Remove C++ dependency from thread state

[ggeier]

Summary

• Convert the thread state implementation from C++ to C so libcoldtrace no longer pulls in libc++/libstdc++.
• Replace std::vector stack storage with malloc/free-backed stack storage while preserving stack diff serialization.
• Update the coldtrace source list to build thread.c instead of thread.cpp.

Verification

• cmake --build build -j
• ldd build/src/libcoldtrace.so shows only libc and the dynamic loader
• ctest --test-dir build --output-on-failure currently reports 32/45 passing with existing trace-checker failures

[db7]

LGTM. I think this is a good improvement. It removes one potential source of problems.

[ggeier]

There's just one problem left to solve: apparently, this uncovers a race between the first trace entry and file creation and the attribute constructor that calls the cleanup. At least the pipeline seems to show that:
https://github.com/open-s4c/coldtrace/actions/runs/25734106254/job/75566434211

coldtrace: copy procmaps
coldtrace: fopen '/home/runner/work/coldtrace/coldtrace/build/test/unmap__traces/freezer_log_1_0.bin (deleted)': No such file or directory
coldtrace: Failed to copy mapped files to '/home/runner/work/coldtrace/coldtrace/build/test/unmap__traces', coldtrace dump will be incomplete

For now, the pipeline just passes, because I disabled the cleanup. I assume it is a race, because I don't have this issue on my machine. However, it could be also deterministically first alloc/trace entry, constructor and cleanup, and then the remaining tracing.

[db7]

There's just one problem left to solve: apparently, this uncovers a race between the first trace entry and file creation and the attribute constructor that calls the cleanup. At least the pipeline seems to show that: https://github.com/open-s4c/coldtrace/actions/runs/25734106254/job/75566434211

coldtrace: copy procmaps
coldtrace: fopen '/home/runner/work/coldtrace/coldtrace/build/test/unmap__traces/freezer_log_1_0.bin (deleted)': No such file or directory
coldtrace: Failed to copy mapped files to '/home/runner/work/coldtrace/coldtrace/build/test/unmap__traces', coldtrace dump will be incomplete

For now, the pipeline just passes, because I disabled the cleanup. I assume it is a race, because I don't have this issue on my machine. However, it could be also deterministically first alloc/trace entry, constructor and cleanup, and then the remaining tracing.

I will try to reproduce on my side.

[db7]

There's just one problem left to solve: apparently, this uncovers a race between the first trace entry and file creation and the attribute constructor that calls the cleanup. At least the pipeline seems to show that: https://github.com/open-s4c/coldtrace/actions/runs/25734106254/job/75566434211

coldtrace: copy procmaps

coldtrace: fopen '/home/runner/work/coldtrace/coldtrace/build/test/unmap__traces/freezer_log_1_0.bin (deleted)': No such file or directory

coldtrace: Failed to copy mapped files to '/home/runner/work/coldtrace/coldtrace/build/test/unmap__traces', coldtrace dump will be incomplete

For now, the pipeline just passes, because I disabled the cleanup. I assume it is a race, because I don't have this issue on my machine. However, it could be also deterministically first alloc/trace entry, constructor and cleanup, and then the remaining tracing.

I will try to reproduce on my side.

I couldnt reproduce on my Ubuntu system

[ggeier]

I ran once the pipeline without changes to confirm it's not because github changed the runners #110. And I made the current change no longer be triggered by no longer calling thread_stack_push (#109) to make sure our code change is not triggering this. It seems that the change from having a libc++ dependency to no longer having it is actually the culprit.

[db7]

@ggeier - I am quite confused now. Why is this branching #108 passing now?

[ggeier]

070e188 'fixed' it: disabling the cleanup in the pipeline, but this can't be the solution

[db7]

070e188 'fixed' it: disabling the cleanup in the pipeline, but this can't be the solution

Ok, if that was the change than I really cannot reproduce it.