Merge pull request #14010 from mshinwell/string-ops-machtypes

[dijkstracula]

Summary

Backport ocaml/ocaml#14010
to 5.4.0-semgrep.

The fix preserves address-typed Cmm expressions when computing addresses for
string/bytes operations and unaligned loads and stores. Without this distinction,
Flambda may fail to preserve the liveness relationship between an interior
address and its underlying allocation, resulting in GC-related memory
corruption.

This was encountered as nondeterministic corruption in an allocation-heavy
property test.

Backport Notes

The compiler change applies cleanly to OCaml 5.4. No source adaptation was
required: the resulting asmcomp/cmm_helpers.ml patch matches upstream ocaml#14010
exactly.

The conflicting upstream Changes entry was omitted because it records
inclusion in an upstream OCaml release, whereas this is a Semgrep-fork
backport.

No other fork code is changed.

Validation

• Clean Flambda bootstrap of the exact branch:
  ./configure --enable-flambda && make -j8
• Confirmed the resulting compiler reports:
  • version: 5.4.0+semgrep-fork@23106c6472
  • flambda: true
• Compiled the current Semgrep preprocessor evaluator with the resulting
  compiler at -O3.
• Ran four seeded allocation-heavy evaluator checks at 5,000,000 iterations
  each: 20,000,000 total cases passed.

The only changed source file is /Users/ntaylor/code/ocaml/asmcomp/cmm_helpers.ml:154.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}