On Mon, Mar 01, 2021 at 03:08:58PM -0500, Jason Merrill wrote:
> On 3/1/21 2:54 PM, Marek Polacek wrote:
> > On Thu, Feb 25, 2021 at 10:45:29PM -0500, Jason Merrill via Gcc-patches
> > wrote:
> > > On 2/25/21 5:41 PM, Marek Polacek wrote:
> > > > On Thu, Feb 25, 2021 at 10:59:49AM -0500, Jason Merrill wrote:
> > > > > On 2/12/21 6:12 PM, Marek Polacek wrote:
> > > > > > We represent deduction guides with FUNCTION_DECLs, but they are
> > > > > > built
> > > > > > without DECL_CONTEXT
> > > > >
> > > > > Hmm, that seems wrong: "A deduction-guide shall be declared in the
> > > > > same scope as the corresponding class template and, for a member class
> > > > > template, with the same access." But it probably isn't necessary to
> > > > > change
> > > > > this:
> > > > >
> > > > > > leading to an ICE in type_dependent_expression_p
> > > > > > on the assert that the type of a function template with no dependent
> > > > > > (innermost!) template arguments must be non-dependent. Consider the
> > > > > > attached class-deduction79.C: we create a deduction guide:
> > > > > >
> > > > > > template<class T> G(T)-> E<Z>::G<T>
> > > > > >
> > > > > > we deduce T and create a partial instantiation:
> > > > > >
> > > > > > G(T) -> E<Z>::G<T> [with T = int]
> > > > > >
> > > > > > And then do_class_deduction wants to create a CALL_EXPR from the
> > > > > > above
> > > > > > using build_new_function_call -> build_over_call which calls
> > > > > > mark_used
> > > > > > -> maybe_instantiate_noexcept -> type_dependent_expression_p.
> > > > > >
> > > > > > There, the innermost template arguments are non-dependent (<int>),
> > > > > > but
> > > > > > the fntype is dependent -- the return type is a TYPENAME_TYPE, and
> > > > > > since we have no DECL_CONTEXT, this check holds:
> > > > > >
> > > > > > /* Otherwise, if the function decl isn't from a dependent
> > > > > > scope, it can't be
> > > > > > type-dependent. Checking this is important for functions
> > > > > > with auto return
> > > > > > type, which looks like a dependent type. */
> > > > > > if (TREE_CODE (expression) == FUNCTION_DECL
> > > > > > && !(DECL_CLASS_SCOPE_P (expression)
> > > > > > && dependent_type_p (DECL_CONTEXT (expression)))
> > > > > >
> > > > > > whereupon we ICE.
> > > > > >
> > > > > > Experiments with setting DECL_CONTEXT didn't pan out.
> > > > >
> > > > > In c8 of the PR it looks like you were using the class itself as
> > > > > DECL_CONTEXT; the quote above says that the right context is the
> > > > > enclosing
> > > > > scope of the class.
> > > >
> > > > Sadly, using CP_TYPE_CONTEXT (type) would result in a crash in
> > > > retrieve_specialization:
> > > >
> > > > /* There should be as many levels of arguments as there are
> > > > levels of parameters. */
> > > > gcc_assert (TMPL_ARGS_DEPTH (args)
> > > > == (TREE_CODE (tmpl) == TEMPLATE_DECL
> > > > ? TMPL_PARMS_DEPTH (DECL_TEMPLATE_PARMS (tmpl))
> > > > : template_class_depth (DECL_CONTEXT (tmpl))));
> > >
> > > Yeah, probably simpler not to bother.
> > >
> > > > > > So perhaps we
> > > > > > just want to skip the assert for deduction guides, because they are
> > > > > > a little special. Better ideas solicited.
> > > > >
> > > > > In c3 you mention that one of the variants broke with r269093; this is
> > > > > because my change to check CLASSTYPE_TEMPLATE_INSTANTIATION is false
> > > > > for the
> > > > > template pattern itself (E<Z>).
> > > >
> > > > And the original test started with my r11-1713 because using TREE_TYPE
> > > > directly instead of decltype (which is a non-deduced context) means we
> > > > can deduced from the argument.
> > > > > But I think probably the right answer is to defer this deduction
> > > > > until the
> > > > > enclosing scope is non-dependent, i.e. (untested)
> > > >
> > > > Thanks. That mostly works, except the new class-deduction-aggr[89].C
> > > > tests. Consider 8:
> > > >
> > > > namespace N {
> > > > template <typename, typename> struct S {
> > > > template <typename T, typename U> S(T, U);
> > > > };
> > > > } // namespace N
> > > > template <int> struct E {
> > > > template <typename T> struct G { T t; };
> > > > void fn() { G{N::S<char, int>{'a', 1}}; }
> > > > };
> > > >
> > > > void
> > > > g ()
> > > > {
> > > > E<1> e;
> > > > e.fn ();
> > > > }
> > > >
> > > > With your patch, when in do_class_deduction when
> > > > processing_template_decl,
> > > > we just return. When we call do_class_deduction again when p_t_d is 0,
> > > > maybe_aggr_guide returns early here:
> > > >
> > > > if (!CP_AGGREGATE_TYPE_P (type))
> > > > return NULL_TREE
> > > >
> > > > because G is not complete (and rightly so, we didn't instantiate it).
> > > > So
> > > > we aren't able to deduce the template parameters. I'm not sure if I
> > > > should
> > > > pursue this direction further. :(
> > >
> > > I think so; we just need to test CP_AGGREGATE_TYPE_P on the original
> > > template pattern instead of the instantiation E<1>::G.
> >
> > I'm sorry, I've got stuck again.
> >
> > Yes, using the original template pattern helps us get past the
> > CP_AGGREGATE_TYPE_P check.
> >
> > However, using TREE_TYPE (DECL_TI_TEMPLATE (tmpl)) as the type of the
> > deduction guide
> > means the guide will be "template<class T> G(T)-> E<<anonymous> >::G<T>"
> > which
> > results in
> >
> > class-deduction-aggr8.C:11:15: error: invalid use of dependent type
> > 'typename E<<anonymous> >::G<N::S<char, int> >'
> >
> > which makes sense I guess: when we defer building up the guide until
> > we've instantiated E<1>, finding the dependent type E<> is not expected.
>
> Yeah, I was only thinking to use the pattern for the aggregate check.
Ack. Though I think I also have to use the pattern here:
init = reshape_init (type, init, complain);
otherwise reshape_init returns a TARGET_EXPR and we immediately
crash in collect_ctor_idx_types because that only expects a CONSTRUCTOR.
And what we need to get is the type T -- of the constructor's index.
> > But creating the guide with "struct E<1>::G<T>" as its type seems
> > wrong; I'm not even sure if a guide like
> >
> > template<class T> G(T)-> E<1>::G<T>
> >
> > makes sense.
>
> It looks fine to me.
>
> > In any case the deduction fails (when we call
> > build_new_function_call in do_class_deduction), because we've got
> > a mismatch: the template parameter T has level 1, but the template
> > function parameter has level 2.
>
> Sure, because E<1>::G<T> has been partially instantiated, so the T has been
> reduced from level 2 to 1.
Right.
> You'll need to do a similar partial instantiation for building the deduction
> guide, either as part of the deduction guide rewriting or on the constructor
> before rewriting.
So I've tried. I can't actually tsubst the constructor itself, because it
at this point contains an AGGR_INIT_EXPR, which tsubst* can't handle. But
what I could do is
parms = tsubst (parms, DECL_TI_ARGS (tmpl), complain, init);
just after the call to collect_ctor_idx_types. After all, this is what
we care about and create the function template parameters from. So now
T's level is reduced to 1, and the guide we create is
template<class T> G(T)-> E<1>::G<T>
This guide is then chosen in do_class_deduction -> build_new_function_call,
but we crash in fn_type_unification -> instantiate_template (after we've
deduced T to N::S<char, int>) in retrieve_specialization:
gcc_assert (TMPL_ARGS_DEPTH (args)
== (TREE_CODE (tmpl) == TEMPLATE_DECL
? TMPL_PARMS_DEPTH (DECL_TEMPLATE_PARMS (tmpl))
: template_class_depth (DECL_CONTEXT (tmpl))));
args is <1, S> (depth 2), tmpl is our deduction guide, and
DECL_TEMPLATE_PARMS (tmpl) is [1 T] (depth 1).
:/
> It might also work to do something tricky like giving the E template
> parameter a default argument of 1, but that seems like it would need more
> invention.
I haven't tried this, but I'm not sure if it'd avoid the problem above.
Marek
