Do not report inlining decision to runtime in DEBUG mode

[kunalspathak]

We should report inlining decisions to the runtime only in non-DEBUG mode. If done in DEBUG mode, we can see discrepancies in inlining decisions that can further diverge the code produced in RELEASE vs. DEBUG mode. The inlining reporting happened at below code. The fix is to not report the inlining decision.

runtime/src/coreclr/jit/fginline.cpp

Lines 862 to 866 in 5e7d681

	#if defined(DEBUG)
	// In debug builds we want the inline tree to show all failed
	// inlines.
	fgWalkTreePre(stmt->GetRootNodePointer(), fgFindNonInlineCandidate, stmt);
	#endif

Fixes: #113546

[Copilot]

Copilot wasn't able to review any files in this pull request.

Files not reviewed (1)

• src/coreclr/jit/fginline.cpp: Language not supported

[AndyAyersMS]

Note this code is trying to reconstruct (after the fact) why we didn't inline something. The initial inline attempt happened (and failed) and (if necessary) was reported before we even built the set of inline candidates. So there is no need to report it again here.

Now, reporting failure a second time normally would be harmless and not lead to codegen diffs.

But what is happening here is that the debug-only reconstruction comes up with a different explanation for why the inline failed, and that explanation is one that precludes ever inlining the method, and so if we end up reporting on the reconstructed failure it triggers a divergence in debug and release.

As for why we even bother reconstruction -- the idea was that in debug the inline tree should reflect all the inline decisions, both successes and failures, and since some inlines "fail early" we need to recover why they failed later when building the tree, and evidently this reconstruction process is a bit shaky.

[kunalspathak]

/azp run runtime-coreclr superpmi-asmdiffs-checked-release

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[kunalspathak]

/ba-g failure is unrelated, one of them is #112264