type_of metafunction should dealias the type
#93
Conversation
|
Hi @katzdm , can you help me take a look when you have time? |
changkhothuychung
changed the title
type_of metafunction should dealias the type
clang/lib/Sema/Metafunctions.cpp
Outdated
| @@ -2007,7 +2007,15 @@ bool type_of(APValue &Result, Sema &S, EvalFn Evaluator, DiagFn Diagnoser, | |||
|
|
|||
| switch (RV.getReflectionKind()) { | |||
| case ReflectionKind::Null: | |||
| case ReflectionKind::Type: | |||
There was a problem hiding this comment.
I probably should have pointed to this earlier, but this is the spec for type_of:
https://brevzin.github.io/cpp_proposals/2996_reflection/d2996r6.html#pnum_210
type_of(^^int) fails to be a constant expression, so the existing behavior (i.e., diagnosing an error) is correct.
The cases that need modification are Object, Value, Declaration, BaseSpecifier, and DataMemberSpec.
(also for return_type_of below)
Ping me tomorrow if there's anything I can help with!
clang/lib/Sema/Metafunctions.cpp
Outdated
| bool UnwrapAlises = isa<UsingType>(currentType) || | ||
| isa<TypedefType>(currentType) || | ||
| isa<TemplateSpecializationType>(currentType); |
There was a problem hiding this comment.
Out of curiosity, what happens if we just pass /*UnwrapAliases=*/true unconditionally? Are we avoiding an undesirable behavior with this check, or is this an optimization? If the later, I'd prefer passing unconditional true.
(also for return_type_of below)
There was a problem hiding this comment.
I updated to unconditional true and it looks fine to me. I updated some tests in libcxx to reflect the new behavior.
|
I have this test fails locally - The reason is probably due to the arch of my laptop (mac M1). From the log - Can you help me run that test with my change to see if that runs ok? |
@changkhothuychung I believe I fixed this yesterday evening, while merging latest changes from upstream. Let me know if pulling from latest fixes the issue for you. |
|
@katzdm I pulled the latest from p2996, but it doesnt seem to fix the error above. |
Fix #91
Issue number of the reported bug or feature request: #
Describe your changes
Adding code to dealias the underlying type in case of using typedef-name
Testing performed
Updating current tests in libcxx to reflect the new expected behaviors.
Additional context