C++: Fix PR86083

Message ID 20180620114107.11570-1-krebbel@linux.vnet.ibm.com
State New
Headers show
Series
  • C++: Fix PR86083
Related show

Commit Message

Andreas Krebbel June 20, 2018, 11:41 a.m.
When turning a user-defined numerical literal into an operator
invocation the literal needs to be translated to the execution
character set.

Bootstrapped and regtested on s390x. x86_64 still running.
Ok to apply if x86_64 is clean?

Bye,

-Andreas-

gcc/cp/ChangeLog:

2018-06-20  Andreas Krebbel  <krebbel@linux.ibm.com>

	PR C++/86082
	* parser.c (make_char_string_pack):
	(cp_parser_userdef_numeric_literal):

gcc/testsuite/ChangeLog:

2018-06-20  Andreas Krebbel  <krebbel@linux.ibm.com>

	PR C++/86082
	* g++.dg/pr86082.C: New test.
---
 gcc/cp/parser.c                | 17 ++++++++++++++++-
 gcc/testsuite/g++.dg/pr86082.C | 18 ++++++++++++++++++
 2 files changed, 34 insertions(+), 1 deletion(-)
 create mode 100644 gcc/testsuite/g++.dg/pr86082.C

-- 
2.9.1

Comments

Jason Merrill June 20, 2018, 3:05 p.m. | #1
On Wed, Jun 20, 2018 at 7:41 AM, Andreas Krebbel
<krebbel@linux.vnet.ibm.com> wrote:
> When turning a user-defined numerical literal into an operator

> invocation the literal needs to be translated to the execution

> character set.

>

> Bootstrapped and regtested on s390x. x86_64 still running.

> Ok to apply if x86_64 is clean?

>

> Bye,

>

> -Andreas-

>

> gcc/cp/ChangeLog:

>

> 2018-06-20  Andreas Krebbel  <krebbel@linux.ibm.com>

>

>         PR C++/86082

>         * parser.c (make_char_string_pack):

>         (cp_parser_userdef_numeric_literal):


You're missing a description of the changes.  OK with that corrected.

Jason
Rainer Orth June 26, 2018, 9:17 a.m. | #2
Hi Andreas,

> When turning a user-defined numerical literal into an operator

> invocation the literal needs to be translated to the execution

> character set.

>

> Bootstrapped and regtested on s390x. x86_64 still running.

> Ok to apply if x86_64 is clean?


the new testcase FAILs on Solaris

FAIL: g++.dg/pr86082.C   (test for excess errors)

Excess errors:
cc1plus: error: conversion from UTF-8 to IBM1047 not supported by iconv

and, according to gcc-testresults, on powerpc-ibm-aix7.2.0.0.

Fixed as follows, tested on i386-pc-solaris2.11, installed on mainline.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University


2018-06-26  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>

	* g++.dg/pr86082.C: Require IBM1047 support.
# HG changeset patch
# Parent  eeab29b53ad0ae8a16bb9e8d4d0f934e9d5ebebe
Require IBM1047 support in g++.dg/pr86082.C

diff --git a/gcc/testsuite/g++.dg/pr86082.C b/gcc/testsuite/g++.dg/pr86082.C
--- a/gcc/testsuite/g++.dg/pr86082.C
+++ b/gcc/testsuite/g++.dg/pr86082.C
@@ -1,4 +1,5 @@
 /* { dg-do link } */
+/* { dg-require-iconv "IBM1047" } */
 /* { dg-options "-fexec-charset=IBM1047 -std=c++11" } */
 
 /* When turning 123_test into an operator invocation the literal 123
Andreas Krebbel June 26, 2018, 9:19 a.m. | #3
On 06/26/2018 11:17 AM, Rainer Orth wrote:
> Hi Andreas,

> 

>> When turning a user-defined numerical literal into an operator

>> invocation the literal needs to be translated to the execution

>> character set.

>>

>> Bootstrapped and regtested on s390x. x86_64 still running.

>> Ok to apply if x86_64 is clean?

> 

> the new testcase FAILs on Solaris

> 

> FAIL: g++.dg/pr86082.C   (test for excess errors)

> 

> Excess errors:

> cc1plus: error: conversion from UTF-8 to IBM1047 not supported by iconv

> 

> and, according to gcc-testresults, on powerpc-ibm-aix7.2.0.0.

> 

> Fixed as follows, tested on i386-pc-solaris2.11, installed on mainline.


Thanks for fixing it!

-Andreas-
Andreas Krebbel July 6, 2018, 7:20 a.m. | #4
On 06/20/2018 01:41 PM, Andreas Krebbel wrote:
> When turning a user-defined numerical literal into an operator

> invocation the literal needs to be translated to the execution

> character set.

> 

> Bootstrapped and regtested on s390x. x86_64 still running.

> Ok to apply if x86_64 is clean?

> 

> Bye,

> 

> -Andreas-

> 

> gcc/cp/ChangeLog:

> 

> 2018-06-20  Andreas Krebbel  <krebbel@linux.ibm.com>

> 

> 	PR C++/86082

> 	* parser.c (make_char_string_pack):

> 	(cp_parser_userdef_numeric_literal):

> 

> gcc/testsuite/ChangeLog:

> 

> 2018-06-20  Andreas Krebbel  <krebbel@linux.ibm.com>

> 

> 	PR C++/86082

> 	* g++.dg/pr86082.C: New test.


I've tested the patch also on GCC 7 and 8 branch. Ok to apply there as well?

The backport will include the testcase fix from Rainer:
https://gcc.gnu.org/ml/gcc-patches/2018-06/msg01601.html

-Andreas-
Jason Merrill July 8, 2018, 12:44 p.m. | #5
On Fri, Jul 6, 2018 at 5:20 PM, Andreas Krebbel <krebbel@linux.ibm.com> wrote:
> On 06/20/2018 01:41 PM, Andreas Krebbel wrote:

>> When turning a user-defined numerical literal into an operator

>> invocation the literal needs to be translated to the execution

>> character set.

>>

>> Bootstrapped and regtested on s390x. x86_64 still running.

>> Ok to apply if x86_64 is clean?

>>

>> Bye,

>>

>> -Andreas-

>>

>> gcc/cp/ChangeLog:

>>

>> 2018-06-20  Andreas Krebbel  <krebbel@linux.ibm.com>

>>

>>       PR C++/86082

>>       * parser.c (make_char_string_pack):

>>       (cp_parser_userdef_numeric_literal):

>>

>> gcc/testsuite/ChangeLog:

>>

>> 2018-06-20  Andreas Krebbel  <krebbel@linux.ibm.com>

>>

>>       PR C++/86082

>>       * g++.dg/pr86082.C: New test.

>

> I've tested the patch also on GCC 7 and 8 branch. Ok to apply there as well?


Hmm, it seems safe enough, but also seems like a change with a limited
audience.  Would it work for you to include the patch just in your
builds of GCC 7 and 8?

Jason

Patch

diff --git a/gcc/cp/parser.c b/gcc/cp/parser.c
index c6206fc..9522861 100644
--- a/gcc/cp/parser.c
+++ b/gcc/cp/parser.c
@@ -4291,7 +4291,16 @@  make_char_string_pack (tree value)
   /* Fill in CHARVEC with all of the parameters.  */
   charvec = make_tree_vec (len);
   for (i = 0; i < len; ++i)
-    TREE_VEC_ELT (charvec, i) = build_int_cst (char_type_node, str[i]);
+    {
+      unsigned char s[3] = { '\'', str[i], '\'' };
+      cpp_string in = { 3, s };
+      cpp_string out = { 0, 0 };
+      if (!cpp_interpret_string (parse_in, &in, 1, &out, CPP_STRING))
+	return NULL_TREE;
+      gcc_assert (out.len == 2);
+      TREE_VEC_ELT (charvec, i) = build_int_cst (char_type_node,
+						 out.text[0]);
+    }
 
   /* Build the argument packs.  */
   SET_ARGUMENT_PACK_ARGS (argpack, charvec);
@@ -4407,6 +4416,12 @@  cp_parser_userdef_numeric_literal (cp_parser *parser)
   if (decl && decl != error_mark_node)
     {
       tree tmpl_args = make_char_string_pack (num_string);
+      if (tmpl_args == NULL_TREE)
+	{
+	  error ("failed to translate literal to execution character set %qT",
+		 num_string);
+	  return error_mark_node;
+	}
       decl = lookup_template_function (decl, tmpl_args);
       result = finish_call_expr (decl, &args, false, true,
 				 tf_warning_or_error);
diff --git a/gcc/testsuite/g++.dg/pr86082.C b/gcc/testsuite/g++.dg/pr86082.C
new file mode 100644
index 0000000..c7247cd
--- /dev/null
+++ b/gcc/testsuite/g++.dg/pr86082.C
@@ -0,0 +1,18 @@ 
+/* { dg-do link } */
+/* { dg-options "-fexec-charset=IBM1047 -std=c++11" } */
+
+/* When turning 123_test into an operator invocation the literal 123
+   needs to be translated to the execution character set.  Failing to
+   do so results in a link error since '1', '2', and '3' in the
+   specialization will be translated as string literals.  */
+
+template <char... p> void q();
+template <> void q<'1','2','3'>() {}
+
+template <char... p> void operator""_test() { q<p...> (); }
+
+int
+main ()
+{
+  123_test;
+}