[clang] [serialization] no transitive decl change (PR #92083)

[Ilya Biryukov via cfe-commits]

ilya-biryukov wrote:

To clarify: we observe this regression on x86, not on Arm. @alexfh is now trying to profile the compiler to locate what's going wrong and I will try to poke at PCM files and come up with a reproducer.

However, the regressions is definitely real and will block our internal releases, preventing testing of further commits that go in (and these things may pile up). Local revert could be an option, but I'm worried that follow up changes will have conflicts (changing DeclID to 64bit seems like it should, at least potentially, have wide-reaching implications).

It is also hard to predict how long it would take to get to the bottom of it, we will still prioritize finding the problem and the reproducer if this commit is reverted, but we won't have to worry about the conflicts and such.

I do fully acknowledge that it's frustrating when problems are not found during the review or shortly after commits land, and that not having this upstream may make follow up work more difficult (because of the need to have those commits on a branch and rebase/merge more).

We try hard to be prompt with out testing, but sometimes things aren't as fast as we'd like.

https://github.com/llvm/llvm-project/pull/92083