This is the mail archive of the binutils@sourceware.org mailing list for the binutils project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Your gas changes caused weakref failures for cris-elf,supposedly others too

>>> Alan Modra <amodra@bigpond.net.au> 04.11.05 02:57:47 >>>
>On Mon, Oct 31, 2005 at 10:15:49AM +0100, Jan Beulich wrote:
>> >While you two discuss where and how to fix this, I've entered
>[snip]
>> 
>> Thanks. And yes, I suppose the failure exists everywhere. And no,
>> neither of the two of use could really see the problem while
testing
>> since the submissions and check-ins were too close to one another
in
>> time.
>
>gas/read.c
>
>revision 1.111
>date: 2005/10/27 07:40:07;  author: jbeulich;  state: Exp;  lines: +1
-1
>
>revision 1.110
>date: 2005/10/24 17:51:41;  author: aoliva;  state: Exp;  lines: +119
-0
>
>62 hours is hardly "too close".  There seems to be no progress on
>resolving these annoying testsuite failures on all targets, even
though
>the fix is fairly simple.  I will revert your last patch if something
is
>not done soon.

This you just look at the commit time, not the submission time. If I
hadn't checked in that patch, another (x86) test would have failed as
was realized and reported
(http://sourceware.org/ml/binutils/2005-10/msg00393.html), since the
OK for a dependent x86 patch was given before the OK for the general
patch (and I'm sorry for having forgotten about the dependency). Also,
from my perspective, the weakref patch was already broken when it got
checked in, the broken-ness was just hidden (it silently assumed symbols
can be redefined without making sure that really worked, much like other
parts of the assembler before the adjustment
[http://sourceware.org/ml/binutils/2005-09/msg00377.html]). Finally,
Alexandre promised to address the problems (attached, unfortunately not
copied to the list), which is why I didn't spend any time on addressing
these (not to mention that I didn't really have time to do so).
Meanwhile I suggest the following (probably I could even consider this
obvious and don't wait for approval...):

gas/testsuite/
2005-11-04  Jan Beulich  <jbeulich@novell.com>

	* gas/all/gas.exp: xfail weakref dump tests for all targets.

---
/home/jbeulich/src/binutils/mainline/2005-11-04/gas/testsuite/gas/all/gas.exp	2005-11-04
08:29:48.000000000 +0100
+++ 2005-11-04/gas/testsuite/gas/all/gas.exp	2005-11-04
08:49:50.000000000 +0100
@@ -256,15 +256,15 @@ if { ![istarget "i960-*-*"] } {
 case $target_triplet in {
     { z80-*-* } { }
     default {
-	setup_xfail cris-*-* mmix-*-*
+	setup_xfail *-*-*
 	run_dump_test weakref1
-	setup_xfail cris-*-* mmix-*-*
+	setup_xfail *-*-*
 	run_dump_test weakref1g
-	setup_xfail cris-*-* mmix-*-*
+	setup_xfail *-*-*
 	run_dump_test weakref1l
-	setup_xfail cris-*-* mmix-*-*
+	setup_xfail *-*-*
 	run_dump_test weakref1u
-	setup_xfail cris-*-* mmix-*-*
+	setup_xfail *-*-*
 	run_dump_test weakref1w
     }
 }

Attachment: binutils-mainline-weakref-xfail.patch
Description: Binary data

--- Begin Message --- 
On Oct 28, 2005, "Jan Beulich" <JBeulich@novell.com> wrote:

> Looking at the code it doesn't seem so.

Maybe it diverted a lot, or I changed my mind during the
implementation, or I just misremember :-( Sorry.  I'll get to it ASAP.

> If I was to
> do anything with it, I'd just remove the entire r# section.

If redefinitions are not going to be supported (which is exactly what
r# was testing), that sounds like a plan :-)

> But as said
> before, following the description in as.texinfo I can't see why none of
> the nr* symbols appear in the symbol table.

If the weakref isn't ultimately referenced, or if it was redirected to
something else or redefined, then the symbol is not necessary.

> And in addition to that I also wonder how the test completes on some
> targets that have special .set semantics

No idea, I didn't know about that.

-- 
Alexandre Oliva         http://www.lsd.ic.unicamp.br/~oliva/
Red Hat Compiler Engineer   aoliva@{redhat.com, gcc.gnu.org}
Free Software Evangelist  oliva@{lsd.ic.unicamp.br, gnu.org}

--- End Message ---
Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]