AST,Sema: track and emit un-deserialized members #40994
Conversation
|
@swift-ci please test |
include/swift/AST/ASTContext.h
Outdated
| @@ -338,6 +338,9 @@ class ASTContext final { | |||
| llvm::SmallPtrSet<DerivativeAttr *, 1>> | |||
| DerivativeAttrs; | |||
|
|
|||
| llvm::DenseMap<ClassDecl *, llvm::SmallVector<DeclName, 2>> | |||
| MissingClassVTableEntries; | |||
There was a problem hiding this comment.
Can you store the DeclName in the MissingMemberDecl instead of adding another table to the ASTContext?
There was a problem hiding this comment.
While possible to iterate the members and pull out the MissingMemberDecl, it may limit any additional context from being saved off (due to Decls being bump ptr allocated). If you still think that it's better to do it that way, LMK and I can change this pretty easily.
There was a problem hiding this comment.
Can you explain what you mean by limiting context from being saved off?
What I'm suggesting is iterating over the members in TypeCheckDeclPrimary, yeah.
There was a problem hiding this comment.
Adding any additional context like the XRefTrace or the module which failed to load would not be possible AIUI. (Anything that is not trivially destructible is not embeddable).
There was a problem hiding this comment.
I think you can store that context in the MissingMemberDecl itself (it's fine for bump-allocated Decls to point to other things. You can either bump-allocate the additional context, or heap-allocate it and register a global destructor with the ASTContext on the MissingMemberDecl field in question).
|
Build failed |
|
Build failed |
This adds tracking of the vtable holes due to a failure to deserialize vtable entries. This allows for the user to be able to identify what member failed to be deserialied and can aid in understanding why an `open` class may not be subclassed. Future improvements here would allow tracing the XRefPath which failed to be deserialied. However, this still provides an improvement over the existing experience where there is no available information on why the class cannot be inherited from.
compnerd
force-pushed
the
undeserialized
branch
from
87ee5bd to
86de45a
Compare
|
@swift-ci please test |
|
Build failed |
|
@swift-ci please test macOS platform |
I wasn't particularly enthused by the use of |
| // RUN: %empty-directory(%t) | ||
| // RUN: %empty-directory(%t/swift) | ||
| // RUN: %target-swift-frontend -emit-module -emit-module-path %t/swift/SwiftLibrary.swiftmodule %s -module-name SwiftLibrary -I%S/Inputs | ||
| // RUN: %target-typecheck-verify-swift -DCLIENT -c %s -module-name client -I%t/swift |
There was a problem hiding this comment.
What does -c do? I couldn't find it in either options list.
There was a problem hiding this comment.
It's an alias for -emit-object: https://github.com/apple/swift/blob/ac875d00f04126d8a579e307d052df9752a18286/include/swift/Option/Options.td#L1015-L1016
There was a problem hiding this comment.
Thanks, should have looked at the modes too.
This adds tracking of the vtable holes due to a failure to deserialize
vtable entries. This allows for the user to be able to identify what
member failed to be deserialied and can aid in understanding why an
openclass may not be subclassed.Future improvements here would allow tracing the XRefPath which failed
to be deserialied. However, this still provides an improvement over the
existing experience where there is no available information on why the
class cannot be inherited from.
Replace this paragraph with a description of your changes and rationale. Provide links to external references/discussions if appropriate.
Resolves SR-NNNN.