Fix use-after-free in LoopUnrollPatterns.inc

[taalexander]

Description

In LoopUnrollPatterns.inc, after replacing a cc.continue op, the code would unconditionally check if (allowBreak) and attempt to cast term (the original terminator) to a cc.break op. However, replaceOpWithNewOp erases the original operation, leaving term as a dangling pointer. This caused failures in tests on MacOS that are otherwise not normally detectable on Linux, presumably due to MacOS's more aggressive allocator.

I wasn't able to prepare a regression test as a reproducer with scribbling like in #3748 on Linux in the devcontainer. However, I could reproduce with ASAN manually on Linux (pointing to the need for a CI job to automate this):

cd build
cmake .. \
  -DCMAKE_CXX_FLAGS="-fsanitize=address -fno-omit-frame-pointer" \
  -DCMAKE_C_FLAGS="-fsanitize=address -fno-omit-frame-pointer" \
  -DCMAKE_EXE_LINKER_FLAGS="-fsanitize=address" \
  -DCMAKE_SHARED_LINKER_FLAGS="-fsanitize=address"

ninja cudaq-opt

lit -v test/Transforms/unroll_loop_with_break.qke

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[lmondada]

Thanks!

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.