bug#9262: test.c runs afoul of GCC 4.6.1 bug if --enable-gcc-warnings

[Jim Meyering]

Paul Eggert wrote:
> I'm running into a problem with GCC 4.6.1 x86-64 when
> compiling coreutils (latest git version), configured
> with --enable-gcc-warnings.  A bogus warning is generated,
> causing "make" to fail.  The warning is "function might be
> candidate for attribute 'pure' if it is known to return
> normally [-Wsuggest-attribute=pure]", and it points to the
> "advance" function in test.c.  The warning is also generated
> for lbracket.c, which includes test.c.
>
> This is due to a GCC bug that has just been confirmed; see
> .
>
> For now I'm leaving this alone, but I thought I'd give
> others a heads-up.  The coreutils code is pretty
> crufty but on the other hand I dislike modifying code just
> to work around bugs in static checkers.

Thanks for the heads up.
This came up in July:

http://thread.gmane.org/gmane.comp.gnu.coreutils.general/1399

This is why I now build with gcc-4.7.x.
With it, I no longer get that invalid warning.

Ideally, configure would check for the buggy compiler
and disable the offending option when found.

If you're not going to work on that, what do you
think about sending a message to cont...@debbugs.gnu.org
with this in the body?

tags 9262 + notabug
close 9262

Since this isn't a bug in coreutils, and I'm trying to close
issues we're not likely to work on.

bug#9140: [PATCH 2/3] large-inode: New module

[Bruno Haible]

Eric Blake wrote:
> It will probably break any other 
> packages that have picked up the latest gnulib in the last few days.

Indeed. It broke libiconv-1.14. But fortunately it's not a regression.
Large file support in the 'iconv' program would have been an undocumented
improvement. Now it's delayed to libiconv-1.15...

Thanks for having discovered and fixed this, Eric!!

Bruno
-- 
In memoriam Edward Pimental 

bug#9263: mv overwrite message when owner different

[jidanni]

This message,
mv: try to overwrite `/cf/updates/F.cpio.gz', overriding mode 0644 (rw-r--r--)? 
y
does not make it clear that the file owner is different.
mv (GNU coreutils) 8.5

bug#9262: test.c runs afoul of GCC 4.6.1 bug if --enable-gcc-warnings

[Paul Eggert]

I'm running into a problem with GCC 4.6.1 x86-64 when
compiling coreutils (latest git version), configured
with --enable-gcc-warnings.  A bogus warning is generated,
causing "make" to fail.  The warning is "function might be
candidate for attribute 'pure' if it is known to return
normally [-Wsuggest-attribute=pure]", and it points to the
"advance" function in test.c.  The warning is also generated
for lbracket.c, which includes test.c.

This is due to a GCC bug that has just been confirmed; see
.

For now I'm leaving this alone, but I thought I'd give
others a heads-up.  The coreutils code is pretty
crufty but on the other hand I dislike modifying code just
to work around bugs in static checkers.

bug#9140: [PATCH 2/3] large-inode: New module

[Eric Blake]

On 07/22/2011 04:10 PM, Paul Eggert wrote:

On 07/22/11 10:45, Pádraig Brady wrote:

If it works it would be better in AC_SYS_LARGEFILE (as well) I think.


OK, thanks, I pushed this into gnulib.  Assuming it works (and I don't
see why it'd hurt) I'll push it into autoconf.


No problem with me to pushing the AC_SYS_LARGEFILE improvements into 
autoconf.


--
Eric Blake   ebl...@redhat.com+1-801-349-2682
Libvirt virtualization library http://libvirt.org

bug#9140: [PATCH 2/3] large-inode: New module

[Eric Blake]

On 07/22/2011 04:10 PM, Paul Eggert wrote:

On 07/22/11 10:45, Pádraig Brady wrote:

If it works it would be better in AC_SYS_LARGEFILE (as well) I think.


OK, thanks, I pushed this into gnulib.  Assuming it works (and I don't
see why it'd hurt) I'll push it into autoconf.


Ouch.  This commit broke libvirt 0.9.4, because it unintentionally broke 
large-file support altogether.  It will probably break any other 
packages that have picked up the latest gnulib in the last few days.



Pádraig Brady suggested this in .
* MODULES.html.sh: Add largefile, remove large-inode.
* modules/largefile, m4/largefile.m4: New files.
* modules/large-inode, m4/large-inode.m4: Remove.



-
-configure.ac-early:
-AC_REQUIRE([gl_SYS_LARGE_INODE])
-
-configure.ac:
-



+
+configure.ac:
+AC_REQUIRE([AC_SYS_LARGEFILE])
+
+configure.ac:
+


The first configure.ac should have been configure.ac-early; as a result 
of the duplicate listing, gnulib-tool completely omits AC_SYS_LARGEFILE 
from any configure.ac that was using the broken largefile module.  Test 
by looking at './configure --help | grep large' pre- and post-patch. 
I'm pushing this:


diff --git i/ChangeLog w/ChangeLog
index 72e7611..3fd0c47 100644
--- i/ChangeLog
+++ w/ChangeLog
@@ -1,3 +1,8 @@
+2011-08-08  Eric Blake  
+
+   largefile: fix typo that regressed large file support
+   * modules/largefile (configure.ac-early): Fix section name.
+
 2011-08-06  Karl Berry  

* MODULES.html.sh (func_all_files): _Noreturn is no longer
diff --git i/modules/largefile w/modules/largefile
index d4a2e4b..ca10d48 100644
--- i/modules/largefile
+++ w/modules/largefile
@@ -6,7 +6,7 @@ m4/largefile.m4

 Depends-on:

-configure.ac:
+configure.ac-early:
 AC_REQUIRE([AC_SYS_LARGEFILE])

 configure.ac:


--
Eric Blake   ebl...@redhat.com+1-801-349-2682
Libvirt virtualization library http://libvirt.org

bug#9237: [PATCH] tests: get rid of obsolete 'error_' shell function

[Stefano Lattarini]

On Monday 08 August 2011, Jim Meyering wrote:
> Stefano Lattarini wrote:
> > Date: Thu, 4 Aug 2011 20:48:06 +0200
> > Subject: [PATCH] tests: complete the renaming framework_failure -> 
> > framework_failure_
> >
> > * tests/init.cfg (framework_failure): Remove, `framework_failure_'
> > from init.sh should be used instead in the tests.
> > Remove leading obsolete "FIXME" comment.
> > (is_local_dir_, require_strace_, require_membership_in_two_groups_,
> > require_sparse_support_, skip_if_mcstransd_is_running_,
> > mkfifo_or_skip_) Use `framework_failure_', not `framework_failure'.
> > * Many test scripts: Likewise.
> 
> Thanks.
> I've confirmed that the modified command below does
> the same job and requires less manual editing.
> I've modified your commit log to include the command
> and to fix typos, so include the new version below.
>
Thanks!  And BTW, allow me to apologize for the embarassing number of
typos in both the patch and my previous mail.

> Also, I have queued this additional change set:
> 
> From 601c4d9209e506b92ebfbc9d0fec0f679f5fe5d3 Mon Sep 17 00:00:00 2001
> From: Jim Meyering 
> Date: Mon, 8 Aug 2011 08:54:52 +0200
> Subject: [PATCH 2/2] maint: prevent accidental future use of the old shell
>  function name
> 
> This is especially important for an error-handling shell function
> like this that is actually called only rarely.
>
> [SNIP]
>
Good addition.  Should have thought about it myself ...

Regards,
  Stefano

bug#6683: mktemp foo.XXXXXXXXXXX is not sufficiently random

[Paul Eggert]

On 08/07/2011 10:04 AM, Jim Meyering wrote:

> Yes, please do.

OK, thanks, I installed the one-line change as change to the diff.
This is the first time I've updated a diff file in a while (ever?),
so I hope I did it right.  I'm marking the bug done.

>From 8e2767a3f0c279d355f067e53be2c63173959eb1 Mon Sep 17 00:00:00 2001
From: Paul Eggert 
Date: Mon, 8 Aug 2011 00:29:46 -0700
Subject: [PATCH] mktemp: stir in enough entropy (Bug#6683)

* gl/lib/tempname.c.diff (gen_tempname_len):
Use x_suffix_len bytes' worth of entropy, not 8 bytes.
---
 gl/lib/tempname.c.diff |2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/gl/lib/tempname.c.diff b/gl/lib/tempname.c.diff
index fcacf53..3e30c97 100644
--- a/gl/lib/tempname.c.diff
+++ b/gl/lib/tempname.c.diff
@@ -100,7 +100,7 @@ index 2da5afe..562955a 100644
 -  }
 -#endif
 -  value += random_time_bits ^ __getpid ();
-+  rand_src = randint_all_new (NULL, 8);
++  rand_src = randint_all_new (NULL, x_suffix_len);
 +  if (! rand_src)
 +return -1;
 
-- 
1.7.4.4