ABI test baseline status

ABI test baseline status

[Joseph S. Myers]

Looking at the status of check-abi baseline data, since check-abi was 
enabled by default and the data moved into sysdeps, some architectures 
either have no data or have not had it updated since the move.  It might 
of course be the case that the data was accurate with no changes, but 
experience with other architectures suggests this in unlikely.  Though 
checking the data is something architecture maintainers should do after 
the release freeze, that's to detect any ABI breakage over the course of 
the release cycle; setting up the baselines (cross-checked with old 
releases where possible) shouldn't wait for the freeze.

* sh data hasn't been updated and probably needs to be.

* s390 data (both s390-32 and s390-64) hasn't been updated and probably 
needs to be.

* When the x32 port is complete, that will need ABI baseline data.

* Likewise, when Hurd support is properly functional a Hurd ABI baseline 
will be needed.

In ports:

* Ryan, will you be adding ABI baseline data for powerpc-nofpu or shall I 
do so?

* alpha needs data added.

* hppa needs data added.

* tile needs data added (probably three baselines - tilepro, tilegx32, 
tilegx64).  In addition, other changes to tile and linux-generic are 
needed to bring them up to date with libc changes since those ports were 
added.  (If you are maintaining a port, you really need to follow 
glibc-cvs, watch out for libc changes relevant to the port and try to get 
it updated within days unless you're away at the time.  We should also try 
to make clear in patch postings if they are relevant to ports, but 
following glibc-cvs is a good idea as well.)  Specifically, the 
linux-generic bits/typesizes.h (and the alpha one) doesn't seem to have 
been updated for

2012-04-10  H.J. Lu  <hongjiu.lu@intel.com>

        * bits/typesizes.h (__SNSECONDS_T_TYPE): New macro.
        * sysdeps/unix/bsd/bsd4.4/freebsd/bits/typesizes.h
        (__SNSECONDS_T_TYPE): Likewise.
        * sysdeps/unix/sysv/linux/s390/bits/typesizes.h
        (__SNSECONDS_T_TYPE): Likewise.
        * sysdeps/unix/sysv/linux/sparc/bits/typesizes.h
        (__SNSECONDS_T_TYPE): Likewise.

There is still a sysdeps/tile/elf directory although the use of elf/ 
directories was eliminated on 2012-03-27.  tile needs crti.S and crtn.S - 
the old initfini system was removed 2012-02-21.  tile's bits/siginfo.h 
needs updating for the namespace changes on 2012-02-26/2012-02-27 (not 
using the "struct siginfo" name, defining / using pthread_attr_t 
appropriately) - as, it seems, does ia64; likewise tile's pthreadtypes.h.  
(And noticed in the course of seeing just how out of date tile had become 
in a few months: ia64 has a stray __ELF__ conditional in 
sysdeps/ia64/fpu/libm-symbols.h.  alpha and hppa need rather more 
substantial checks of their sysdeps files to get them properly back in 
sync.)

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: ABI test baseline status

[Chris Metcalf]

On 5/13/2012 6:42 AM, Joseph S. Myers wrote:
> * tile needs data added (probably three baselines - tilepro, tilegx32, 
> tilegx64).  In addition, other changes to tile and linux-generic are 
> needed to bring them up to date with libc changes since those ports were 
> added.  (If you are maintaining a port, you really need to follow 
> glibc-cvs, watch out for libc changes relevant to the port and try to get 
> it updated within days unless you're away at the time.  We should also try 
> to make clear in patch postings if they are relevant to ports, but 
> following glibc-cvs is a good idea as well.)  Specifically, the 
> linux-generic bits/typesizes.h (and the alpha one) doesn't seem to have 
> been updated for

I have all the changes queued up in my git tree, plus other optimization,
etc., changes that have been added to our 2.12 branch for our CentOS
6.1-compatible release.  I am just struggling with the abilist stuff over
this weekend and should have the whole patch series out on Monday.

I'll have some further questions to ask about the ulps-generating stuff,
though.  I hacked the examples to work around the missing exception
#defines on tile to generate some plausible ulps data, but as things stand
the presence of all the FE_xxx tests means the tests don't compile.

-- 
Chris Metcalf, Tilera Corp.
http://www.tilera.com

Re: ABI test baseline status

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 1829 bytes --]

On Sunday 13 May 2012 06:42:12 Joseph S. Myers wrote:
> (And noticed in the course of seeing just how out of date tile had become
> in a few months: ia64 has a stray __ELF__ conditional in
> sysdeps/ia64/fpu/libm-symbols.h.

pushed the patch below
-mike

From b9564c38ea4aa7415b472d5d58760930a616ca3d Mon Sep 17 00:00:00 2001
From: Mike Frysinger <vapier@gentoo.org>
Date: Sun, 13 May 2012 15:09:05 -0400
Subject: [PATCH] ia64: delete __ELF__ check in libm-symbols.h

We assume ELF everywhere now, so drop support for non-ELF.

Reported-by: Joseph S. Myers <joseph@codesourcery.com>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
---
 ChangeLog.ia64                  |    4 ++++
 sysdeps/ia64/fpu/libm-symbols.h |    9 ++-------
 2 files changed, 6 insertions(+), 7 deletions(-)

diff --git a/ChangeLog.ia64 b/ChangeLog.ia64
index 11ba030..fd90f34 100644
--- a/ChangeLog.ia64
+++ b/ChangeLog.ia64
@@ -1,3 +1,7 @@
+2012-05-14  Mike Frysinger  <vapier@gentoo.org>
+
+	* sysdeps/ia64/fpu/libm-symbols.h: Delete __ELF__ checks.
+
 2012-05-07  Mike Frysinger  <vapier@gentoo.org>
 
 	* sysdeps/unix/sysv/linux/ia64/sysdep.h: Include errno.h.
diff --git a/sysdeps/ia64/fpu/libm-symbols.h b/sysdeps/ia64/fpu/libm-symbols.h
index 5b5e4b7..81084bd 100644
--- a/sysdeps/ia64/fpu/libm-symbols.h
+++ b/sysdeps/ia64/fpu/libm-symbols.h
@@ -3,13 +3,8 @@
 
 /* Support for compatible assembler handling.  */
 
-#ifdef __ELF__
-# define ASM_SIZE_DIRECTIVE(name) .size name,.-name
-# define ASM_TYPE_DIRECTIVE(name,T) .type name,T
-#else
-# define ASM_SIZE_DIRECTIVE(name)
-# define ASM_TYPE_DIRECTIVE(name,T)
-#endif
+#define ASM_SIZE_DIRECTIVE(name) .size name,.-name
+#define ASM_TYPE_DIRECTIVE(name,T) .type name,T
 
 #define LOCAL_LIBM_ENTRY(name)			\
 	.proc name;				\
-- 
1.7.9.7

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: ABI test baseline status

[Joseph S. Myers]

On Mon, 14 May 2012, Mike Frysinger wrote:

> On Sunday 13 May 2012 06:42:12 Joseph S. Myers wrote:
> > (And noticed in the course of seeing just how out of date tile had become
> > in a few months: ia64 has a stray __ELF__ conditional in
> > sysdeps/ia64/fpu/libm-symbols.h.
> 
> pushed the patch below

Thanks - note there was another ia64 sysdeps issue mentioned in my message 
as well (bits/siginfo.h needs updating for the namespace changes on 
2012-02-26/2012-02-27 (not using the "struct siginfo" name, defining / 
using pthread_attr_t appropriately)).

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: ABI test baseline status

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 724 bytes --]

On Tuesday 15 May 2012 06:27:31 Joseph S. Myers wrote:
> On Mon, 14 May 2012, Mike Frysinger wrote:
> > On Sunday 13 May 2012 06:42:12 Joseph S. Myers wrote:
> > > (And noticed in the course of seeing just how out of date tile had
> > > become in a few months: ia64 has a stray __ELF__ conditional in
> > > sysdeps/ia64/fpu/libm-symbols.h.
> > 
> > pushed the patch below
> 
> Thanks - note there was another ia64 sysdeps issue mentioned in my message
> as well (bits/siginfo.h needs updating for the namespace changes on
> 2012-02-26/2012-02-27 (not using the "struct siginfo" name, defining /
> using pthread_attr_t appropriately)).

this file seem ripe for splitting up like we did with epoll.h ...
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

[PATCH] ia64: update pthread_attr_t handling in bits/siginfo.h

[Mike Frysinger]

This matches the changes made to the common bits/siginfo.h file.

Reported-by: Joseph Myers <joseph@codesourcery.com>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
---
 ChangeLog.ia64                              |    5 +++++
 sysdeps/unix/sysv/linux/ia64/bits/siginfo.h |   10 ++++++++--
 2 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/ChangeLog.ia64 b/ChangeLog.ia64
index 3130214..d9c3b47 100644
--- a/ChangeLog.ia64
+++ b/ChangeLog.ia64
@@ -1,3 +1,8 @@
+2012-05-16  Mike Frysinger  <vapier@gentoo.org>
+
+	* sysdeps/unix/sysv/linux/ia64/bits/siginfo.h: Add forward
+	declaration of pthread_attr_t and use it in sigevent.
+
 2012-05-15  Mike Frysinger  <vapier@gentoo.org>
 
 	* sysdeps/ia64/elf/stackguard-macros.h: Delete unused file.
diff --git a/sysdeps/unix/sysv/linux/ia64/bits/siginfo.h b/sysdeps/unix/sysv/linux/ia64/bits/siginfo.h
index 40f4862..fbb07b5 100644
--- a/sysdeps/unix/sysv/linux/ia64/bits/siginfo.h
+++ b/sysdeps/unix/sysv/linux/ia64/bits/siginfo.h
@@ -1,5 +1,5 @@
 /* siginfo_t, sigevent and constants.  Linux/ia64 version.
-   Copyright (C) 2000-2004, 2009, 2011 Free Software Foundation, Inc.
+   Copyright (C) 2000-2012 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by David Mosberger-Tang <davidm@hpl.hp.com>.
 
@@ -299,6 +299,12 @@ enum
 # define __SIGEV_MAX_SIZE	64
 # define __SIGEV_PAD_SIZE	((__SIGEV_MAX_SIZE / sizeof (int)) - 4)
 
+/* Forward declaration.  */
+# ifndef __have_pthread_attr_t
+typedef union pthread_attr_t pthread_attr_t;
+#  define __have_pthread_attr_t	1
+# endif
+
 typedef struct sigevent
   {
     sigval_t sigev_value;
@@ -316,7 +322,7 @@ typedef struct sigevent
 	struct
 	  {
 	    void (*_function) (sigval_t);	/* Function to start.  */
-	    void *_attribute;			/* Really pthread_attr_t.  */
+	    pthread_attr_t *_attribute;		/* Thread attributes.  */
 	  } _sigev_thread;
       } _sigev_un;
   } sigevent_t;
-- 
1.7.8.6

Re: ABI test baseline status

[Joseph S. Myers]

On Wed, 16 May 2012, Mike Frysinger wrote:

> On Tuesday 15 May 2012 06:27:31 Joseph S. Myers wrote:
> > On Mon, 14 May 2012, Mike Frysinger wrote:
> > > On Sunday 13 May 2012 06:42:12 Joseph S. Myers wrote:
> > > > (And noticed in the course of seeing just how out of date tile had
> > > > become in a few months: ia64 has a stray __ELF__ conditional in
> > > > sysdeps/ia64/fpu/libm-symbols.h.
> > > 
> > > pushed the patch below
> > 
> > Thanks - note there was another ia64 sysdeps issue mentioned in my message
> > as well (bits/siginfo.h needs updating for the namespace changes on
> > 2012-02-26/2012-02-27 (not using the "struct siginfo" name, defining /
> > using pthread_attr_t appropriately)).
> 
> this file seem ripe for splitting up like we did with epoll.h ...

In principle it would be good to share more of it (and bits/fcntl.h) - in 
practice I think it will be harder than for sys/epoll.h to work out 
exactly what the areas of essential variation are between the different 
versions, and so exactly how it should be split up.

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: [PATCH] ia64: update pthread_attr_t handling in bits/siginfo.h

[Joseph S. Myers]

On Wed, 16 May 2012, Mike Frysinger wrote:

> This matches the changes made to the common bits/siginfo.h file.

Are you going to commit this ...?

(IA64 now needs ABI baseline and bits/mman.h updates as well.)

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: [PATCH] ia64: update pthread_attr_t handling in bits/siginfo.h

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 327 bytes --]

On Wednesday 23 May 2012 06:22:22 Joseph S. Myers wrote:
> On Wed, 16 May 2012, Mike Frysinger wrote:
> > This matches the changes made to the common bits/siginfo.h file.
> 
> Are you going to commit this ...?

guess i forgot to push it

> (IA64 now needs ABI baseline and bits/mman.h updates as well.)

yeah
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]