GCC testsuite differences with and without gas on IRIX 6

GCC testsuite differences with and without gas on IRIX 6

[Rainer Orth]

I've just completed two bootstraps of mainline GCC on IRIX 6.5.20 (N32 ABI
only) with the native as and gas 2.14.90 with this patch:

	http://sources.redhat.com/ml/binutils/2003-07/msg00415.html

The only new testsuite failures with gas are in libjava:

+FAIL: PR295 compilation from source

This test (and a few others) fail due to the following gas warning:

/var/tmp//cc80GL5k.s:19: Warning: .space repeat count is zero, ignored

for this code:

        .align  3
        .space  0
        .section        .rodata

The warning is from from gas/read.c (s_space), where it is only emitted if
!flag_mri.

The question is how best to fix this:

* The native as doesn't complain, so one may just remove the gas warning.

* One could ignore the warning via dejagnu's target.exp, but that doesn't
  seem friendly to a normal user seeing that warning when compiling his
  code with GCJ.

* One could change GCC not to emit this useless .space directive in the
  first place?

What are people's preferences here?

	Rainer

-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University

Re: GCC testsuite differences with and without gas on IRIX 6

[Joe Buck]

On Thu, Jul 24, 2003 at 09:06:24PM +0200, Rainer Orth wrote:
> I've just completed two bootstraps of mainline GCC on IRIX 6.5.20 (N32 ABI
> only) with the native as and gas 2.14.90 with this patch:
> 
> 	http://sources.redhat.com/ml/binutils/2003-07/msg00415.html
> 
> The only new testsuite failures with gas are in libjava:
> 
> +FAIL: PR295 compilation from source
> 
> This test (and a few others) fail due to the following gas warning:
> 
> /var/tmp//cc80GL5k.s:19: Warning: .space repeat count is zero, ignored
> 
> for this code:
> 
>         .align  3
>         .space  0
>         .section        .rodata
> 
> The warning is from from gas/read.c (s_space), where it is only emitted if
> !flag_mri.
> 
> The question is how best to fix this:
> 
> * The native as doesn't complain, so one may just remove the gas warning.
> 
> * One could ignore the warning via dejagnu's target.exp, but that doesn't
>   seem friendly to a normal user seeing that warning when compiling his
>   code with GCJ.
> 
> * One could change GCC not to emit this useless .space directive in the
>   first place?
> 
> What are people's preferences here?

It would seem that the simplest solution would be to make gas shut up.
When seeing ".space 0", it could just silently create zero space as it
is told.

Re: GCC testsuite differences with and without gas on IRIX 6

[Eric Christopher]

> > What are people's preferences here?
> 
> It would seem that the simplest solution would be to make gas shut up.
> When seeing ".space 0", it could just silently create zero space as it
> is told.

Agreed. I'm curious as to why the warning is in there, but I can't see a
reason for it myself.

-eric

-- 
Eric Christopher <echristo@redhat.com>

Re: GCC testsuite differences with and without gas on IRIX 6

[Thiemo Seufer]

Eric Christopher wrote:
> 
> > > What are people's preferences here?
> > 
> > It would seem that the simplest solution would be to make gas shut up.
> > When seeing ".space 0", it could just silently create zero space as it
> > is told.
> 
> Agreed. I'm curious as to why the warning is in there, but I can't see a
> reason for it myself.

Hm. Is .space supposed to impose some alignment on the following code?


Thiemo

Re: GCC testsuite differences with and without gas on IRIX 6

[Eric Christopher]

On Thu, 2003-07-24 at 13:26, Thiemo Seufer wrote:
> Eric Christopher wrote:
> > 
> > > > What are people's preferences here?
> > > 
> > > It would seem that the simplest solution would be to make gas shut up.
> > > When seeing ".space 0", it could just silently create zero space as it
> > > is told.
> > 
> > Agreed. I'm curious as to why the warning is in there, but I can't see a
> > reason for it myself.
> 
> Hm. Is .space supposed to impose some alignment on the following code?

In the docs I have it doesn't say so... 'course this is the K&H book
which has quite a few typos and omissions.

-eric

-- 
Eric Christopher <echristo@redhat.com>

Re: GCC testsuite differences with and without gas on IRIX 6

[Rainer Orth]

Eric Christopher writes:

> > Hm. Is .space supposed to impose some alignment on the following code?
> 
> In the docs I have it doesn't say so... 'course this is the K&H book
> which has quite a few typos and omissions.

Neither does as.texinfo, Assembler Directives, .space size, fill.

So since nobody seems to see a reason why gas should warn about .space 0,
what about the following patch then?

	Rainer

-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University


Mon Jul 28 22:12:55 2003  Rainer Orth  <ro@TechFak.Uni-Bielefeld.DE>

	* read.c (s_space): Don't warn about .space 0.

Index: gas/read.c
===================================================================
RCS file: /cvs/src/src/gas/read.c,v
retrieving revision 1.66
diff -u -p -r1.66 read.c
--- gas/read.c	22 Jul 2003 15:27:06 -0000	1.66
+++ gas/read.c	28 Jul 2003 20:20:17 -0000
@@ -2970,9 +2970,7 @@ s_space (mult)
 	  bytes = repeat;
 	  if (repeat <= 0)
 	    {
-	      if (!flag_mri)
-		as_warn (_(".space repeat count is zero, ignored"));
-	      else if (repeat < 0)
+	      if (repeat < 0)
 		as_warn (_(".space repeat count is negative, ignored"));
 	      goto getout;
 	    }

Re: GCC testsuite differences with and without gas on IRIX 6

[Eric Christopher]

> So since nobody seems to see a reason why gas should warn about .space 0,
> what about the following patch then?

> 
> Mon Jul 28 22:12:55 2003  Rainer Orth  <ro@TechFak.Uni-Bielefeld.DE>
> 
> 	* read.c (s_space): Don't warn about .space 0.

Good for me.

-eric

-- 
Eric Christopher <echristo@redhat.com>

Re: GCC testsuite differences with and without gas on IRIX 6

[Rainer Orth]

Eric Christopher writes:

> > Mon Jul 28 22:12:55 2003  Rainer Orth  <ro@TechFak.Uni-Bielefeld.DE>
> > 
> > 	* read.c (s_space): Don't warn about .space 0.
> 
> Good for me.

Thanks, installed.

	Rainer

-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University

Re: GCC testsuite differences with and without gas on IRIX 6

[Rainer Orth]

Rainer Orth writes:

> Eric Christopher writes:
> 
> > > Mon Jul 28 22:12:55 2003  Rainer Orth  <ro@TechFak.Uni-Bielefeld.DE>
> > > 
> > > 	* read.c (s_space): Don't warn about .space 0.
> > 
> > Good for me.
> 
> Thanks, installed.

It has just been brought to my attention that this hadn't been an approval,
but just an opinion on the patch.  I had only looked in the toplevel
MAINTAINERS file for a list of global-write-privs binutils maintainers, but
neglected to consult gas/MAINTAINERS or binutils/MAINTAINERS (which
unfortunately aren't listed in the toplevel file, unlike
e.g. gdb/MAINTAINERS).

What should I do about this patch?  Revert, or is this ok nonetheless?

Sorry about this.

	Rainer

-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University

Re: GCC testsuite differences with and without gas on IRIX 6

[Thiemo Seufer]

Rainer Orth wrote:
> Rainer Orth writes:
> 
> > Eric Christopher writes:
> > 
> > > > Mon Jul 28 22:12:55 2003  Rainer Orth  <ro@TechFak.Uni-Bielefeld.DE>
> > > > 
> > > > 	* read.c (s_space): Don't warn about .space 0.
> > > 
> > > Good for me.
> > 
> > Thanks, installed.
> 
> It has just been brought to my attention that this hadn't been an approval,
> but just an opinion on the patch.

Well, I consider it to be an approval.

> I had only looked in the toplevel
> MAINTAINERS file for a list of global-write-privs binutils maintainers, but
> neglected to consult gas/MAINTAINERS or binutils/MAINTAINERS (which
> unfortunately aren't listed in the toplevel file, unlike
> e.g. gdb/MAINTAINERS).
> 
> What should I do about this patch?  Revert, or is this ok nonetheless?

It is ok.


Thiemo

Re: GCC testsuite differences with and without gas on IRIX 6

[Eric Christopher]

On Tue, 2003-07-29 at 05:34, Thiemo Seufer wrote:
> Rainer Orth wrote:
> > Rainer Orth writes:
> > 
> > > Eric Christopher writes:
> > > 
> > > > > Mon Jul 28 22:12:55 2003  Rainer Orth  <ro@TechFak.Uni-Bielefeld.DE>
> > > > > 
> > > > > 	* read.c (s_space): Don't warn about .space 0.
> > > > 
> > > > Good for me.
> > > 
> > > Thanks, installed.
> > 
> > It has just been brought to my attention that this hadn't been an approval,
> > but just an opinion on the patch.
> 
> Well, I consider it to be an approval.

Right. Unfortunately neither Theimo or I can approve these. However, I
figure someone would have spoken up by now if they hated it so I say we
leave it :)

-eric

-- 
Eric Christopher <echristo@redhat.com>

Re: GCC testsuite differences with and without gas on IRIX 6

[Rainer Orth]

Eric Christopher writes:

> Right. Unfortunately neither Theimo or I can approve these. However, I
> figure someone would have spoken up by now if they hated it so I say we
> leave it :)

Except for H.J. Lu's report that the patch may have broken two testcases on
alpha-linux and hppa-linux.  I've already replied to that; let's see where
we get with this.

	Rainer

Re: GCC testsuite differences with and without gas on IRIX 6

[Eric Christopher]

On Tue, 2003-07-29 at 13:02, Rainer Orth wrote:
> Eric Christopher writes:
> 
> > Right. Unfortunately neither Theimo or I can approve these. However, I
> > figure someone would have spoken up by now if they hated it so I say we
> > leave it :)
> 
> Except for H.J. Lu's report that the patch may have broken two testcases on
> alpha-linux and hppa-linux.  I've already replied to that; let's see where
> we get with this.

Yup, I agree.

-eric

-- 
Eric Christopher <echristo@redhat.com>

Re: GCC testsuite differences with and without gas on IRIX 6

[Richard Henderson]

On Mon, Jul 28, 2003 at 10:24:17PM +0200, Rainer Orth wrote:
> So since nobody seems to see a reason why gas should warn about .space 0,
> what about the following patch then?

Well, the thing is, this is almost certainly a bug in gcc that
it is being emitted in the first place.  I've seen this before
and that has *always* been the case.

I'd prefer that we continue to warn.


r~

Re: GCC testsuite differences with and without gas on IRIX 6

[Rainer Orth]

Richard Henderson writes:

> Well, the thing is, this is almost certainly a bug in gcc that
> it is being emitted in the first place.  I've seen this before
> and that has *always* been the case.
> 
> I'd prefer that we continue to warn.

Ok, given that the patch seems to have cause a gas testsuite regression on
HP-PA, I've reverted the patch.  I'll investigate the root cause when I get
to this.

	Rainer

-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University

Re: GCC testsuite differences with and without gas on IRIX 6

[Eric Christopher]

On Wed, 2003-07-30 at 09:59, Richard Henderson wrote:
> On Mon, Jul 28, 2003 at 10:24:17PM +0200, Rainer Orth wrote:
> > So since nobody seems to see a reason why gas should warn about .space 0,
> > what about the following patch then?
> 
> Well, the thing is, this is almost certainly a bug in gcc that
> it is being emitted in the first place.  I've seen this before
> and that has *always* been the case.
> 
> I'd prefer that we continue to warn.

Okie. Thanks :)

-eric

-- 
Eric Christopher <echristo@redhat.com>