diff mbox series

c++: alias_ctad_tweaks and constrained dguide [PR95486]

Message ID 20200724134955.3681531-1-ppalka@redhat.com
State New
Headers show
Series c++: alias_ctad_tweaks and constrained dguide [PR95486] | expand

Commit Message

Patrick Palka July 24, 2020, 1:49 p.m. UTC
In the below testcase, we're ICEing from alias_ctad_tweaks ultimately
because the implied deduction guide for X's user-defined constructor
already has constraints associated with it.  We then carry over these
constraints to 'fprime', the overlying deduction guide for the alias
template Y, via tsubst_decl from alias_ctad_tweaks.  Later in
alias_ctad_tweaks we call get_constraints followed by set_constraints on
this overlying deduction guide without doing remove_constraints in
between, which triggers the !found assert in set_constraints.

This patch fixes this issue by adding an intervening call to
remove_constraints.  And for consistency, it changes the call to
get_constraints to go through fprime instead of f.  Since the
DECL_TEMPLATE_RESULT of fprime is f, this consistency change makes
no functional difference.

Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK
commit?

gcc/cp/ChangeLog:

	PR c++/95486
	* pt.c (alias_ctad_tweaks): Obtain the associated constraints
	through fprime instead of through f.  Call remove_constraints
	before calling set_constraints.

gcc/testsuite/ChangeLog:

	PR c++/95486
	* g++.dg/cpp2a/class-deduction-alias3.C: New test.
---
 gcc/cp/pt.c                                         |  7 +++++--
 gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C | 11 +++++++++++
 2 files changed, 16 insertions(+), 2 deletions(-)
 create mode 100644 gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C

Comments

Jason Merrill July 29, 2020, 7:54 p.m. UTC | #1
On 7/24/20 9:49 AM, Patrick Palka wrote:
> In the below testcase, we're ICEing from alias_ctad_tweaks ultimately
> because the implied deduction guide for X's user-defined constructor
> already has constraints associated with it.  We then carry over these
> constraints to 'fprime', the overlying deduction guide for the alias
> template Y, via tsubst_decl from alias_ctad_tweaks.  Later in
> alias_ctad_tweaks we call get_constraints followed by set_constraints on
> this overlying deduction guide without doing remove_constraints in
> between, which triggers the !found assert in set_constraints.
> 
> This patch fixes this issue by adding an intervening call to
> remove_constraints.

This hunk is OK.

> And for consistency, it changes the call to
> get_constraints to go through fprime instead of f.

This hunk is not, as what we're doing is substituting the constraints of 
f to produce the constraints of f'.  This change makes it less clear.

OK without that change.

> Since the DECL_TEMPLATE_RESULT of fprime is f, this consistency change makes
> no functional difference.

The DECL_TEMPLATE_RESULT of fprime is g, not f.

> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK
> commit?
> 
> gcc/cp/ChangeLog:
> 
> 	PR c++/95486
> 	* pt.c (alias_ctad_tweaks): Obtain the associated constraints
> 	through fprime instead of through f.  Call remove_constraints
> 	before calling set_constraints.
> 
> gcc/testsuite/ChangeLog:
> 
> 	PR c++/95486
> 	* g++.dg/cpp2a/class-deduction-alias3.C: New test.
> ---
>   gcc/cp/pt.c                                         |  7 +++++--
>   gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C | 11 +++++++++++
>   2 files changed, 16 insertions(+), 2 deletions(-)
>   create mode 100644 gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C
> 
> diff --git a/gcc/cp/pt.c b/gcc/cp/pt.c
> index b6df87a8a2e..4d955c555dc 100644
> --- a/gcc/cp/pt.c
> +++ b/gcc/cp/pt.c
> @@ -28598,7 +28598,7 @@ alias_ctad_tweaks (tree tmpl, tree uguides)
>   	  DECL_PRIMARY_TEMPLATE (fprime) = fprime;
>   
>   	  /* Substitute the associated constraints.  */
> -	  tree ci = get_constraints (f);
> +	  tree ci = get_constraints (fprime);
>   	  if (ci)
>   	    ci = tsubst_constraint_info (ci, targs, complain, in_decl);
>   	  if (ci == error_mark_node)
> @@ -28616,7 +28616,10 @@ alias_ctad_tweaks (tree tmpl, tree uguides)
>   	    }
>   
>   	  if (ci)
> -	    set_constraints (fprime, ci);
> +	    {
> +	      remove_constraints (fprime);
> +	      set_constraints (fprime, ci);
> +	    }
>   	}
>         else
>   	{
> diff --git a/gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C b/gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C
> new file mode 100644
> index 00000000000..318d4c942ce
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C
> @@ -0,0 +1,11 @@
> +// PR c++/95486
> +// { dg-do compile { target c++20 } }
> +
> +template<class T, class U>
> +struct X { X(U) requires __is_same(U, int) {} };
> +
> +template<class U>
> +using Y = X<void, U>;
> +
> +Y y{1};
> +Y z{'a'}; // { dg-error "failed|no match" }
>
Patrick Palka July 29, 2020, 9:02 p.m. UTC | #2
On Wed, 29 Jul 2020, Jason Merrill wrote:

> On 7/24/20 9:49 AM, Patrick Palka wrote:
> > In the below testcase, we're ICEing from alias_ctad_tweaks ultimately
> > because the implied deduction guide for X's user-defined constructor
> > already has constraints associated with it.  We then carry over these
> > constraints to 'fprime', the overlying deduction guide for the alias
> > template Y, via tsubst_decl from alias_ctad_tweaks.  Later in
> > alias_ctad_tweaks we call get_constraints followed by set_constraints on
> > this overlying deduction guide without doing remove_constraints in
> > between, which triggers the !found assert in set_constraints.
> > 
> > This patch fixes this issue by adding an intervening call to
> > remove_constraints.
> 
> This hunk is OK.
> 
> > And for consistency, it changes the call to
> > get_constraints to go through fprime instead of f.
> 
> This hunk is not, as what we're doing is substituting the constraints of f to
> produce the constraints of f'.  This change makes it less clear.
> 
> OK without that change.

That makes sense, consider that hunk gone then.

> 
> > Since the DECL_TEMPLATE_RESULT of fprime is f, this consistency change makes
> > no functional difference.
> 
> The DECL_TEMPLATE_RESULT of fprime is g, not f.

Ah yes, sorry for this thinko.  I'll commit just the one hunk after a
round of testing; thanks for the review.

> 
> > Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK
> > commit?
> > 
> > gcc/cp/ChangeLog:
> > 
> > 	PR c++/95486
> > 	* pt.c (alias_ctad_tweaks): Obtain the associated constraints
> > 	through fprime instead of through f.  Call remove_constraints
> > 	before calling set_constraints.
> > 
> > gcc/testsuite/ChangeLog:
> > 
> > 	PR c++/95486
> > 	* g++.dg/cpp2a/class-deduction-alias3.C: New test.
> > ---
> >   gcc/cp/pt.c                                         |  7 +++++--
> >   gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C | 11 +++++++++++
> >   2 files changed, 16 insertions(+), 2 deletions(-)
> >   create mode 100644 gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C
> > 
> > diff --git a/gcc/cp/pt.c b/gcc/cp/pt.c
> > index b6df87a8a2e..4d955c555dc 100644
> > --- a/gcc/cp/pt.c
> > +++ b/gcc/cp/pt.c
> > @@ -28598,7 +28598,7 @@ alias_ctad_tweaks (tree tmpl, tree uguides)
> >   	  DECL_PRIMARY_TEMPLATE (fprime) = fprime;
> >     	  /* Substitute the associated constraints.  */
> > -	  tree ci = get_constraints (f);
> > +	  tree ci = get_constraints (fprime);
> >   	  if (ci)
> >   	    ci = tsubst_constraint_info (ci, targs, complain, in_decl);
> >   	  if (ci == error_mark_node)
> > @@ -28616,7 +28616,10 @@ alias_ctad_tweaks (tree tmpl, tree uguides)
> >   	    }
> >     	  if (ci)
> > -	    set_constraints (fprime, ci);
> > +	    {
> > +	      remove_constraints (fprime);
> > +	      set_constraints (fprime, ci);
> > +	    }
> >   	}
> >         else
> >   	{
> > diff --git a/gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C
> > b/gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C
> > new file mode 100644
> > index 00000000000..318d4c942ce
> > --- /dev/null
> > +++ b/gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C
> > @@ -0,0 +1,11 @@
> > +// PR c++/95486
> > +// { dg-do compile { target c++20 } }
> > +
> > +template<class T, class U>
> > +struct X { X(U) requires __is_same(U, int) {} };
> > +
> > +template<class U>
> > +using Y = X<void, U>;
> > +
> > +Y y{1};
> > +Y z{'a'}; // { dg-error "failed|no match" }
> > 
> 
>
diff mbox series

Patch

diff --git a/gcc/cp/pt.c b/gcc/cp/pt.c
index b6df87a8a2e..4d955c555dc 100644
--- a/gcc/cp/pt.c
+++ b/gcc/cp/pt.c
@@ -28598,7 +28598,7 @@  alias_ctad_tweaks (tree tmpl, tree uguides)
 	  DECL_PRIMARY_TEMPLATE (fprime) = fprime;
 
 	  /* Substitute the associated constraints.  */
-	  tree ci = get_constraints (f);
+	  tree ci = get_constraints (fprime);
 	  if (ci)
 	    ci = tsubst_constraint_info (ci, targs, complain, in_decl);
 	  if (ci == error_mark_node)
@@ -28616,7 +28616,10 @@  alias_ctad_tweaks (tree tmpl, tree uguides)
 	    }
 
 	  if (ci)
-	    set_constraints (fprime, ci);
+	    {
+	      remove_constraints (fprime);
+	      set_constraints (fprime, ci);
+	    }
 	}
       else
 	{
diff --git a/gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C b/gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C
new file mode 100644
index 00000000000..318d4c942ce
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp2a/class-deduction-alias3.C
@@ -0,0 +1,11 @@ 
+// PR c++/95486
+// { dg-do compile { target c++20 } }
+
+template<class T, class U>
+struct X { X(U) requires __is_same(U, int) {} };
+
+template<class U>
+using Y = X<void, U>;
+
+Y y{1};
+Y z{'a'}; // { dg-error "failed|no match" }