Patchwork RFA: Fix bootstrap/44637

login
register
mail settings
Submitter Joern Rennecke
Date July 10, 2010, 8:04 a.m.
Message ID <20100710040444.im0gaauyo4woo0go-nzlynne@webmail.spamcop.net>
Download mbox | patch
Permalink /patch/58465/
State New
Headers show

Comments

Joern Rennecke - July 10, 2010, 8:04 a.m.
I've tried to do bootstrapping on a playstation, but I couldn't even get a
baseline - although the compilers were built on the first day, it spent six
more days building java libraries (and swapping, CPU usage around 1%),
before the shell aborted for some unknown reason, and the machine locked up.

With the dearth of PPC connectivity / CPU cycles on the compile farm,
I'm afraid I can only offer cross tests.

built all-gcc for i686-pc-linux-gnu X ppc-linux-gnu with configure option
--enable-werror-always and gcc (GCC) 4.6.0 20100630 (experimental);

built all-gcc for i686-pc-linux-gnu X ppc-linux-gnu with configure options
--enable-werror-always --enable-build-with-cxx and
g++ (GCC) 4.6.0 20100630 (experimental).
2010-06-23  Joern Rennecke  <joern.rennecke@embecosm.com>

	PR bootstrap/44637
	* config/rs6000/rs6000.c (rs6000_xcoff_strip_dollar): Use const char *
	variable for const char * strchr result.
Richard Guenther - July 10, 2010, 8:19 a.m.
On Sat, Jul 10, 2010 at 10:04 AM, Joern Rennecke <amylaar@spamcop.net> wrote:
> I've tried to do bootstrapping on a playstation, but I couldn't even get a
> baseline - although the compilers were built on the first day, it spent six
> more days building java libraries (and swapping, CPU usage around 1%),
> before the shell aborted for some unknown reason, and the machine locked up.
>
> With the dearth of PPC connectivity / CPU cycles on the compile farm,
> I'm afraid I can only offer cross tests.
>
> built all-gcc for i686-pc-linux-gnu X ppc-linux-gnu with configure option
> --enable-werror-always and gcc (GCC) 4.6.0 20100630 (experimental);
>
> built all-gcc for i686-pc-linux-gnu X ppc-linux-gnu with configure options
> --enable-werror-always --enable-build-with-cxx and
> g++ (GCC) 4.6.0 20100630 (experimental).
>
> 2010-06-23  Joern Rennecke  <joern.rennecke@embecosm.com>
>
>        PR bootstrap/44637
>        * config/rs6000/rs6000.c (rs6000_xcoff_strip_dollar): Use const char
> *
>        variable for const char * strchr result.
>
> Index: config/rs6000/rs6000.c
> ===================================================================
> --- config/rs6000/rs6000.c      (revision 161213)
> +++ config/rs6000/rs6000.c      (working copy)
> @@ -21252,12 +21252,13 @@ toc_hash_eq (const void *h1, const void
>  const char *
>  rs6000_xcoff_strip_dollar (const char *name)
>  {
> +  const char *cp;
>   char *strip, *p;
>   int len;
>
> -  p = strchr (name, '$');
> +  cp = strchr (name, '$');
>
> -  if (p == 0 || p == name)
> +  if (cp == 0 || cp == name)
>     return name;
>
>   len = strlen (name);

Huh, I wonder why the same diagnostic is not issued for the code below which
reads

  strip = (char *) alloca (len + 1);
  strcpy (strip, name);
  p = strchr (strip, '$');
  while (p)
    {
      *p = '_';
      p = strchr (p + 1, '$');
    }

why not use our "friendly" CONST_CAST macro here?

Richard.
>
Jakub Jelinek - July 10, 2010, 8:32 a.m.
On Sat, Jul 10, 2010 at 10:19:49AM +0200, Richard Guenther wrote:
> > --- config/rs6000/rs6000.c      (revision 161213)
> > +++ config/rs6000/rs6000.c      (working copy)
> > @@ -21252,12 +21252,13 @@ toc_hash_eq (const void *h1, const void
> >  const char *
> >  rs6000_xcoff_strip_dollar (const char *name)
> >  {
> > +  const char *cp;
> >   char *strip, *p;
> >   int len;
> >
> > -  p = strchr (name, '$');
> > +  cp = strchr (name, '$');
> >
> > -  if (p == 0 || p == name)
> > +  if (cp == 0 || cp == name)
> >     return name;
> >
> >   len = strlen (name);
> 
> Huh, I wonder why the same diagnostic is not issued for the code below which
> reads
> 
>   strip = (char *) alloca (len + 1);
>   strcpy (strip, name);
>   p = strchr (strip, '$');
>   while (p)
>     {
>       *p = '_';
>       p = strchr (p + 1, '$');
>     }
> 

Because strchr is char *strchr (const char *, int); for C
and char *strchr (char *, int); const char *strchr (const char *, int); for C++?
In the first snippet the argument is const char *, so the result is const
char * too (for C++ only, otherwise char *), while in the second snippet
the argument is char *, so the result is char * too.

> why not use our "friendly" CONST_CAST macro here?

Why?

	Jakub
Richard Guenther - July 10, 2010, 8:34 a.m.
On Sat, Jul 10, 2010 at 10:32 AM, Jakub Jelinek <jakub@redhat.com> wrote:
> On Sat, Jul 10, 2010 at 10:19:49AM +0200, Richard Guenther wrote:
>> > --- config/rs6000/rs6000.c      (revision 161213)
>> > +++ config/rs6000/rs6000.c      (working copy)
>> > @@ -21252,12 +21252,13 @@ toc_hash_eq (const void *h1, const void
>> >  const char *
>> >  rs6000_xcoff_strip_dollar (const char *name)
>> >  {
>> > +  const char *cp;
>> >   char *strip, *p;
>> >   int len;
>> >
>> > -  p = strchr (name, '$');
>> > +  cp = strchr (name, '$');
>> >
>> > -  if (p == 0 || p == name)
>> > +  if (cp == 0 || cp == name)
>> >     return name;
>> >
>> >   len = strlen (name);
>>
>> Huh, I wonder why the same diagnostic is not issued for the code below which
>> reads
>>
>>   strip = (char *) alloca (len + 1);
>>   strcpy (strip, name);
>>   p = strchr (strip, '$');
>>   while (p)
>>     {
>>       *p = '_';
>>       p = strchr (p + 1, '$');
>>     }
>>
>
> Because strchr is char *strchr (const char *, int); for C
> and char *strchr (char *, int); const char *strchr (const char *, int); for C++?
> In the first snippet the argument is const char *, so the result is const
> char * too (for C++ only, otherwise char *), while in the second snippet
> the argument is char *, so the result is char * too.

Ah, that makes sense.  I didn't look closely enough.

>> why not use our "friendly" CONST_CAST macro here?
>
> Why?

Not necessary in that case then.

Richard.

>        Jakub
>

Patch

Index: config/rs6000/rs6000.c
===================================================================
--- config/rs6000/rs6000.c	(revision 161213)
+++ config/rs6000/rs6000.c	(working copy)
@@ -21252,12 +21252,13 @@  toc_hash_eq (const void *h1, const void 
 const char *
 rs6000_xcoff_strip_dollar (const char *name)
 {
+  const char *cp;
   char *strip, *p;
   int len;
 
-  p = strchr (name, '$');
+  cp = strchr (name, '$');
 
-  if (p == 0 || p == name)
+  if (cp == 0 || cp == name)
     return name;
 
   len = strlen (name);