Incorrect handling of incomplete types in std::derived_from
#121278
Labels
Comments
cor3ntin
added
libc++
|
Note that https://cplusplus.github.io/LWG/lwg-active.html#3929 and https://cplusplus.github.io/LWG/lwg-active.html#2939 are related The fact that LWG suggest making it a Mandates (rather than a constraint) makes me question whether GCC is correct and this should always be a hard error (in which case it might be a clang bug) |
|
IMO we should reject such code, since it can easily result in really hard-to-find bugs. Given that the LWG issues you linked seem to agree with me here, I'd rather have Clang reject both implementations of |
cor3ntin
added a commit
to cor3ntin/llvm-project
that referenced
this issue
Dec 30, 2024
…inae-friendly error LWG3929 suggests that passing incomplete types to __is_base_of and other builtins supporting [meta.unary] should result in a non-sfinaeable error. This is consistent with GCC's behavior and avoid inconsistency when using a builtin instead of a standard trait in a concept-definition. Fixes llvm#121278
cor3ntin
added a commit
to cor3ntin/llvm-project
that referenced
this issue
Dec 30, 2024
…inae-friendly error LWG3929 suggests that passing incomplete types to __is_base_of and other builtins supporting [meta.unary] should result in a non-sfinaeable error. This is consistent with GCC's behavior and avoid inconsistency when using a builtin instead of a standard trait in a concept-definition. Fixes llvm#121278
cor3ntin
added a commit
to cor3ntin/llvm-project
that referenced
this issue
Dec 30, 2024
…inae-friendly error LWG3929 suggests that passing incomplete types to __is_base_of and other builtins supporting [meta.unary] should result in a non-sfinaeable error. This is consistent with GCC's behavior and avoid inconsistency when using a builtin instead of a standard trait in a concept-definition. Fixes llvm#121278
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Per https://eel.is/c++draft/meta#lib:is_base_of,
This is generally enforced by
__is_base_of(the builtin used to implementis_base_of_v)However, in libc++,
derived_fromis defined asllvm-project/libcxx/include/__concepts/derived_from.h
Line 27 in cb1ad98
And because the constraints are only enforced in
__is_base_ofwhich is not in the immediate context of theis_base_of_vconstraint, substituting an incomplete class inis_base_of_vdoes not result in a substitution failure in the immediate context (which should just causederived_fromto not be satisfied per https://eel.is/c++draft/temp.constr.atomic#3.sentence-2) but instead makes the program ill-formed, which i believe is non-conforming.Note that GCC diverges in all cases. MSVC and EDG exposes what I believe to be the correct behavior.
I think there are 2 solutions:
derived_fromto use__is_base_ofdirectly (which is what libstdc++ and MS STL do)std::is_base_of_vto check for completenesshttps://godbolt.org/z/zGrKEKEao
The text was updated successfully, but these errors were encountered: