulated stack from the embedding application.
- Inserted spaces around PRIx64, SCNx64 and friends, for c++11 compatibility.
- Fix sprintf formatting warning in MaybeDumpProfileLocked
- Fix logging issues in android
- Changed DEFINE_foo macros to ignore envname unless ENABLE_PROFILING is defined
- Changed DEFINE_string to define const char*s instead of strings
- Disabled HEAPPROFILE envvar unless ENABLE_PROFILING is defined
- Add "ARMv8-a" to the supporting list of ARM architecture
- Add generic.total_physical_bytes property to MallocExtension
- Conditionally define HAVE_VDSO_SUPPORT only on linux_x86 to avoid static initializers
- Add TC_MALLOPT_IS_OVERRIDDEN_BY_TCMALLOC mallopt() arg
- Added tc_malloc_skip_new_handler.
- Added TCMALLOC_DONT_REPLACE_SYSTEM_ALLOC which bypasses the libc_override logic.
- Backported 7df7f14 "issue-693: enable futex usage on arm" from upstream.
- Don't use the tls model 'initial-exec' in chromeos on arm with gcc.
- Update addr2line-pdb.c to fix format string errors and use relative addresses
matching linux's behavior more closely.
- Changed kint64min to not depend on undefined behavior.
- Fix potential missing nul character in symbol names produced by addr2line-pdb.
- Remove superfluous size_t value >= 0 check.
- Make kFooType in tcmalloc.cc truly const.
- Added support for mips64el.
- Pulled SuggestedDelayNS() implementation for 32bit architectures which do not support 64bit atomicity
- Pulled several mipsel related changes from lss project to fix compile errors
- Fixed line endings in vendor/README_windows.txt to Unix (LF)
- Removed have_futex from spinlock_linux-inl.h; assume it's always there
- Add thread-safe support to query and update the heap sampling period
- Remove use of LINKER_INITIALIZED in page_heap_allocator.h, to enable the
compiler to optimize away a static constructor
- Include <atomic> in malloc_extension.h due to std::atomic usage