C++20 coroutine with stateful allocator results in ASAN stack-use-after-scope starting with clang 19

[bartdesmet]

When trying to upgrade to clang 19, we ran into an issue where an ASAN violation is reported in code using a stateful allocator on a promise type. I've tried to reduce the repro in https://godbolt.org/z/soPP6x9rx which shows 18 and 19 side-by-side. It looks like the stack-use-after-scope occurs during the construction of the coroutine frame.

==1==ERROR: AddressSanitizer: stack-use-after-scope on address 0x723e446091a0 at pc 0x5842f8f8642e bp 0x7fff8c00cc10 sp 0x7fff8c00cc08
READ of size 16 at 0x723e446091a0 thread T0
    #0 0x5842f8f8642d in callback_converter<my_coro_allocator<std::byte>> wrap<main::$_0, my_coro_allocator<std::byte>>(my_coro_allocator<std::byte>, main::$_0) /app/example.cpp:200:31
    #1 0x5842f8f85e04 in std::__n4861::coroutine_handle<void> callback_to_coroutine_handle<main::$_0, my_coro_allocator<std::byte>>(main::$_0, my_coro_allocator<std::byte> const&) /app/example.cpp:209:12
    #2 0x5842f8f85e04 in main /app/example.cpp:215:14
    #3 0x723e46429d8f  (/lib/x86_64-linux-gnu/libc.so.6+0x29d8f) (BuildId: 490fef8403240c91833978d494d39e537409b92e)
    #4 0x723e46429e3f in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x29e3f) (BuildId: 490fef8403240c91833978d494d39e537409b92e)
    #5 0x5842f8ea5374 in _start (/app/output.s+0x2c374)

Address 0x723e446091a0 is located in stack of thread T0 at offset 160 in frame
    #0 0x5842f8f85fbf in callback_converter<my_coro_allocator<std::byte>> wrap<main::$_0, my_coro_allocator<std::byte>>(my_coro_allocator<std::byte>, main::$_0) /app/example.cpp:201

  This frame has 3 object(s):
    [32, 64) 'agg.tmp.i.i' (line 78)
    [96, 128) 'agg.tmp.i' (line 157)
    [160, 192) 'agg.tmp1.i' (line 157) <== Memory access at offset 160 is inside this variable
HINT: this may be a false positive if your program uses some custom stack unwind mechanism, swapcontext or vfork
      (longjmp and C++ exceptions *are* supported)
SUMMARY: AddressSanitizer: stack-use-after-scope /app/example.cpp:200:31 in callback_converter<my_coro_allocator<std::byte>> wrap<main::$_0, my_coro_allocator<std::byte>>(my_coro_allocator<std::byte>, main::$_0)

[llvmbot]

@llvm/issue-subscribers-coroutines

Author: Bart J.F. De Smet (bartdesmet)

When trying to upgrade to clang 19, we ran into an issue where an ASAN violation is reported in code using a stateful allocator on a promise type. I've tried to reduce the repro in https://godbolt.org/z/soPP6x9rx which shows 18 and 19 side-by-side. It looks like the `stack-use-after-scope` occurs during the construction of the coroutine frame.

[NewSigma]

Cleaned reproducer: https://godbolt.org/z/4Y7r7dadr

[llvmbot]

@llvm/issue-subscribers-clang-codegen

Author: Bart J.F. De Smet (bartdesmet)

When trying to upgrade to clang 19, we ran into an issue where an ASAN violation is reported in code using a stateful allocator on a promise type. I've tried to reduce the repro in https://godbolt.org/z/soPP6x9rx which shows 18 and 19 side-by-side. It looks like the `stack-use-after-scope` occurs during the construction of the coroutine frame.

==1==ERROR: AddressSanitizer: stack-use-after-scope on address 0x723e446091a0 at pc 0x5842f8f8642e bp 0x7fff8c00cc10 sp 0x7fff8c00cc08
READ of size 16 at 0x723e446091a0 thread T0
    #<!-- -->0 0x5842f8f8642d in callback_converter&lt;my_coro_allocator&lt;std::byte&gt;&gt; wrap&lt;main::$_0, my_coro_allocator&lt;std::byte&gt;&gt;(my_coro_allocator&lt;std::byte&gt;, main::$_0) /app/example.cpp:200:31
    #<!-- -->1 0x5842f8f85e04 in std::__n4861::coroutine_handle&lt;void&gt; callback_to_coroutine_handle&lt;main::$_0, my_coro_allocator&lt;std::byte&gt;&gt;(main::$_0, my_coro_allocator&lt;std::byte&gt; const&amp;) /app/example.cpp:209:12
    #<!-- -->2 0x5842f8f85e04 in main /app/example.cpp:215:14
    #<!-- -->3 0x723e46429d8f  (/lib/x86_64-linux-gnu/libc.so.6+0x29d8f) (BuildId: 490fef8403240c91833978d494d39e537409b92e)
    #<!-- -->4 0x723e46429e3f in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x29e3f) (BuildId: 490fef8403240c91833978d494d39e537409b92e)
    #<!-- -->5 0x5842f8ea5374 in _start (/app/output.s+0x2c374)

Address 0x723e446091a0 is located in stack of thread T0 at offset 160 in frame
    #<!-- -->0 0x5842f8f85fbf in callback_converter&lt;my_coro_allocator&lt;std::byte&gt;&gt; wrap&lt;main::$_0, my_coro_allocator&lt;std::byte&gt;&gt;(my_coro_allocator&lt;std::byte&gt;, main::$_0) /app/example.cpp:201

  This frame has 3 object(s):
    [32, 64) 'agg.tmp.i.i' (line 78)
    [96, 128) 'agg.tmp.i' (line 157)
    [160, 192) 'agg.tmp1.i' (line 157) &lt;== Memory access at offset 160 is inside this variable
HINT: this may be a false positive if your program uses some custom stack unwind mechanism, swapcontext or vfork
      (longjmp and C++ exceptions *are* supported)
SUMMARY: AddressSanitizer: stack-use-after-scope /app/example.cpp:200:31 in callback_converter&lt;my_coro_allocator&lt;std::byte&gt;&gt; wrap&lt;main::$_0, my_coro_allocator&lt;std::byte&gt;&gt;(my_coro_allocator&lt;std::byte&gt;, main::$_0)

[bartdesmet]

@EugeneZelenko can this issue be assigned to someone familiar with coroutine compilation? This issue still reproduces in clang 20 and trunk, so we're holding off on a toolchain update on our end (but I may attempt some noinline or optnone workarounds to unblock us).

[EugeneZelenko]

@bartdesmet: You try to look on coroutines code changes history and ping developers here.

[bartdesmet]

@zmodem any thoughts on this issue?

[zmodem]

It's on my radar, but I haven't had time to look at it yet, sorry. @ChuanqiXu9 too.

[bartdesmet]

Gently ping to see if this could be looked at and get fixed for 21?