Eric Blake ebb9 at byu.net writes:
So if I can pinpoint
the failure to a cygwin bug and get it fixed before the formal release of
1.7,
then gnulib does not have to worry about working around anything.
Update: the bug was indeed in cygwin, and is now fixed - POSIX requires
implementations
file with locale-specific quotes?
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http
has to be an interface change to gnulib's fts to make it
obvious when a directory is changing, in order to make it easier for find
to group multiple files found from a single directory.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP
variable being used as your
followup message)? Without using the cache, you are paying the cost of a
link test on every subsequent configure run?
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin
.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
not comply with POSIX. For example, Solaris
/bin/sh. This portability pitfall is documented in the autoconf manual.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key
work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
to this, see
m4's bootstrap script, which sees if git-merge-changelog is available in
the path, and if so, adds the git changelog merge driver if it is not
already configured, and if not, suggests that the user look into using it.
- --
Don't work too hard, make some time for fun as well!
Eric Blake
too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
that does not append your employer's
legalese.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla
in the user's pre-seed of the cache:
if test x$ac_cv_func_utime_null != xno; then
--
Eric Blake
constants.
Anyone object to me pushing this?
Make fcntl module detect O_NOATIME, O_NOFOLLOW on GNU/Linux.
Looks correct to me.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin
to provide both the rsync approach while keeping a
fallback using wget, for scenarios where rsync is blocked by firewalls but
http can still get through?
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9
?
--
Eric Blake
into git submodules?
From: Eric Blake e...@byu.net
Date: Fri, 24 Apr 2009 10:54:03 -0600
Subject: [PATCH] maint.mk: import improvements from m4
* top/maint.mk (VC-tag): Use signing key from cfg.mk.
(move_if_change): Delete unused macro.
(news-date-check, vc-diff-check): Support VPATH builds
if those symlinks point into git submodules?
Yes, it sounds like that would be worthwhile -- at least to you.
How about this patch?
From: Eric Blake e...@byu.net
Date: Fri, 24 Apr 2009 14:58:02 -0600
Subject: [PATCH] vc-list-files: ignore git symlinks
* build-aux/vc-list-files (.git): Use ls
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Jim Meyering on 4/24/2009 3:22 PM:
Yes, it sounds like that would be worthwhile -- at least to you.
How about this patch?
From: Eric Blake e...@byu.net
Date: Fri, 24 Apr 2009 14:58:02 -0600
Subject: [PATCH] vc-list-files: ignore git
as separate
arguments to sed. I think it should be:
eval exec git ls-tree -r 'HEAD:$dir' \
\| sed -n 's!^100[^ ]*.!$dir!p' $postprocess
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Eric Blake on 4/25/2009 6:05 AM:
+ eval exec git ls-tree -r 'HEAD:$dir' \
+\| sed -n 's!^100[^]*.!'$dir'!p' $postprocess
Is that the right quoting? Remember, this line is going through an eval,
and we don't want a $dir
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Eric Blake on 4/25/2009 7:14 AM:
I convinced myself. If $dir contains $, then you have the wrong quoting.
So I'm applying the above patch.
Except that when I committed, the timestamp went backwards! We need to be
more consistent
, and m4 currently do), while still allowing a cfg.mk override
(the way autoconf currently does).
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake
on useless-if-before-free for projects that choose to
skip that particular syntax check, provided we also rewrite that check to
first validate that build-aux/useless-if-before-free has been imported.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
iEYEARECAAYFAkn1rwgACgkQ84KuGfSFAYChFgCdFDVX5VdU8nkL+v3qwEArd3CJ
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Jim Meyering on 4/26/2009 11:41 PM:
Eric Blake wrote:
Right now, maint.mk is hard-coded to check for config.h as the first
include of each .c file. But what about m4's situation, where every file
includes a common header (m4.h
? On the other hand, do we really care that such old
systems can't create sparse files?
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
correctly, mbstate_t can be
a numeric type rather than a struct.
But this declaration is at global scope. Why use an initializer at all, when
this should work regardless of the underlying type?
mbstate_t _gl_mbsrtowcs_state;
--
Eric Blake
.
Then whatever we end up using, we need to add a comment saying why we aren't
relying on bss initialization.
--
Eric Blake
for another?
--
Eric Blake
are preserved.
| +notice and this notice are preserved. This file is offered as-is,
| +without warranty of any kind.
| @end smallexample
Should we go ahead and modify all of gnulib's m4/*.m4 scripts to add that
sentence?
--
Eric Blake
.
Maybe some improvements can be made in that area.
Also a question - does your compiler support _Bool, even though it lacks
stdbool.h?
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment
? $(srcdir)/.git
+ diff=$$(git submodule -q foreach git diff-index --name-only HEAD) \
diff=$$(cd $(srcdir) git submodule...
--
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
signature.asc
Description: OpenPGP digital signature
Looks good to me.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
this?
From: Eric Blake e...@byu.net
Date: Sun, 3 May 2009 15:39:00 +0200
Subject: [PATCH] maint: make the new no-submodule-changes rule VPATH-safe
* top/maint.mk (no-submodule-changes): Don't assume a srcdir build.
Thanks for writing the patch up for me.
- --
Don't work too hard, make
it might
be specific to the fact that I built on cygwin; it is just a re-ordering of
arguments, but any hints on how to debug before I spend more time looking into
it would be appreciated.
From: Eric Blake e...@byu.net
Date: Fri, 8 May 2009 08:59:46 -0600
Subject: [PATCH] argp: avoid undefined
an additional gl_list RB-tree implementation that uses
threading, such that it would be easy to for clients to experiment between
implementations and determine whether they care more about memory and insertion
speed, vs. better iteration speed?
--
Eric Blake
the effort.
--
Eric Blake
, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
(the AIX workaround guesses wrong for char and short) for another
(alignof() must not be used to initialize an enumerator, but can be used
in all other contexts). Shouldn't this at least be documented in the
alignof.h header?
- --
Don't work too hard, make some time for fun as well!
Eric Blake
) 'long long int', particularly since
these types often have smaller alignment than size?
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake
':
../../lib/memchr2.c:57: warning: return makes pointer from integer without a
cast
An instance of missing memchr, also on gnulib's obsolete list.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9
as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
iEYEARECAAYFAkoKtQIACgkQ84KuGfSFAYCNZwCfT4nehMaz/5KP/YpyXEQ6yemh
of
gl_USE_SYSTEM_EXTENSIONS; should we revisit those?
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG
calls, are locale-agnostic.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http
math.h instead of the intended system math.h.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla
. It also risks
issues with CDPATH in $(gnulib-version) (in a non-VPATH build, cd
./gnulib is silent, but cd gnulib can be vocal or even do the wrong
thing). I'm pushing this followup.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP
, not just NULL).
Should we be adding that to the list of pre-defined macros in
gl_USE_SYSTEM_EXTENSIONS (and push it to autoconf as well)?
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment
collecting multiple unquoted m4 arguments into a single
string).
What do you think of this patch?
From: Eric Blake e...@byu.net
Date: Wed, 20 May 2009 14:45:07 -0600
Subject: [PATCH] Another try at making math.h work for AIX xlc.
* m4/include_next.m4: Avoid redundant m4_quote.
(gl_CHECK_NEXT_HEADERS
Eric Blake ebb9 at byu.net writes:
What do you think of this patch?
From: Eric Blake ebb9 at byu.net
Date: Wed, 20 May 2009 14:45:07 -0600
Subject: [PATCH] Another try at making math.h work for AIX xlc.
* m4/include_next.m4: Avoid redundant m4_quote.
(gl_CHECK_NEXT_HEADERS): Also define
macros for mode_t values.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http
for AC_CHECK_SIZEOF([mode_t]), then
use the preprocessor to skip the unneeded branch altogether using the
appropriate macro from config.h.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key
,
if required by the configure.ac.
Would it be possible to change gnulib-tool to generate the appropriate
$(AM_V_GEN) output itself, rather than making bootstrap do a post-process
pass? After all, not all projects use the bootstrap script.
- --
Don't work too hard, make some time for fun as well!
Eric
words, we aren't removing a level of
m4 quoting.
--
Eric Blake
this
comment, to explain things:
+ dnl Don't redefine SIZE_MAX in config.h if config.h is re-included after
+ dnl stdint.h. Remember that the #undef in AH_VERBATIM gets replaced with
+ dnl #define by AC_DEFINE_UNQUOTED.
The patch is correct as-is.
--
Eric Blake
as possible (m4_dquote, on the other hand, is very
important).
$ m4 -Ilib m4sugar/m4sugar.m4 -
m4_divert
m4_define([a], [A, B])
a
A, B
m4_quote(a)
A,B
m4_expand([a])
A, B
m4_len(a)
m4:stdin:6: warning: m4_len: extra arguments ignored: 2 1
1
m4_len(m4_quote(a))
3
m4_len(m4_expand([a]))
4
--
Eric Blake
to beef up our strstr.m4 checks to catch
platform bugs, instead of the current assumption that if you don't care
about quadratic behavior, that the platform's strstr is always sufficient?
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP
in
gnulib's strstr replacement?
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http
*input = strdup (aBaaax);
Did you take the proper precautions to ensure strdup is available? I'd
almost feel better if you rewrote this to use malloc/strcpy, rather than
dragging in dependencies just for the test.
- --
Don't work too hard, make some time for fun as well!
Eric Blake
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Eric Blake on 5/26/2009 6:56 AM:
Yes, adding to the test is a good first step, and we can use the feedback
to decide how much else we might need to do. Is the bug also present in
gnulib's strstr replacement?
To answer my own question
memchr implementations out there; comments are welcome on whether we should
polish and apply this patch.
From: Eric Blake e...@byu.net
Date: Tue, 26 May 2009 10:03:26 -0600
Subject: [PATCH] memchr: detect broken x86_64 and alpha implementations
* modules/memchr-tests (Depends-on): Move mmap
/canonicalize-lgpl.c: dest = strchr (rpath, '\0');
lib/canonicalize.c: dest = strchr (rname, '\0');
lib/glob.c: end_name = strchr (unescape, '\0');
From: Eric Blake e...@byu.net
Date: Tue, 26 May 2009 12:21:17 -0600
Subject: [PATCH] strtok_r: use rawmemchr for speed.
* lib
Bruno Haible bruno at clisp.org writes:
Eric Blake wrote:
http://www.alphalinux.org/archives/axp-list/March2001/0337.shtml
It looks like the bug is alpha-specific in memchr
I don't think it is a bug. memchr could also be implemented by doing
a backwards search and still
, and therefore (correctly)
rejected as a non-POSIX extension when using --posix with newer sed. It
was a bug in earlier sed that \| was not identified as non-portable.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version
$ echo hi | POSIXLY_CORRECT=1 sed/sed v || echo $?
hi
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG
hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Bruno Haible on 5/26/2009 4:05 PM:
Eric Blake wrote:
http://www.alphalinux.org/archives/axp-list/March2001/0337.shtml
It looks like the bug is alpha-specific in memchr
I don't think it is a bug. memchr could also be implemented
-tuned assembly, but hands down faster than a byte at a time.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG
.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
Wouldn't it be better to store this in gnulib.git/.gitattributes, so that
everyone can get this exception rather than just savannah?
--
Eric Blake
/gnulib/doc//ld-output-def.texi:24: Misplaced }.
This appears to be fixed by the attached patch.
Thanks. I added a ChangeLog entry, and pushed it.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Bruno Haible on 5/31/2009 1:51 PM:
Eric Blake wrote:
I noticed you just pushed test-alignof.c. Should this file also check
'long double', and (where available) 'long long int', particularly since
these types often have smaller
time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
should be able to find one. The broken alpha
version may have been around longer, and I'm not sure whether it is fixed
yet, but alpha seems to be a less popular architecture.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Eric Blake on 6/7/2009 9:51 PM:
According to Jim Meyering on 6/6/2009 2:41 PM:
A few of the function declarations in hash.h could benefit from
gcc's warn_unused_result attribute, so I'm adding it:
Including one in hash.c itself
a memory fence for a *printf test of %.*s, so I'll leave that to you
if you'd like to start checking for it.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net
/msg00253.html
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
sticks with abort() (no additional
message) or uses your new proposed macro (a new message in addition to the
bug reporting address triggered by the SIGABRT).
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG
testsuite to accept
either spelling.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http
: Eric Blake e...@byu.net
Date: Mon, 15 Jun 2009 11:04:21 -0600
Subject: [PATCH] memchr: add valgrind exception
* lib/memchr.valgrind: New file.
* modules/memchr (Files): Distribute it.
Signed-off-by: Eric Blake e...@byu.net
---
ChangeLog |4
lib/memchr.valgrind | 14
rehash when shrinking the
table on hash_delete can be ignored?
I'm debating just deleting the USE_OBSTACK code altogether; it's even buggier
at memory management, and probably not worth maintaining.
From: Eric Blake e...@byu.net
Date: Mon, 8 Jun 2009 05:56:37 -0600
Subject: [PATCH] hash: avoid
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Eric Blake on 6/15/2009 11:42 AM:
With the recent clarification on memchr behavior, it is conceivable that we
could trip false positives under valgrind when using memchr to find the
location of a known byte with an overestimated
? The
logs you pointed to were for SASL, which did not use the strstr module.
If test-strstr fails (which I suspect it will), then I need to also modify
strstr.m4 to declare that the system strstr needs replacing if memchr is
broken.
- --
Don't work too hard, make some time for fun as well!
Eric Blake
`install_handlers':
fatal-signal.c:189: error: `SA_NODEFER' undeclared (first use in this
function)
I'm not worried about fixing older m4 versions. Let's focus on getting
the latest to work.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP
in that
category other than the Debian bug report.
All that said, I think I will still commit this patch, unless anyone objects:
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Eric Blake on 6/15/2009 5:17 PM:
Additionally, it looks like hash_rehash has a memory leak - if new_table
is allocated, but we later fail to allocate new_entry, the function
returns immediately without reclaiming new_table. Which
Eric Blake ebb9 at byu.net writes:
Subject: [PATCH] hash: improve memory management
* lib/hash.c (hash_delete): Free entries if resize failed.
(hash_insert): Don't leave entry on table when returning failure.
(hash_rehash): Avoid memory leak, by guaranteeing that we won't
allocate once we
statement after the label? Is this worth a syntax
check in maint.mk?
--
Eric Blake
gnulib changes are still
portable to Interix.
- --
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Bruno Haible on 6/17/2009 2:17 AM:
Eric Blake wrote:
Hmm. I'm now starting to wonder whether I should create a new module
strstr-simple, which fixes broken strstr (in the presence of broken
memchr) but without worrying about
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Bruno Haible on 6/17/2009 2:17 AM:
Eric Blake wrote:
Hmm. I'm now starting to wonder whether I should create a new module
strstr-simple, which fixes broken strstr (in the presence of broken
memchr) but without worrying about
Eric Blake ebb9 at byu.net writes:
I got bored and attempted this. This implementation still passes the m4
testsuite, although there might be corner cases not exercised by m4 (such as
shrinking the table on hash_delete) that I got wrong, so I'm posting it for
review now. In particular
with m4, the very next call to hash_insert would notice the
table was at 100%, and force another resize above 907 buckets.
Thoughts, before I implement something along these lines?
--
Eric Blake
for fun as well!
Eric Blake e...@byu.net
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
iEYEARECAAYFAko5PKkACgkQ84KuGfSFAYBLmQCgrpRV4v3+a73
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Eric Blake on 6/17/2009 12:21 PM:
Before the rehash, about 20% of the buckets were empty. But after changing
the
prime factor, the hasher function scatters better, and _every single bucket_
is
occupied. Which means ALL
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Eric Blake on 6/17/2009 1:02 PM:
I'm now playing with this simple code motion patch, also available at:
http://repo.or.cz/w/gnulib/ericb.git
$ git pull git://repo.or.cz/gnulib/ericb.git master
This appears to work; and I've now
Jim Meyering jim at meyering.net writes:
Eric Blake wrote:
+ /* Guarantee that once we start moving entries into the new table,
+ we will not need any further memory allocations. We only need
+ new allocations if the hasher consolidates multiple buckets from
+ the old
'free_entry_list', rather than 'overflow' as in
my alt implementation). Here's the effective rebase of that patch (although it
will be a couple hours before I can escape the firewall of my current machine
that prevents me from posting it to a public repository).
From: Eric Blake e...@byu.net
Date: Wed, 17
catch. So the simple code motion was a bit too simple; here's the updated
patch (and I've now run it through the same m4 test that found the original
problem):
From: Eric Blake e...@byu.net
Date: Wed, 17 Jun 2009 13:01:41 -0600
Subject: [PATCH] hash: check for resize before insertion
* lib
Jim Meyering jim at meyering.net writes:
Slight improvement.
This adds a key assertion at the end of the loop.
Looks good so far; how about also adding a test of hash_get_entries and
hash_do_for_each?
--
Eric Blake
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Bruno Haible on 6/17/2009 2:37 AM:
Eric Blake wrote:
gnulib can work around missing ESTALE, but we
would need to port errno_h.m4 to detect interix's situation
Done as follows:
With that in place, Jay, would you mind testing
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
According to Eric Blake on 6/17/2009 4:05 PM:
Jim Meyering jim at meyering.net writes:
Slight improvement.
This adds a key assertion at the end of the loop.
Looks good so far; how about also adding a test of hash_get_entries
501 - 600 of 3907 matches
Mail list logo