make check failure on x86-64 with abilist

make check failure on x86-64 with abilist

[Andreas Jaeger]

Roland,

I get this failure during make check on Linux/x86-64 now:
--- -   2003-03-19 10:15:18.000000000 +0100
+++ /usr/src/aj/build/libc/main/libc.symlist    2003-03-19 10:15:18.000000000 +0100
@@ -427 +427 @@ GLIBC_2.2.5
- _nl_default_dirname D 0x12
+ _nl_default_dirname D 0x26
make[2]: *** [check-abi-libc] Error 1

What's the best way to fix this?  Will update-abi handle this?

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj

Re: make check failure on x86-64 with abilist

[Jakub Jelinek]

On Wed, Mar 19, 2003 at 10:26:30AM +0100, Andreas Jaeger wrote:
> 
> Roland,
> 
> I get this failure during make check on Linux/x86-64 now:
> --- -   2003-03-19 10:15:18.000000000 +0100
> +++ /usr/src/aj/build/libc/main/libc.symlist    2003-03-19 10:15:18.000000000 +0100
> @@ -427 +427 @@ GLIBC_2.2.5
> - _nl_default_dirname D 0x12
> + _nl_default_dirname D 0x26
> make[2]: *** [check-abi-libc] Error 1
> 
> What's the best way to fix this?  Will update-abi handle this?

First of all, it is unfortunate a string as char array is exported.
If a binary accesses _nl_default_dirname, it will have a
copy reloc for exactly this size.
0x12 is sizeof "/usr/share/locale" which is what the ABI should specify.

	Jakub

Re: make check failure on x86-64 with abilist

[Andreas Jaeger]

Jakub Jelinek <jakub@redhat.com> writes:

> On Wed, Mar 19, 2003 at 10:26:30AM +0100, Andreas Jaeger wrote:
>> 
>> Roland,
>> 
>> I get this failure during make check on Linux/x86-64 now:
>> --- -   2003-03-19 10:15:18.000000000 +0100
>> +++ /usr/src/aj/build/libc/main/libc.symlist    2003-03-19 10:15:18.000000000 +0100
>> @@ -427 +427 @@ GLIBC_2.2.5
>> - _nl_default_dirname D 0x12
>> + _nl_default_dirname D 0x26
>> make[2]: *** [check-abi-libc] Error 1
>> 
>> What's the best way to fix this?  Will update-abi handle this?

update-abi handles this fine.

> First of all, it is unfortunate a string as char array is exported.
> If a binary accesses _nl_default_dirname, it will have a
> copy reloc for exactly this size.
> 0x12 is sizeof "/usr/share/locale" which is what the ABI should specify.

I see - and I build with a different configure line.  So, this error
will always appear if something different from --prefix=/usr is used
:-(.  Can we avoid this somehow?

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj

Re: make check failure on x86-64 with abilist

[Roland McGrath]

> I see - and I build with a different configure line.  So, this error
> will always appear if something different from --prefix=/usr is used
> :-(.  

Yes.

> Can we avoid this somehow?

We never should have exported a char array like that in the first place. :-(
I wonder why that is a public symbol at all.  _nl_default_dirname looks
internal to me.

Re: make check failure on x86-64 with abilist

[Andreas Jaeger]

Roland McGrath <roland@redhat.com> writes:

>> I see - and I build with a different configure line.  So, this error
>> will always appear if something different from --prefix=/usr is used
>> :-(.  
>
> Yes.
>
>> Can we avoid this somehow?
>
> We never should have exported a char array like that in the first place. :-(
> I wonder why that is a public symbol at all.  _nl_default_dirname looks
> internal to me.

So, should we make it internal now?  I don't see where it's used and
it's not exported in /usr/include.

Btw. make check will always abort if the abi check fails, it's not
possible to rerun make check and continue - like it's done with all
other tests that create a .out file,

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj

Re: make check failure on x86-64 with abilist

[Roland McGrath]

> So, should we make it internal now?  I don't see where it's used and
> it's not exported in /usr/include.

Sounds fine to me.

> Btw. make check will always abort if the abi check fails, it's not
> possible to rerun make check and continue - like it's done with all
> other tests that create a .out file,

That means you didn't lose anything.  make check will always update all
.out files, and then do check-abi (show you the diff).

Re: make check failure on x86-64 with abilist

[Andreas Jaeger]

Roland McGrath <roland@redhat.com> writes:

>> So, should we make it internal now?  I don't see where it's used and
>> it's not exported in /usr/include.
>
> Sounds fine to me.

Patch is appended.  Ok to commit?

>> Btw. make check will always abort if the abi check fails, it's not
>> possible to rerun make check and continue - like it's done with all
>> other tests that create a .out file,
>
> That means you didn't lose anything.  make check will always update all
> .out files, and then do check-abi (show you the diff).

Excellent!

Andreas

2003-03-19  Andreas Jaeger  <aj@suse.de>

	* intl/Versions: Make _nl_msg_cat_cntr,  _nl_default_dirname and
	_nl_domain_bindings as private.

============================================================
Index: intl/Versions
--- intl/Versions	1 May 2002 16:42:08 -0000	1.4
+++ intl/Versions	19 Mar 2003 11:01:08 -0000
@@ -1,8 +1,5 @@
 libc {
   GLIBC_2.0 {
-    # global variables
-    _nl_msg_cat_cntr; _nl_default_dirname; _nl_domain_bindings;
-
     # functions used in inline functions or macros
     __dcgettext;
 
@@ -27,5 +24,9 @@ libc {
 
     # n*
     ngettext;
+  }
+  GLIBC_PRIVATE {
+    # global variables
+    _nl_msg_cat_cntr; _nl_default_dirname; _nl_domain_bindings;
   }
 }

-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj

Re: make check failure on x86-64 with abilist

[Roland McGrath]

> Patch is appended.  Ok to commit?

We'll wait for Ulrich and Jakub to comment.  

> > That means you didn't lose anything.  make check will always update all
> > .out files, and then do check-abi (show you the diff).
> 
> Excellent!

Actually if you use -j but not -k, check-abi could fail first and make it bail.
But it's unlikely.

Re: make check failure on x86-64 with abilist

[Andreas Schwab]

Andreas Jaeger <aj@suse.de> writes:

|> 2003-03-19  Andreas Jaeger  <aj@suse.de>
|> 
|> 	* intl/Versions: Make _nl_msg_cat_cntr,  _nl_default_dirname and
|> 	_nl_domain_bindings as private.

I don't think you can make _nl_msg_cat_cntr private, it was once part of
the libintl ABI (see revision 1.12 of intl/libintl.h).

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: make check failure on x86-64 with abilist

[Andreas Jaeger]

Andreas Schwab <schwab@suse.de> writes:

> Andreas Jaeger <aj@suse.de> writes:
>
> |> 2003-03-19  Andreas Jaeger  <aj@suse.de>
> |> 
> |> 	* intl/Versions: Make _nl_msg_cat_cntr,  _nl_default_dirname and
> |> 	_nl_domain_bindings as private.
>
> I don't think you can make _nl_msg_cat_cntr private, it was once part of
> the libintl ABI (see revision 1.12 of intl/libintl.h).

Can we make the other two private?

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj

Re: make check failure on x86-64 with abilist

[Andreas Schwab]

Andreas Jaeger <aj@suse.de> writes:

|> Andreas Schwab <schwab@suse.de> writes:
|> 
|> > Andreas Jaeger <aj@suse.de> writes:
|> >
|> > |> 2003-03-19  Andreas Jaeger  <aj@suse.de>
|> > |> 
|> > |> 	* intl/Versions: Make _nl_msg_cat_cntr,  _nl_default_dirname and
|> > |> 	_nl_domain_bindings as private.
|> >
|> > I don't think you can make _nl_msg_cat_cntr private, it was once part of
|> > the libintl ABI (see revision 1.12 of intl/libintl.h).
|> 
|> Can we make the other two private?

Yes, I think we can.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: make check failure on x86-64 with abilist

[Ulrich Drepper]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Roland McGrath wrote:
>>So, should we make it internal now?  I don't see where it's used and
>>it's not exported in /usr/include.
> 
> 
> Sounds fine to me.

It's used in (old?) gettext code if I remember correctly.

- -- 
- --------------.                        ,-.            444 Castro Street
Ulrich Drepper \    ,-----------------'   \ Mountain View, CA 94041 USA
Red Hat         `--' drepper at redhat.com `---------------------------
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (GNU/Linux)

iD8DBQE+eKe62ijCOnn/RHQRAmBFAKDJivIMd3EaJd6IJvScptHQEe9YfgCfddKy
tTKFQT+7QGKmMsFoknD3oTI=
=Zu6g
-----END PGP SIGNATURE-----

Re: make check failure on x86-64 with abilist

[Andreas Jaeger]

Andreas Schwab <schwab@suse.de> writes:

> Andreas Jaeger <aj@suse.de> writes:
>
> |> Andreas Schwab <schwab@suse.de> writes:
> |> 
> |> > Andreas Jaeger <aj@suse.de> writes:
> |> >
> |> > |> 2003-03-19  Andreas Jaeger  <aj@suse.de>
> |> > |> 
> |> > |> 	* intl/Versions: Make _nl_msg_cat_cntr,  _nl_default_dirname and
> |> > |> 	_nl_domain_bindings as private.
> |> >
> |> > I don't think you can make _nl_msg_cat_cntr private, it was once part of
> |> > the libintl ABI (see revision 1.12 of intl/libintl.h).
> |> 
> |> Can we make the other two private?
>
> Yes, I think we can.

Ok, is the obvious change then ok to commit?

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj

Re: make check failure on x86-64 with abilist

[Andreas Jaeger]

Ulrich Drepper <drepper@redhat.com> writes:

> Roland McGrath wrote:
>>>So, should we make it internal now?  I don't see where it's used and
>>>it's not exported in /usr/include.
>> 
>> 
>> Sounds fine to me.
>
> It's used in (old?) gettext code if I remember correctly.

So, what can be done?

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj

Re: make check failure on x86-64 with abilist

[Ulrich Drepper]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Andreas Jaeger wrote:
> Ok, is the obvious change then ok to commit?

Not so fast, we'll have to check some old code first.

- -- 
- --------------.                        ,-.            444 Castro Street
Ulrich Drepper \    ,-----------------'   \ Mountain View, CA 94041 USA
Red Hat         `--' drepper at redhat.com `---------------------------
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (GNU/Linux)

iD8DBQE+eKf/2ijCOnn/RHQRAqatAKC8ta6iZIx3KDYKyN+izMEsUinprACePUFi
4dMikCpuO8SB/OPRkKV8p94=
=+BoT
-----END PGP SIGNATURE-----

Re: make check failure on x86-64 with abilist

[Jakub Jelinek]

On Wed, Mar 19, 2003 at 09:25:19AM -0800, Ulrich Drepper wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Andreas Jaeger wrote:
> > Ok, is the obvious change then ok to commit?
> 
> Not so fast, we'll have to check some old code first.

E.g. /usr/bin/gettext from RHL5.2 has:
  0804ce78  01b05 R_386_COPY            0804ce78  _nl_domain_bindings
while in 6.2 it already does not.
I have seen _nl_msg_cat_cntr and _nl_default_dirname in all
gettext versions I looked at as != SHN_UND symbols though.

	Jakub

Re: make check failure on x86-64 with abilist

[Andreas Jaeger]

Jakub Jelinek <jakub@redhat.com> writes:

> On Wed, Mar 19, 2003 at 09:25:19AM -0800, Ulrich Drepper wrote:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>> 
>> Andreas Jaeger wrote:
>> > Ok, is the obvious change then ok to commit?
>> 
>> Not so fast, we'll have to check some old code first.
>
> E.g. /usr/bin/gettext from RHL5.2 has:
>   0804ce78  01b05 R_386_COPY            0804ce78  _nl_domain_bindings
> while in 6.2 it already does not.
> I have seen _nl_msg_cat_cntr and _nl_default_dirname in all
> gettext versions I looked at as != SHN_UND symbols though.

So, we cannot get rid of it - but how do we fix the abi-check failure?

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj