[RemoveDIs] Add additional debug-mode verifier checks

[OCHyams]

Separated from #83251

[llvmbot]

@llvm/pr-subscribers-llvm-ir

Author: Orlando Cazalet-Hyams (OCHyams)

Changes

Separated from #83251

---

Full diff: https://github.com/llvm/llvm-project/pull/84308.diff

1 Files Affected:

• (modified) llvm/lib/IR/Verifier.cpp (+10)

diff --git a/llvm/lib/IR/Verifier.cpp b/llvm/lib/IR/Verifier.cpp
index e0de179e57146f..e6e617cc40cb2d 100644
--- a/llvm/lib/IR/Verifier.cpp
+++ b/llvm/lib/IR/Verifier.cpp
@@ -2674,6 +2674,11 @@ void Verifier::visitFunction(const Function &F) {
   Check(verifyAttributeCount(Attrs, FT->getNumParams()),
         "Attribute after last parameter!", &F);
 
+  Check(F.IsNewDbgInfoFormat == F.getParent()->IsNewDbgInfoFormat,
+        "Fn debug format should match parent", &F,
+        F.IsNewDbgInfoFormat, F.getParent(),
+        F.getParent()->IsNewDbgInfoFormat);
+
   bool IsIntrinsic = F.isIntrinsic();
 
   // Check function attributes.
@@ -3017,6 +3022,11 @@ void Verifier::visitBasicBlock(BasicBlock &BB) {
     Check(I.getParent() == &BB, "Instruction has bogus parent pointer!");
   }
 
+  Check(BB.IsNewDbgInfoFormat == BB.getParent()->IsNewDbgInfoFormat,
+        "BB debug format should match parent", &BB,
+        BB.IsNewDbgInfoFormat, BB.getParent(),
+        BB.getParent()->IsNewDbgInfoFormat);
+
   // Confirm that no issues arise from the debug program.
   if (BB.IsNewDbgInfoFormat)
     CheckDI(!BB.getTrailingDPValues(), "Basic Block has trailing DbgRecords!",

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[SLTozer]

Should this be Check, or CheckDI? On the one hand this is a check that shouldn't fail due to a bad input, but only due to an error in LLVM itself; on the other hand, if it could be resolved by stripping debug info then it might be better to allow that in case someone encounters this in the wild? I'm not strongly opinionated either way, since there's precedent for internal debug info errors being a verifier failure.

[CarlosAlbertoEnciso]

May be instead of "Fn ..." use "F ..." or "Function ...".

[OCHyams]

This relanded in 2953d9c