[diagnostic patch] PR 54941

[diagnostic patch] PR 54941

[Paolo Carlini]

[-- Attachment #1: Type: text/plain, Size: 480 bytes --]

Hi all, Gaby,

in this bug Manuel noticed that the zeros in the diagnostic lines of the 
form:

<built-in>:0:0: ....

don't provide useful information. Thus the below just avoids printing 
the zeros basing directly on the file name: admittedly, it may seem a 
bit gross, but in practice the strcmp would often fail early, and should 
be rather efficient anyway because the second argument is known at 
compile-time.

Tested x86_64-linux.

Thanks,
Paolo.

////////////////////////

[-- Attachment #2: CL_54941 --]
[-- Type: text/plain, Size: 295 bytes --]

2013-09-08  Paolo Carlini  <paolo.carlini@oracle.com>

	PR c++/54941
	* diagnostic.c (diagnostic_build_prefix): When s.file is
	"<built-in>" don't output line and column numbers.

/testsuite
2013-09-08  Paolo Carlini  <paolo.carlini@oracle.com>

	PR c++/54941
	* g++.dg/overload/new1.C: Adjust.

[-- Attachment #3: patch_54941 --]
[-- Type: text/plain, Size: 1226 bytes --]

Index: diagnostic.c
===================================================================
--- diagnostic.c	(revision 202358)
+++ diagnostic.c	(working copy)
@@ -245,6 +245,9 @@ diagnostic_build_prefix (diagnostic_context *conte
     (s.file == NULL
      ? build_message_string ("%s%s:%s %s%s%s", locus_cs, progname, locus_ce,
 			     text_cs, text, text_ce)
+     : !strcmp (s.file, N_("<built-in>"))
+     ? build_message_string ("%s%s:%s %s%s%s", locus_cs, s.file, locus_ce,
+			     text_cs, text, text_ce)
      : context->show_column
      ? build_message_string ("%s%s:%d:%d:%s %s%s%s", locus_cs, s.file, s.line,
 			     s.column, locus_ce, text_cs, text, text_ce)
Index: testsuite/g++.dg/overload/new1.C
===================================================================
--- testsuite/g++.dg/overload/new1.C	(revision 202358)
+++ testsuite/g++.dg/overload/new1.C	(working copy)
@@ -17,6 +17,5 @@ void f(X *x = new (3) X(6));   // { dg-error "" }
 
 void f(X *x = new (2) X[10]);  // { dg-error "" } 
 // { dg-message "candidate" "candidate note" { target *-*-* } 18 }
-// { dg-message "operator new|candidate expects" "match candidate text" { target *-*-* } 00 }
 
 void f(X *x = new X[10][5]);   // { dg-error "" } 

Re: [diagnostic patch] PR 54941

[Gabriel Dos Reis]

On Sun, Sep 8, 2013 at 8:00 AM, Paolo Carlini <paolo.carlini@oracle.com> wrote:
> Hi all, Gaby,
>
> in this bug Manuel noticed that the zeros in the diagnostic lines of the
> form:
>
> <built-in>:0:0: ....
>
> don't provide useful information. Thus the below just avoids printing the
> zeros basing directly on the file name: admittedly, it may seem a bit gross,
> but in practice the strcmp would often fail early, and should be rather
> efficient anyway because the second argument is known at compile-time.

For builtins, we probably do not want to print the line number
and the column number (both of which are zero).

>
> Tested x86_64-linux.
>
> Thanks,
> Paolo.
>
> ////////////////////////

Re: [diagnostic patch] PR 54941

[Paolo Carlini]

Hi Gaby,

On 09/08/2013 03:41 PM, Gabriel Dos Reis wrote:
> On Sun, Sep 8, 2013 at 8:00 AM, Paolo Carlini <paolo.carlini@oracle.com> wrote:
>> Hi all, Gaby,
>>
>> in this bug Manuel noticed that the zeros in the diagnostic lines of the
>> form:
>>
>> <built-in>:0:0: ....
>>
>> don't provide useful information. Thus the below just avoids printing the
>> zeros basing directly on the file name: admittedly, it may seem a bit gross,
>> but in practice the strcmp would often fail early, and should be rather
>> efficient anyway because the second argument is known at compile-time.
> For builtins, we probably do not want to print the line number
> and the column number (both of which are zero).
I agree. This is exactly what my patch does, isn't it?

Thanks,
Paolo.

Re: [diagnostic patch] PR 54941

[Gabriel Dos Reis]

On Sun, Sep 8, 2013 at 8:59 AM, Paolo Carlini <paolo.carlini@oracle.com> wrote:
> Hi Gaby,
>
>
> On 09/08/2013 03:41 PM, Gabriel Dos Reis wrote:
>>
>> On Sun, Sep 8, 2013 at 8:00 AM, Paolo Carlini <paolo.carlini@oracle.com>
>> wrote:
>>>
>>> Hi all, Gaby,
>>>
>>> in this bug Manuel noticed that the zeros in the diagnostic lines of the
>>> form:
>>>
>>> <built-in>:0:0: ....
>>>
>>> don't provide useful information. Thus the below just avoids printing the
>>> zeros basing directly on the file name: admittedly, it may seem a bit
>>> gross,
>>> but in practice the strcmp would often fail early, and should be rather
>>> efficient anyway because the second argument is known at compile-time.
>>
>> For builtins, we probably do not want to print the line number
>> and the column number (both of which are zero).
>
> I agree. This is exactly what my patch does, isn't it?

What I meant is that we should not showing a location at all.
I misread the original hunk

+     : !strcmp (s.file, N_("<built-in>"))
+     ? build_message_string ("%s%s:%s %s%s%s", locus_cs, s.file, locus_ce,
+     text_cs, text, text_ce)

Sorry.  Patch OK.

>
> Thanks,
> Paolo.