On Fri, 23 Apr 2021, Jason Merrill wrote:
> On 4/22/21 9:46 AM, Patrick Palka wrote:
> > On Wed, 21 Apr 2021, Patrick Palka wrote:
> >
> > > On Wed, 21 Apr 2021, Jason Merrill wrote:
> > >
> > > > On 4/12/21 1:20 PM, Patrick Palka wrote:
> > > > > Here we're crashing during deduction for a template placeholder from a
> > > > > dependent initializer because one of the initializer's elements has an
> > > > > empty TREE_TYPE, something which resolve_args and later
> > > > > unify_one_argument
> > > > > don't expect. And if the deduction from a dependent initializer
> > > > > otherwise fails, we prematurely issue an error rather than
> > > > > reattempting
> > > > > the deduction at instantiation time.
> > > > >
> > > > > This patch makes do_class_deduction more tolerant about dependent
> > > > > initializers, in a manner similar to what do_auto_deduction does: if
> > > > > deduction from a dependent initializer fails, just return the original
> > > > > placeholder unchanged.
> > > >
> > > > Why doesn't the type_dependent_expression_p check in do_auto_deduction
> > > > catch
> > > > this already?
> > >
> > > That check applies only when context != adc_unify, but here we have
> > > context == adc_unify since we're being called from
> > > convert_template_argument.
> > >
> > > And currently, when 'auto' deduction fails for a dependent initializer,
> > > do_auto_deduction will just silently return the original placeholder:
> > >
> > > int val = type_unification_real (tparms, targs, parms, &init, 1, 0,
> > > DEDUCE_CALL,
> > > NULL, /*explain_p=*/false);
> > > if (val > 0)
> > > {
> > > if (processing_template_decl)
> > > /* Try again at instantiation time. */
> > > return type;
> > >
> > > so I suppose this patch just makes do_class_deduction behave more
> > > similarly to do_auto_deduction in this situation.
> >
> > On second thought, I think attempting CTAD a dependent initializer as the
> > patch
> > does might sometimes give us the wrong answer. If e.g. the class template
> > in
> > question has the deduction guides
> >
> > template <class T> A(T) -> A<char>;
> > A(int) -> A<void>;
> >
> > then ahead-of-time CTAD for A{v} where v is type-dependent will succeed and
> > resolve to A<char>, but at instantiation time the type of v might be int.
> > So
> > perhaps we should just have do_class_deduction punt on all type-dependent
> > expressions, e.g.
>
> OK.
Thanks a lot, patch committed as r12-100.
I wonder, would this be suitable for backporting to the 11 branch given
the amount of PRs this fixes (~16 if dups are counted)? The patch
should affect only C++20 code (since do_class_deduction should see a
dependent init only when context=adc_unify, so the added early exit
check should be dead code unless class NTTP placeholders are used), and
there doesn't seem to be a general workaround for the ICEs this fixes.
>
> > -- >8 --
> >
> > gcc/cp/ChangeLog:
> >
> > PR c++/89565
> > PR c++/93383
> > PR c++/99200
> > * pt.c (do_class_deduction): Give up if the initializer is
> > type-dependent.
> >
> > gcc/testsuite/ChangeLog:
> >
> > PR c++/89565
> > PR c++/93383
> > PR c++/99200
> > * g++.dg/cpp2a/nontype-class39.C: Remove dg-ice.
> > * g++.dg/cpp2a/nontype-class45.C: New test.
> > * g++.dg/cpp2a/nontype-class46.C: New test.
> > ---
> > gcc/cp/pt.c | 5 +++
> > gcc/testsuite/g++.dg/cpp2a/nontype-class39.C | 2 --
> > gcc/testsuite/g++.dg/cpp2a/nontype-class45.C | 32 ++++++++++++++++++++
> > gcc/testsuite/g++.dg/cpp2a/nontype-class46.C | 11 +++++++
> > 4 files changed, 48 insertions(+), 2 deletions(-)
> > create mode 100644 gcc/testsuite/g++.dg/cpp2a/nontype-class45.C
> > create mode 100644 gcc/testsuite/g++.dg/cpp2a/nontype-class46.C
> >
> > diff --git a/gcc/cp/pt.c b/gcc/cp/pt.c
> > index 7bcbe6dc3ce..6673f935ab6 100644
> > --- a/gcc/cp/pt.c
> > +++ b/gcc/cp/pt.c
> > @@ -29362,6 +29362,11 @@ do_class_deduction (tree ptype, tree tmpl, tree
> > init,
> > return error_mark_node;
> > }
> > + /* If the initializer is dependent, we can't resolve the class template
> > + placeholder ahead of time. */
> > + if (type_dependent_expression_p (init))
> > + return ptype;
> > +
> > tree type = TREE_TYPE (tmpl);
> > bool try_list_ctor = false;
> > diff --git a/gcc/testsuite/g++.dg/cpp2a/nontype-class39.C
> > b/gcc/testsuite/g++.dg/cpp2a/nontype-class39.C
> > index 512afad8e4f..9b4da4f02ea 100644
> > --- a/gcc/testsuite/g++.dg/cpp2a/nontype-class39.C
> > +++ b/gcc/testsuite/g++.dg/cpp2a/nontype-class39.C
> > @@ -1,7 +1,5 @@
> > // PR c++/89565
> > // { dg-do compile { target c++20 } }
> > -// { dg-additional-options "-fchecking" }
> > -// { dg-ice "resolve_args" }
> > template <auto>
> > struct N{};
> > diff --git a/gcc/testsuite/g++.dg/cpp2a/nontype-class45.C
> > b/gcc/testsuite/g++.dg/cpp2a/nontype-class45.C
> > new file mode 100644
> > index 00000000000..e7addf5f291
> > --- /dev/null
> > +++ b/gcc/testsuite/g++.dg/cpp2a/nontype-class45.C
> > @@ -0,0 +1,32 @@
> > +// PR c++/99200
> > +// { dg-do compile { target c++20 } }
> > +
> > +template <int N>
> > +struct A
> > +{
> > + constexpr A (const char (&s)[N]) { for (int i = 0; i < N; i++) v[i] =
> > s[i]; v[N] = 0; }
> > + char v[N + 1];
> > +};
> > +
> > +template <A s>
> > +struct B
> > +{
> > + constexpr operator const char *() { return s.v; }
> > +};
> > +
> > +template <typename T>
> > +const char *
> > +foo ()
> > +{
> > + return B<__PRETTY_FUNCTION__>{};
> > +}
> > +
> > +template <typename T>
> > +const char *
> > +bar ()
> > +{
> > + return B<__FUNCTION__>{};
> > +}
> > +
> > +auto a = foo <int> ();
> > +auto b = bar <double> ();
> > diff --git a/gcc/testsuite/g++.dg/cpp2a/nontype-class46.C
> > b/gcc/testsuite/g++.dg/cpp2a/nontype-class46.C
> > new file mode 100644
> > index 00000000000..d91e800424f
> > --- /dev/null
> > +++ b/gcc/testsuite/g++.dg/cpp2a/nontype-class46.C
> > @@ -0,0 +1,11 @@
> > +// PR c++/93383
> > +// { dg-do compile { target c++20 } }
> > +
> > +template <int> struct A {};
> > +
> > +template <A a> struct B {
> > + void foo(B<+a>);
> > + void bar(B<a.x>);
> > + template <class T> using type = B<T{}>;
> > + template <class> static inline auto y = A{0}; // { dg-error "deduction|no
> > match" }
> > +};
> >
>
>
