I changed order to avoid suggesting that std::deque has annotations not for all allocators.

Including suggestions.

Thx for fast review!

CI rerun and rebase.

I also don't think the issue is related to the patch.

Rebase and suggested changes.

This update removes __grow_by from _LIBCPP_STRING_UNSTABLE_EXTERN_TEMPLATE_LIST in libcxx/include/__string/extern_template_lists.h.

This update:

This patch reverts previous attempt and is fixing the preoblem.

Testing a potential fix.

To test the issue on CI, this update does:

• rebase,
• add _LIBCPP_SUPPRESS_DEPRECATED_PUSH and _LIBCPP_SUPPRESS_DEPRECATED_POP to not duplicate code.

Removing std::__debug_db_invalidate_all(this); from __grow_by_without_replace as it was removed form __grow_by.

Rebase.

A fix for rebase.

This update:

• applies comment change suggested by @ldionne (thank you!),
• resolves conflicts.

I hope I fixed everything requested, cannot find anything else to fix. I hope this time it landed for the last time...
I see buildbot failures, but I don't see how my patch may be the cause (one, two).

Rebase.

Rebase.

rebase

rebase

rebase

Updating libcxx/include/__config in response to _LIBCPP_FUNC_VIS removal.

Rebase.

This update improves a comment.

This update adds tests for the patch.

• Test for different allocators.
• Test confirming that turning off annotations work.

As suggested, this update:

• reverts changes in libcxx/include/__string/extern_template_lists.h,
• in a comment, describes the bug and explains why the function is preserved,
• adds _LIBCPP_HIDE_FROM_ABI to __grow_by_without_replace (),
• removes changes in libcxx/lib/abi.

rebase

To the best of my knowledge, that implementation is correct.

• The same value as in __grow_by_and_replace.
• D132769 is working with that patch.
• I tested it with more than unit tests.

This update:

• fixes annotations after changes in D148693,
• does rebase.

rebase

rebase

This update:

• removes XFAIL as the test should always pass,
• removes _LIBCPP_ENABLE_ASSERTIONS=1 because ASan assertions are catched anyway.

Rebase, I hope it will fix the issue with the fail on MacOS.

rebase

rebase, I hope it fixes the CI fail (doesn't look like issue made by this revision).

Rebase.

This update fixes a naming issue.

Hey @vitalybuka, thanks for the standalone failing test (D151836)! It was really helpful! I fixed the problem and hope there are no more issues with that revision (and for no more reverts).

This update renames the new test and updates it slightly.

This updates the base commit.

This update changes how __annotate_delete() for an empty container is executed and makrs two functions as const.
It solves issues with non-standard allocators.

This update applies changes to the newest version of D132092.
It also adds tests similar to those in D136765.

This update:

• adds a test suggested by @vitalybuka (extended to do that kind of caterpillar from every side),
• removes the issue,
• reimplementes __annotate_whole_block (last arguments removed; no longer calling __annotate_from_to).

This update (hopefully) brings back old behavior of deque with const elements.

Because D132092 got reverted, I will include that change there.

This update resolves merge conflicts with new head.

Ok, I just learned that deque objects have to be assignable/movable, so std::deque<const int> is not correct (from what I understand). Therefore tests for it cannot exist. But it's not my intention to change behavior of std::deque<const T> and therefore I created D151777. The new patch adds a few const. All examples working previously seem to work with D151777, but I found some not working even before this patch.

@alexfh it's not by design. A const_cast is missing (or should I use (void *)?). We may add it in the very place where error happens or later before calling API, but type has to be changed here.

This update resolves merge conflicts with the new head commin on the main branch.

I hope I understood all comments. I'm in the middle of testing right now.

This update fixes issues mentioned in the code review:

• adds _LIBCPP_DEPRECATED_("use __grow_by_without_replace"),
• adds _LIBCPP_HIDE_FROM_ABI_AFTER_V1,
• changes deprecated call to __grow_by in __grow_by_without_replace with a copy-paste code of the function.
• in libcxx/include/__string/extern_template_lists.h removes references to __grow_by and __grow_by_without_replace,
• clang-format, I hope this time I used the correct config!

This update fixes the issue with incorrect arguments during deque annotations. Now the patch should work without a problem after landing.

I have a hard time trying to create a stand alone test.
To go to the problematic if branch, push_back should be called when:

• __back_spare() == 0
• __front_spare() < __block_size
• __map_.size() < __map_.capacity()
• __map_.__back_spare() == 0

Should I change something?

It looks like the problem was created when I added __annotate_whole_block. Example buildbot output relevant for this revision: https://lab.llvm.org/buildbot/#/builders/5/builds/33480/steps/13/logs/stdio

==296484==ERROR: AddressSanitizer: bad parameters to __sanitizer_annotate_double_ended_contiguous_container:
      storage_beg        : 0x6210000d0d00
      storage_end        : 0x6210000d1d00
      old_container_beg  : 0x6210000d1b40
      old_container_end  : 0x6210000d1d00
      new_container_beg  : 0x6210000d1b40
      new_container_end  : 0x6210000d0d00
    #0 0x56178e14cac5 in __sanitizer_annotate_double_ended_contiguous_container /b/sanitizer-x86_64-linux-fast/build/llvm-project/compiler-rt/lib/asan/asan_poisoning.cpp:499:5
    #1 0x56178efd3474 in __annotate_double_ended_contiguous_container /b/sanitizer-x86_64-linux-fast/build/libcxx_build_asan/include/c++/v1/deque:893:13
    #2 0x56178efd3474 in __annotate_from_to /b/sanitizer-x86_64-linux-fast/build/libcxx_build_asan/include/c++/v1/deque:991:13
    #3 0x56178efd3474 in __annotate_whole_block /b/sanitizer-x86_64-linux-fast/build/libcxx_build_asan/include/c++/v1/deque:1037:9
    #4 0x56178efd3474 in std::__1::deque<std::__1::pair<std::__1::function<void ()>, llvm::ThreadPoolTaskGroup*>, std::__1::allocator<std::__1::pair<std::__1::function<void ()>, llvm::ThreadPoolTaskGroup*>>>::__add_back_capacity() /b/sanitizer-x86_64-linux-fast/build/libcxx_build_asan/include/c++/v1/deque:2233:13
    #5 0x56178efd1aaf in std::__1::pair<std::__1::function<void ()>, llvm::ThreadPoolTaskGroup*>& std::__1::deque<std::__1::pair<std::__1::function<void ()>, llvm::ThreadPoolTaskGroup*>, std::__1::allocator<std::__1::pair<std::__1::function<void ()>, llvm::ThreadPoolTaskGroup*>>>::emplace_back<std::__1::pair<std::__1::function<void ()>, llvm::ThreadPoolTaskGroup*>>(std::__1::pair<std::__1::function<void ()>, llvm::ThreadPoolTaskGroup*>&&) /b/sanitizer-x86_64-linux-fast/build/libcxx_build_asan/include/c++/v1/deque:1596:9
    [...]

This update fixes an issue with name shadowing and hopefully resolves all CI problems.

Rebase to HEAD of master.

This update the base commit.

This update solves the merge conflict with the newest commit.

This update addresses all nits.

• Adds __map_const_iterator.
• Removes incorrect comments.
• Addresses _mp sufix.