https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122383
Bug ID: 122383
Summary: auto deduction fails when initializing from a template
conversion operator address
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Keywords: rejects-valid
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: attackerj1113 at gmail dot com
Target Milestone: ---
Consider the following code:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
template<class T> struct X {
template<class U> operator X<U>();
};
struct A {};
struct B {};
void test() {
auto (X<A>::*pf)() = &X<A>::operator X<B>;
}
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
GCC rejects this, but other major compilers (Clang, MSVC, EDG) accept the code
as valid.
GCC output:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
<source>: In function 'void test()':
<source>:9:26: error: unable to deduce 'auto (X<A>::*)()' from '&
X<A>::operator X<U>'
9 | auto (X<A>::*pf)() = &X<A>::operator X<B>;
<source>:9:42: note: couldn't deduce template parameter 'auto'
9 | auto (X<A>::*pf)() = &X<A>::operator X<B>;
| ^~~~
Compiler returned: 1
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
See godbolt: https://godbolt.org/z/nsc9394Kh
I believe auto can be easily deduced here as X<B> (X<A>::*)().
If written as:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
auto pf = &X<A>::operator X<B>;
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In that case, 'auto' should likewise deduce to 'X<B> (X<A>::*)()'.
It seems that GCC perhaps attempts template parameter deduction inside the
member conversion operator template, rather than deducing the 'auto' type
directly from the initializer’s full type.
