[darkdi]

allocate_bytes(SIZE_MAX, 16) with an adapted allocator wraps the bookkeeping size, then asserts in Debug or segfaults in Release. Reject the overflow with std::bad_alloc before calling the allocator. Boundary regressions, 1,062 CPU CTest cases, and clang-format 21 pass on macOS arm64, plus ASan/UBSan checks and a no-exceptions smoke test.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[ericniebler]

/ok to test 3f658c2

[ispeters]

@darkdi, I'm curious how you found this bug. You're fixing a problem in the reference implementation for P4223R1 so I'd be interested in any user experiences you have with exec::function.

[darkdi]

Found during a Codex-assisted audit of allocation-size arithmetic. I checked allocate_bytes(SIZE_MAX, 16) with a custom allocator: the bookkeeping size wrapped, followed by an assertion in Debug and a segfault in Release. I haven't used exec::function in an application, so I don't have user experience to report there.

[darkdi]

I also checked the red sanitizer jobs. The same four configurations fail on merged #2359: clang 16 ASAN segfaults while listing test.scratch tests, and gcc 14 TSAN reports an incompatible memory layout/ASLR error during test discovery. This PR's logs show the same failures. Both GCC 14 ASAN configurations and the GPU matrix pass here.