Cherry-pick "[Modules] Record whether VarDecl initializers contain side effects (#143739)" #10925
Conversation
…lvm#143739) Calling `DeclMustBeEmitted` should not lead to more deserialization, as it may occur before previous deserialization has finished. When passed a `VarDecl` with an initializer however, `DeclMustBeEmitted` needs to know whether that initializer contains side effects. When the `VarDecl` is deserialized but the initializer is not, this triggers deserialization of the initializer. To avoid this we add a bit to the serialization format for `VarDecl`s, indicating whether its initializer contains side effects or not, so that the `ASTReader` can query this information directly without deserializing the initializer. rdar://153085264 This is a cherry-pick of 319a51a
Normally expressions passed to EvaluateInPlace have a type, but not when a VarDecl initializer is evaluated before the untyped ParenListExpr is replaced with a CXXConstructExpr. This can happen in LLDB. In these cases consteval fails.
|
@swift-ci please test |
|
@swift-ci please test llvm |
| @@ -2529,12 +2531,12 @@ void ASTWriter::WriteDeclAbbrevs() { | |||
| // VarDecl | |||
| Abv->Add(BitCodeAbbrevOp( | |||
| BitCodeAbbrevOp::Fixed, | |||
| 21)); // Packed Var Decl bits: Linkage, ModulesCodegen, | |||
| 22)); // Packed Var Decl bits: Linkage, ModulesCodegen, | |||
There was a problem hiding this comment.
I'd expect this kind of change to also involve an update to VERSION_MAJOR, but it seems that doesn't happen much for Clang serialization changes? If it's not needed, that's fine.
There was a problem hiding this comment.
I don't know if we've ever written down a policy for this, and we've been inconsistent about it in the past. In general, we don't try to read a pch/module if getClangFullRepositoryVersion doesn't match, and we check that in the same record that we check VERSION_MAJOR. @Bigcheese or @jansvoboda11 any opinion?
There was a problem hiding this comment.
We've always treated Clang as needing an exact commit hash match to consider modules as valid. This probably should have updated VERSION_MAJOR, but also I don't think it's a problem for us if it doesn't.
Calling
DeclMustBeEmittedshould not lead to more deserialization, asit may occur before previous deserialization has finished.
When passed a
VarDeclwith an initializer however,DeclMustBeEmittedneeds to know whether that initializer contains side effects. When the
VarDeclis deserialized but the initializer is not, this triggersdeserialization of the initializer. To avoid this we add a bit to the
serialization format for
VarDecls, indicating whether its initializercontains side effects or not, so that the
ASTReadercan query thisinformation directly without deserializing the initializer.
rdar://153085264
This is a cherry-pick of 319a51a
This unblocks swiftlang/swift#81859