rsmith added inline comments.
================
Comment at: clang/include/clang/Basic/DiagnosticSemaKinds.td:4683-4686
+ "Member '%0' marked with 'exclude_from_explicit_instantiation' attribute is "
+ "not defined but an explicit template instantiation declaration exists. "
+ "Reliance on this member being defined by an explicit template instantiation
"
+ "will lead to link errors.">;
----------------
ldionne wrote:
> rsmith wrote:
> > Diagnostics should start with a lowercase letter and not end with a period.
> >
> > That said, I'm not sure I see why this diagnostic is correct / useful. If
> > the entity is never used, then there's no link error. And if it is ever
> > used, then you should get an implicit instantiation like normal, and we
> > already have a diagnostic for the case where an entity is implicitly
> > instantiated and no definition is available.
> > Diagnostics should start with a lowercase letter and not end with a period.
>
> Done.
>
> > That said, I'm not sure I see why this diagnostic is correct / useful. If
> > the entity is never used, then there's no link error. And if it is ever
> > used, then you should get an implicit instantiation like normal, and we
> > already have a diagnostic for the case where an entity is implicitly
> > instantiated and no definition is available.
>
> This is not what happens right now. If you don't provide a definition but you
> try to call the function, an extern call will be produced (and that will
> result in a link error because any potential explicit instantiation won't
> provide the function). For example:
>
> ```
> cat <<EOF | ./install/bin/clang++ -cc1 -stdlib=libc++ -xc++ -emit-llvm -o - -
> template <class T>
> struct Foo {
> __attribute__((exclude_from_explicit_instantiation)) static void
> static_member_function();
> };
>
> extern template struct Foo<int>;
>
> int main() {
> Foo<int>::static_member_function();
> }
> EOF
> ```
>
> Results in the following LLVM IR:
>
> ```
> ; Function Attrs: noinline norecurse nounwind optnone
> define i32 @main() #0 {
> entry:
> call void @_ZN3FooIiE22static_member_functionEv()
> ret i32 0
> }
>
> declare void @_ZN3FooIiE22static_member_functionEv() #1
> ```
>
> I guess we should be getting a warning or an error on the point of implicit
> instantiation instead, or is this behavior acceptable?
>
I don't think your example is fundamentally any different from:
```
template <class T>
struct Foo {
static void static_member_function();
};
int main() {
Foo<int>::static_member_function();
}
```
which likewise produces a declaration of
`_ZN3FooIiE22static_member_functionEv`. That case produces a
`-Wundefined-func-template` warning; your example should do the same.
Repository:
rC Clang
https://reviews.llvm.org/D51789
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
