ld testsuite breaks with latest GCC-4.0

ld testsuite breaks with latest GCC-4.0

[Greg Schafer]

Hi

In case you haven't already heard, GCC-4.0 no longer installs a "specs"
file. See these threads:

  http://gcc.gnu.org/ml/gcc-patches/2004-12/msg01459.html
  http://gcc.gnu.org/ml/gcc-patches/2004-12/msg01574.html
  
The ld testsuite currently attempts some grovelling around in the gcc specs
file to find out various bits of information. This from ld/configure.host
(excuse the wrap):


  *-*-linux*)
  HOSTING_CRT0='-dynamic-linker `egrep "ld[^ ]*\.so" \`${CC} --print-file-name=specs\` | sed 
-e "s,.*-dynamic-linker[   ][ ]*\(.*/ld[^ ]*\.so..\).*,\1,"` `${CC} --print-file-name=crt1.o` 
`${CC} --print-file-name=crti.o` `if [ -f ../gcc/crtbegin.o ]; then echo ../gcc/crtbegin.o; else 
${CC} --print-file-name=crtbegin.o; fi`'

  
With latest GCC-4.0, `gcc --print-file-name=specs' spits out nothing which
breaks the whole ld testsuite:


Using /temptools/src/binutils-2.15.94.0.2/ld/testsuite/config/default.exp as tool-and-target-specific interface file.
ERROR: tcl error sourcing tool-and-target-specific interface file /temptools/src/binutils-2.15.94.0.2/ld/testsuite/config/default.exp.
Error getting native link files: -dynamic-linker /usr/lib/crt1.o /usr/lib/crti.o /temptools/lib/gcc/i686-pc-linux-gnu/4.0.0/crtbegin.o
grep: specs: No such file or directory
Error getting native link files: -dynamic-linker /usr/lib/crt1.o /usr/lib/crti.o /temptools/lib/gcc/i686-pc-linux-gnu/4.0.0/crtbegin.o
grep: specs: No such file or directory
    while executing


Clearly, changes will be needed. I'm not sure what the best solution is but
I have a feeling that it will probably involve:

  `gcc -dumpspecs'

but that might have the potential to break older GCC's. Dunno.

Thoughts?

Thanks
Greg

PS - I'm using the latest HJ release but the code in question from CVS trunk
looks the same.

Re: ld testsuite breaks with latest GCC-4.0

[Daniel Jacobowitz]

On Sun, Jan 02, 2005 at 03:44:51PM +1100, Greg Schafer wrote:
> Clearly, changes will be needed. I'm not sure what the best solution is but
> I have a feeling that it will probably involve:
> 
>   `gcc -dumpspecs'
> 
> but that might have the potential to break older GCC's. Dunno.

That should work, as far back as I know of.

-- 
Daniel Jacobowitz

[PATCH] Re: ld testsuite breaks with latest GCC-4.0

[Greg Schafer]

[-- Attachment #1: Type: text/plain, Size: 665 bytes --]

On Sun, Jan 02, 2005 at 12:11:44AM -0500, Daniel Jacobowitz wrote:
> On Sun, Jan 02, 2005 at 03:44:51PM +1100, Greg Schafer wrote:
> > Clearly, changes will be needed. I'm not sure what the best solution is but
> > I have a feeling that it will probably involve:
> > 
> >   `gcc -dumpspecs'
> > 
> > but that might have the potential to break older GCC's. Dunno.
> 
> That should work, as far back as I know of.

I was just worried that some installations may be using an altered "on disk"
specs file.

Anyhow, the attached patch makes it work for me on i686-pc-linux-gnu. But it
also affects the ia64-*-aix* target but I have no way of testing that.

Regards
Greg

[-- Attachment #2: ld.patch --]
[-- Type: text/plain, Size: 2246 bytes --]

2005-01-02  Greg Schafer  <gschafer@zip.com.au>

	* configure.host: Use "${CC} -dumpspecs" instead of
	"${CC} --print-file-name=specs" to appease GCC versions >= 4.0.

Index: ld/configure.host
===================================================================
RCS file: /cvs/src/src/ld/configure.host,v
retrieving revision 1.36
diff -u -r1.36 configure.host
--- ld/configure.host	17 May 2004 19:50:16 -0000	1.36
+++ ld/configure.host	2 Jan 2005 06:03:40 -0000
@@ -33,7 +33,7 @@
   ;;
 
 *-*-linux*)
-  HOSTING_CRT0='-dynamic-linker `egrep "ld[^ ]*\.so" \`${CC} --print-file-name=specs\` | sed -e "s,.*-dynamic-linker[ 	][ 	]*\(.*/ld[^ ]*\.so..\).*,\1,"` `${CC} --print-file-name=crt1.o` `${CC} --print-file-name=crti.o` `if [ -f ../gcc/crtbegin.o ]; then echo ../gcc/crtbegin.o; else ${CC} --print-file-name=crtbegin.o; fi`'
+  HOSTING_CRT0='-dynamic-linker `${CC} -dumpspecs | egrep "ld[^ ]*\.so" | sed -e "s,.*-dynamic-linker[ 	][ 	]*\(.*/ld[^ ]*\.so..\).*,\1,"` `${CC} --print-file-name=crt1.o` `${CC} --print-file-name=crti.o` `if [ -f ../gcc/crtbegin.o ]; then echo ../gcc/crtbegin.o; else ${CC} --print-file-name=crtbegin.o; fi`'
   HOSTING_LIBS='-L`dirname \`${CC} --print-file-name=libc.so\`` '"$HOSTING_LIBS"' `if [ -f ../gcc/crtend.o ]; then echo ../gcc/crtend.o; else ${CC} --print-file-name=crtend.o; fi` `${CC} --print-file-name=crtn.o`'
   ;;
 
@@ -146,7 +146,7 @@
   ;;
 
 ia64-*-aix*)
-  HOSTING_CRT0='-dynamic-linker `egrep "libc.so" \`${CC} --print-file-name=specs\` | sed -e "s,.*-dynamic-linker[ 	][ 	]*\(.*/libc.so..\).*,\1,"` `${CC} --print-file-name=crt1.o` `${CC} --print-file-name=crti.o` `if [ -f ../gcc/crtbegin.o ]; then echo ../gcc/crtbegin.o; else ${CC} --print-file-name=crtbegin.o; fi`'
+  HOSTING_CRT0='-dynamic-linker `${CC} -dumpspecs | egrep "libc.so" | sed -e "s,.*-dynamic-linker[ 	][ 	]*\(.*/libc.so..\).*,\1,"` `${CC} --print-file-name=crt1.o` `${CC} --print-file-name=crti.o` `if [ -f ../gcc/crtbegin.o ]; then echo ../gcc/crtbegin.o; else ${CC} --print-file-name=crtbegin.o; fi`'
   HOSTING_LIBS='-L`dirname \`${CC} --print-file-name=libc.so\`` '"$HOSTING_LIBS"' `if [ -f ../gcc/crtend.o ]; then echo ../gcc/crtend.o; else ${CC} --print-file-name=crtend.o; fi` `${CC} --print-file-name=crtn.o`'
   ;;
 

Re: [PATCH] Re: ld testsuite breaks with latest GCC-4.0

[James E Wilson]

On Sat, 2005-01-01 at 22:24, Greg Schafer wrote:
> Anyhow, the attached patch makes it work for me on i686-pc-linux-gnu. But it
> also affects the ia64-*-aix* target but I have no way of testing that.

ia64-aix was never released to the public.  It is a dead target.  Go
ahead and change it anyway you like, no one will care.

The ia64-aix support was already removed from bfd, but I see it is still
in gas and ld, and maybe also other places I didn't look.  I should
really clean that up someday.
-- 
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com

Re: [PATCH] Re: ld testsuite breaks with latest GCC-4.0

[Nick Clifton]

Hi Greg,

> Anyhow, the attached patch makes it work for me on i686-pc-linux-gnu. But it
> also affects the ia64-*-aix* target but I have no way of testing that.



 > 2005-01-02  Greg Schafer  <gschafer@zip.com.au>
 >
 >	* configure.host: Use "${CC} -dumpspecs" instead of
 >	"${CC} --print-file-name=specs" to appease GCC versions >= 4.0.

Approved - please apply.

Cheers
   Nick

Re: [PATCH] Re: ld testsuite breaks with latest GCC-4.0

[Jakub Jelinek]

On Sun, Jan 02, 2005 at 05:24:00PM +1100, Greg Schafer wrote:

> 2005-01-02  Greg Schafer  <gschafer@zip.com.au>
> 
> 	* configure.host: Use "${CC} -dumpspecs" instead of
> 	"${CC} --print-file-name=specs" to appease GCC versions >= 4.0.

If there is a specs file, configure.host should use that and only use
-dumpspecs if it is not present.
-dumpspecs dumps the built-in specs, not the ones it finds on the
filesystem.
This can make a difference if specs is hand edited.

So, I think the patch should use ${CC} --print-file-name=specs,
if that prints just specs, it should use ${CC} -dumpspecs,
otherwise cat the specs file.
  specs=`${CC} --print-file-name=specs`
  { if [ x"$specs" = xspecs ]; then ${CC} -dumpspecs; else cat "$specs"; fi; } \
  | egrep ...

	Jakub

Re: [PATCH] Re: ld testsuite breaks with latest GCC-4.0

[Greg Schafer]

On Tue, Jan 04, 2005 at 11:43:26AM +0100, Jakub Jelinek wrote:

> If there is a specs file, configure.host should use that and only use
> -dumpspecs if it is not present.
> -dumpspecs dumps the built-in specs, not the ones it finds on the
> filesystem.
> This can make a difference if specs is hand edited.

Agreed. That is the exact scenario I was concerned about and alluded to up
thread.

> So, I think the patch should use ${CC} --print-file-name=specs,
> if that prints just specs, it should use ${CC} -dumpspecs,
> otherwise cat the specs file.
>   specs=`${CC} --print-file-name=specs`
>   { if [ x"$specs" = xspecs ]; then ${CC} -dumpspecs; else cat "$specs"; fi; } \
>   | egrep ...

Yes. That would seem saner and definitely more robust. I'll try to make a
new patch and test it but am running short on time right now.. so if someone
beats me to it.. go right ahead. Also, prolly best to remove that ia64-aix cruft
as confirmed by Jim Wilson.

Thanks
Greg

Re: [PATCH] Re: ld testsuite breaks with latest GCC-4.0

[Greg Schafer]

[-- Attachment #1: Type: text/plain, Size: 761 bytes --]

On Wed, Jan 05, 2005 at 09:47:30AM +1100, Greg Schafer wrote:
> On Tue, Jan 04, 2005 at 11:43:26AM +0100, Jakub Jelinek wrote:
> 
> > So, I think the patch should use ${CC} --print-file-name=specs,
> > if that prints just specs, it should use ${CC} -dumpspecs,
> > otherwise cat the specs file.
> >   specs=`${CC} --print-file-name=specs`
> >   { if [ x"$specs" = xspecs ]; then ${CC} -dumpspecs; else cat "$specs"; fi; } \
> >   | egrep ...
> 
> Yes. That would seem saner and definitely more robust. I'll try to make a
> new patch and test it

Here is a new patch that tested out fine for me with gcc-4 and gcc-3.4.3.
Nick approved the previous patch, so if this one is also approved, could
someone please apply it as I don't have commit priv's.

Thanks
Greg

[-- Attachment #2: ld.patch.new --]
[-- Type: text/plain, Size: 2313 bytes --]

2005-01-02  Jakub Jelinek  <jakub@redhat.com>
	    Greg Schafer  <gschafer@zip.com.au>

	* configure.host (*-*-linux*): Allow HOSTING_CRT0 to handle GCC
	versions >= 4.0 that don't install a specs file.
	(ia64-*-aix*): Remove support.

Index: ld/configure.host
===================================================================
RCS file: /cvs/src/src/ld/configure.host,v
retrieving revision 1.36
diff -u -r1.36 configure.host
--- ld/configure.host	17 May 2004 19:50:16 -0000	1.36
+++ ld/configure.host	6 Jan 2005 10:17:37 -0000
@@ -33,7 +33,7 @@
   ;;
 
 *-*-linux*)
-  HOSTING_CRT0='-dynamic-linker `egrep "ld[^ ]*\.so" \`${CC} --print-file-name=specs\` | sed -e "s,.*-dynamic-linker[ 	][ 	]*\(.*/ld[^ ]*\.so..\).*,\1,"` `${CC} --print-file-name=crt1.o` `${CC} --print-file-name=crti.o` `if [ -f ../gcc/crtbegin.o ]; then echo ../gcc/crtbegin.o; else ${CC} --print-file-name=crtbegin.o; fi`'
+  HOSTING_CRT0='-dynamic-linker `specs=\`${CC} --print-file-name=specs\`; { if [ x"$specs" = xspecs ]; then ${CC} -dumpspecs; else cat "$specs"; fi; } | egrep "ld[^ ]*\.so" | sed -e "s,.*-dynamic-linker[ 	][ 	]*\(.*/ld[^ ]*\.so..\).*,\1,"` `${CC} --print-file-name=crt1.o` `${CC} --print-file-name=crti.o` `if [ -f ../gcc/crtbegin.o ]; then echo ../gcc/crtbegin.o; else ${CC} --print-file-name=crtbegin.o; fi`'
   HOSTING_LIBS='-L`dirname \`${CC} --print-file-name=libc.so\`` '"$HOSTING_LIBS"' `if [ -f ../gcc/crtend.o ]; then echo ../gcc/crtend.o; else ${CC} --print-file-name=crtend.o; fi` `${CC} --print-file-name=crtn.o`'
   ;;
 
@@ -145,11 +145,6 @@
   HOSTING_CRT0=`echo "$HOSTING_CRT0" | sed -e "s,ld\[^ \]*\*,ld-linux-ia64,g"`
   ;;
 
-ia64-*-aix*)
-  HOSTING_CRT0='-dynamic-linker `egrep "libc.so" \`${CC} --print-file-name=specs\` | sed -e "s,.*-dynamic-linker[ 	][ 	]*\(.*/libc.so..\).*,\1,"` `${CC} --print-file-name=crt1.o` `${CC} --print-file-name=crti.o` `if [ -f ../gcc/crtbegin.o ]; then echo ../gcc/crtbegin.o; else ${CC} --print-file-name=crtbegin.o; fi`'
-  HOSTING_LIBS='-L`dirname \`${CC} --print-file-name=libc.so\`` '"$HOSTING_LIBS"' `if [ -f ../gcc/crtend.o ]; then echo ../gcc/crtend.o; else ${CC} --print-file-name=crtend.o; fi` `${CC} --print-file-name=crtn.o`'
-  ;;
-
 mips*-sgi-irix4* | mips*-sgi-irix5*)
   HOSTING_CRT0=/usr/lib/crt1.o
   HOSTING_LIBS="$HOSTING_LIBS"' /usr/lib/crtn.o'

Re: [PATCH] Re: ld testsuite breaks with latest GCC-4.0

[Nick Clifton]

Hi Greg,

> Here is a new patch that tested out fine for me with gcc-4 and gcc-3.4.3.
> Nick approved the previous patch, so if this one is also approved, could
> someone please apply it as I don't have commit priv's.

Done.  I used this, updated, ChangeLog entry:

ld/ChangeLog
2005-01-10  Greg Schaffer  <gschafer@zip.com.au>

	* configure.host (linux targets): Use "${CC} -dumpspecs" instead
	of "${CC} --print-file-name=specs" to appease GCC versions >= 4.0.
	(ia64-*-aix): Delete - this target is obsolete, and would be
	broken by this patch.

Cheers
   Nick