MaskRay wrote: > > For reference, can you give a couple examples of code where this is > > triggering? > > If this is triggering in practice, do we want a real driver option to > > control the sanitizer? The alignment attributes themselves?
> I am not sure we need special driver flag for that. > > @MaskRay After reading about amount of unique cases you see, maybe ignore > list is easier? Given the scale of our internal code base, the identified unique failures are moderate. So eventually we can remove the `cl::opt` option, but having the option buys us time to clean up the code base. Many problems identified look like this: https://chromium.googlesource.com/v8/v8.git/+/bb2ac8991c64601be9852d88c22abac2d6a6c39b/src/bigint/bigint.h#120 ```cpp protected: friend class ShiftedDigits; digit_t* digits_; // digits_ may be misaligned int len_; private: // We require externally-provided digits arrays to be 4-byte aligned, but // not necessarily 8-byte aligned; so on 64-bit platforms we use memcpy // to allow unaligned reads. digit_t read_4byte_aligned(int i) { if (sizeof(digit_t) == 4) { return digits_[i]; } else { digit_t result; memcpy(&result, digits_ + i, sizeof(result)); // unspecified behavior identified here return result; } } }; ``` https://github.com/llvm/llvm-project/pull/69240 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
