Re: [PATCH] c++: ICE with -Wmismatched-tags and member template [PR106259]

[Marek Polacek <polacek@redhat.com>]

On Wed, Mar 01, 2023 at 04:44:12PM -0500, Jason Merrill wrote:
> On 3/1/23 16:40, Marek Polacek wrote:
> > On Wed, Mar 01, 2023 at 04:30:16PM -0500, Jason Merrill wrote:
> > > On 3/1/23 15:33, Marek Polacek wrote:
> > > > -Wmismatched-tags warns about the (harmless) struct/class mismatch.
> > > > For, e.g.,
> > > > 
> > > >     template<typename T> struct A { };
> > > >     class A<int> a;
> > > > 
> > > > it works by adding A<T> to the class2loc hash table while parsing the
> > > > class-head and then, while parsing the elaborate type-specifier, we
> > > > add A<int>.  At the end of c_parse_file we go through the table and
> > > > warn about the class-key mismatches.  In this PR we crash though; we
> > > > have
> > > > 
> > > >     template<typename T> struct A {
> > > >       template<typename U> struct W { };
> > > >     };
> > > >     struct A<int>::W<int> w; // #1
> > > > 
> > > > where while parsing A and #1 we've stashed
> > > >      A<T>
> > > >      A<T>::W<U>
> > > >      A<int>::W<int>
> > > > into class2loc.  Then in class_decl_loc_t::diag_mismatched_tags TYPE
> > > > is A<int>::W<int>, and specialization_of gets us A<int>::W<U>, which
> > > > is not in class2loc, so we crash on gcc_assert (cdlguide).  But it's
> > > > OK not to have found A<int>::W<U>, we should just look one "level" up,
> > > > that is, A<T>::W<U>.
> > > > 
> > > > It's important to handle class specializations, so e.g.
> > > > 
> > > >     template<>
> > > >     struct A<char> {
> > > >       template<typename U>
> > > >       class W { };
> > > >     };
> > > > 
> > > > where W's class-key is different than in the primary template above,
> > > > so we should warn depending on whether we're looking into A<char>
> > > > or into a different instantiation.
> > > > 
> > > > Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk?
> > > > 
> > > > 	PR c++/106259
> > > > 
> > > > gcc/cp/ChangeLog:
> > > > 
> > > > 	* parser.cc (class_decl_loc_t::diag_mismatched_tags): If the first
> > > > 	lookup of SPEC didn't find anything, try to look for
> > > > 	most_general_template.
> > > > 
> > > > gcc/testsuite/ChangeLog:
> > > > 
> > > > 	* g++.dg/warn/Wmismatched-tags-11.C: New test.
> > > > ---
> > > >    gcc/cp/parser.cc                              | 30 +++++++++++++++----
> > > >    .../g++.dg/warn/Wmismatched-tags-11.C         | 23 ++++++++++++++
> > > >    2 files changed, 47 insertions(+), 6 deletions(-)
> > > >    create mode 100644 gcc/testsuite/g++.dg/warn/Wmismatched-tags-11.C
> > > > 
> > > > diff --git a/gcc/cp/parser.cc b/gcc/cp/parser.cc
> > > > index 1a124f5395e..b528ee7b1d9 100644
> > > > --- a/gcc/cp/parser.cc
> > > > +++ b/gcc/cp/parser.cc
> > > > @@ -34473,14 +34473,32 @@ class_decl_loc_t::diag_mismatched_tags (tree type_decl)
> > > >    	 be (and inevitably is) at index zero.  */
> > > >          tree spec = specialization_of (type);
> > > >          cdlguide = class2loc.get (spec);
> > > > +      /* It's possible that we didn't find SPEC.  Consider:
> > > > +
> > > > +	   template<typename T> struct A {
> > > > +	     template<typename U> struct W { };
> > > > +	   };
> > > > +	   struct A<int>::W<int> w; // #1
> > > > +
> > > > +	 where while parsing A and #1 we've stashed
> > > > +	   A<T>
> > > > +	   A<T>::W<U>
> > > > +	   A<int>::W<int>
> > > > +	 into CLASS2LOC.  If TYPE is A<int>::W<int>, specialization_of
> > > > +	 will yield A<int>::W<U> which may be in CLASS2LOC if we had
> > > > +	 an A<int> class specialization, but otherwise won't be in it.
> > > > +	 So try to look up A<T>::W<U>.  */
> > > > +      if (!cdlguide)
> > > > +	{
> > > > +	  spec = DECL_TEMPLATE_RESULT (most_general_template (spec));
> > > 
> > > Would it make sense to only look at most_general_template, not A<int>::W<U>
> > > at all?
> > 
> > I think that would break with class specialization, as in...
> > 
> > > > +template<typename T> struct A {
> > > > +  template<typename U>
> > > > +  struct W { };
> > > > +};
> > > > +
> > > > +template<>
> > > > +struct A<char> {
> > > > +  template<typename U>
> > > > +  class W { };
> > > > +};
> > > > +
> > > > +void
> > > > +g ()
> > > > +{
> > > > +  struct A<char>::W<int> w1; // { dg-warning "mismatched" }
> > 
> > ...this, where we should first look into A<char>, and only if not
> > found, go to A<T>.
> 
> I'd expect the
> 
> >       /* Stop if we run into an explicitly specialized class template.  */
> 
> code in most_general_template to avoid that problem.

Ah, I had no idea it does that.  The unconditional most_general_template
works fine for the new test, but some of the existing tests then fail.
Reduced:

template <class Z>   struct S2; // #1
template <class T> class S2<const T>; // #2

extern class  S2<const int> s2ci; // #3
extern struct S2<const int> s2ci;     // { dg-warning "\\\[-Wmismatched-tags" }

where the unconditional most_general_template changes spec from
"class S2<const T>" to "struct S2<Z>" (both of which are in class2loc).
So it regresses the diagnostic, complaining that #3 should have "struct"
since #1 has "struct".  I think we want to keep the current diagnostic,
saying that the last line should have "class" since the specialization
in line #2 has "class".