public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Patrick Palka <ppalka@redhat.com>
To: Patrick Palka <ppalka@redhat.com>
Cc: gcc-patches@gcc.gnu.org, jason@redhat.com
Subject: Re: [PATCH] c++: partial ordering with dependent NTTP type [PR105289]
Date: Fri, 22 Apr 2022 15:27:12 -0400 (EDT)	[thread overview]
Message-ID: <6292a77d-f373-4158-036d-9a56e599bccf@idea> (raw)
In-Reply-To: <20220422175032.3106274-1-ppalka@redhat.com>

On Fri, 22 Apr 2022, Patrick Palka wrote:

> Here ever since r11-6483-ge2e2f3f2c9400f we're rejecting and crashing
> (respectively) on two testcases that we used to accept in C++17 mode.
> Both testcases declare partial specializations for which the primary
> template contains an NTTP with dependent type, but the correctness of
> these partial specializations is unclear according to PR86193.
> 
> This patch restores the previous C++17 behavior for such partial
> specializations by restricting the r11-6483 change to just ordinary
> deduction as opposed to deduction for sake of partial ordering.

Note that if we're okay with rejecting such partial specializations even
in C++17 mode (and thus deeming PR105289 to be ICE-on-invalid instead of
ICE-on-valid), then the fix for the reported ICE is just:

diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
index dde62ee052d..6d65f6ad3cf 100644
--- a/gcc/cp/pt.cc
+++ b/gcc/cp/pt.cc
@@ -24299,7 +24299,7 @@ unify (tree tparms, tree targs, tree parm, tree arg, int strict,
 	      /* Now check whether the type of this parameter is still
 		 dependent, and give up if so.  */
 	      ++processing_template_decl;
-	      tparm = tsubst (tparm, targs, tf_none, NULL_TREE);
+	      tparm = tsubst (TREE_TYPE (parm), targs, tf_none, NULL_TREE);
 	      --processing_template_decl;
 	      if (uses_template_parms (tparm))
 		return unify_success (explain_p);

i.e. we need to substitute into the original NTTP type, not into the
already substituted NTTP type.

> 
> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
> trunk/11?
> 
> 	PR c++/105289
> 	PR c++/86193
> 
> gcc/cp/ChangeLog:
> 
> 	* pt.cc (unify) <case TEMPLATE_PARM_INDEX>: Restrict the
> 	r11-6483 change to just ordinary deduction for function
> 	templates.  When substituting into the NTTP type the second
> 	time, use the original type not the substituted type.  Remove
> 	now unnecessary level check.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* g++.dg/template/partial5.C: Revert r11-6483 change.
> 	* g++.dg/template/partial-specialization11.C: New test.
> 	* g++.dg/template/partial-specialization12.C: New test.
> ---
>  gcc/cp/pt.cc                                  | 25 ++++++++++++-------
>  .../template/partial-specialization11.C       | 10 ++++++++
>  .../template/partial-specialization12.C       | 12 +++++++++
>  gcc/testsuite/g++.dg/template/partial5.C      |  2 +-
>  4 files changed, 39 insertions(+), 10 deletions(-)
>  create mode 100644 gcc/testsuite/g++.dg/template/partial-specialization11.C
>  create mode 100644 gcc/testsuite/g++.dg/template/partial-specialization12.C
> 
> diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
> index dde62ee052d..52bd130b7e7 100644
> --- a/gcc/cp/pt.cc
> +++ b/gcc/cp/pt.cc
> @@ -24287,8 +24287,7 @@ unify (tree tparms, tree targs, tree parm, tree arg, int strict,
>  	  /* We haven't deduced the type of this parameter yet.  */
>  	  if (cxx_dialect >= cxx17
>  	      /* We deduce from array bounds in try_array_deduction.  */
> -	      && !(strict & UNIFY_ALLOW_INTEGER)
> -	      && TEMPLATE_PARM_LEVEL (parm) <= TMPL_ARGS_DEPTH (targs))
> +	      && !(strict & UNIFY_ALLOW_INTEGER))
>  	    {
>  	      /* Deduce it from the non-type argument.  As above, ignore
>  		 top-level quals here too.  */
> @@ -24296,13 +24295,21 @@ unify (tree tparms, tree targs, tree parm, tree arg, int strict,
>  	      RECUR_AND_CHECK_FAILURE (tparms, targs,
>  				       tparm, atype,
>  				       UNIFY_ALLOW_NONE, explain_p);
> -	      /* Now check whether the type of this parameter is still
> -		 dependent, and give up if so.  */
> -	      ++processing_template_decl;
> -	      tparm = tsubst (tparm, targs, tf_none, NULL_TREE);
> -	      --processing_template_decl;
> -	      if (uses_template_parms (tparm))
> -		return unify_success (explain_p);
> +	      if (!processing_template_decl
> +		  && TPARMS_PRIMARY_TEMPLATE (tparms)
> +		  && DECL_FUNCTION_TEMPLATE_P (TPARMS_PRIMARY_TEMPLATE
> +					       (tparms)))
> +		{
> +		  /* If the NTTP's type uses still-undeduced template
> +		     parameters, then don't unify it now.  This gives
> +		     type_unification_real a chance to retry deduction
> +		     with default template arguments substituted in.  */
> +		  ++processing_template_decl;
> +		  tparm = tsubst (TREE_TYPE (parm), targs, tf_none, NULL_TREE);
> +		  --processing_template_decl;
> +		  if (uses_template_parms (tparm))
> +		    return unify_success (explain_p);
> +		}
>  	    }
>  	  else
>  	    /* Try again later.  */
> diff --git a/gcc/testsuite/g++.dg/template/partial-specialization11.C b/gcc/testsuite/g++.dg/template/partial-specialization11.C
> new file mode 100644
> index 00000000000..20da407d422
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/template/partial-specialization11.C
> @@ -0,0 +1,10 @@
> +// PR c++/105289
> +
> +template<class T> struct value_type;
> +
> +template<class T, typename value_type<T>::type V>
> +struct push_front_vlist;
> +
> +template<class T, int V>
> +struct push_front_vlist<T*, V> { };
> +// { dg-error "not more specialized" "PR86193" { target c++14_down } .-1 }
> diff --git a/gcc/testsuite/g++.dg/template/partial-specialization12.C b/gcc/testsuite/g++.dg/template/partial-specialization12.C
> new file mode 100644
> index 00000000000..d70f7592790
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/template/partial-specialization12.C
> @@ -0,0 +1,12 @@
> +// PR c++/105289
> +// { dg-do compile { target c++11 } }
> +
> +template<class T>
> +struct value_type;
> +
> +template <class List, typename value_type<List>::type Element>
> +struct push_front_vlist;
> +
> +template <template <class X, X...> class XList, class T, T Arg, T... Vs>
> +struct push_front_vlist<XList<T, Vs...>, Arg> { };
> +// { dg-error "not more specialized" "PR86193" { target c++14_down } .-1 }
> diff --git a/gcc/testsuite/g++.dg/template/partial5.C b/gcc/testsuite/g++.dg/template/partial5.C
> index 037f684cbd2..7492632f39b 100644
> --- a/gcc/testsuite/g++.dg/template/partial5.C
> +++ b/gcc/testsuite/g++.dg/template/partial5.C
> @@ -14,7 +14,7 @@ template<typename T, typename T::foo V>
>  struct Y { };
>  
>  template<typename T, typename U, U v>
> -struct Y<T, v> { }; // { dg-error "" }
> +struct Y<T, v> { }; // { dg-error "" "PR86193" { target c++14_down } }
>  
>  
>  template<typename T, T V>
> -- 
> 2.36.0.rc2.10.g1ac7422e39
> 
> 


  reply	other threads:[~2022-04-22 19:27 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-04-22 17:50 Patrick Palka
2022-04-22 19:27 ` Patrick Palka [this message]
2022-04-25 15:12   ` Jason Merrill
2022-04-25 18:10     ` Patrick Palka
2022-04-25 21:53       ` Jason Merrill
2022-04-22 18:33 Patrick Palka
2022-04-22 18:41 ` Patrick Palka

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=6292a77d-f373-4158-036d-9a56e599bccf@idea \
    --to=ppalka@redhat.com \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=jason@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).