[PATCH] D131858: [clang] Track the templated entity in type substitution.

[Richard Smith - zygoloid via Phabricator via cfe-commits]

rsmith added a comment.

In D131858#3957630 <https://reviews.llvm.org/D131858#3957630>, @arphaman wrote:

> This change has caused a failure in Clang's stage 2 CI on the green dragon Darwin CI: https://green.lab.llvm.org/green/job/clang-stage2-Rthinlto/6390/console.
>
>   Assertion failed: (lvaluePath->getType() == elemTy && "Unexpected type reference!"), function readAPValue, file /Users/buildslave/jenkins/workspace/clang-stage1-RA/clang-build/tools/clang/include/clang/AST/AbstractBasicReader.inc, line 736.

This assert is simply wrong, and I've removed it in rG2009f2450532450a99c1a03d5e2c30f478121839 <https://reviews.llvm.org/rG2009f2450532450a99c1a03d5e2c30f478121839> -- that change should be safe to cherry-pick into the release branch. It's possible for the recomputation of the type after deserialization to result in a different type than what we saw when serializing, because redeclarations of the same entity can use the same type with different sugar -- or even slightly different types in some cases, such as when an array bound is added in a redeclaration. The dumps of the types provided by @steven_wu confirms that we were just seeing a difference in type sugar in this case.

>   Assertion failed: (BlockScope.empty() && CurAbbrevs.empty() && "Block imbalance"), function ~BitstreamWriter, file /Users/buildslave/jenkins/workspace/clang-stage1-RA/llvm-project/llvm/include/llvm/Bitstream/BitstreamWriter.h, line 119.

Is this still happening? If so, this looks more serious, and will need further investigation.

Can we undo the workaround in https://reviews.llvm.org/D139956 and see if the bot is now happy? Or can someone who was seeing problems before (@steven_wu?) run a test?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D131858/new/

https://reviews.llvm.org/D131858