Re: [PATCH] split: --chunks option
Pádraig Brady wrote: > I got a bit of time for the review last night... > > This was your last interface change for this: > >-b, --bytes=SIZEput SIZE bytes per output file\n\ > + -b, --bytes=/N generate N output files\n\ > + -b, --bytes=K/N print Kth of N chunks of file\n\ >-C, --line-bytes=SIZE put at most SIZE bytes of lines per output file\n\ >-d, --numeric-suffixes use numeric suffixes instead of alphabetic\n\ >-l, --lines=NUMBER put NUMBER lines per output file\n\ > + -l, --lines=/N generate N eol delineated output files\n\ > + -l, --lines=K/N print Kth of N eol delineated chunks\n\ > + -n, --number=N same as --bytes=/N\n\ > + -n, --number=K/Nsame as --bytes=K/N\n\ > + -r, --round-robin=N generate N eol delineated output files using\n\ > +round-robin style distribution.\n\ > + -r. --round-robin=K/N print Kth of N eol delineated chunk as -rN would\n\ > +have generated.\n\ > + -t, --term=CHAR specify CHAR as eol. This will also convert\n\ > +-b to its line delineated equivalent (-C if\n\ > +splitting normally, -l if splitting by\n\ > +chunks). C escape sequences are accepted.\n\ > > Thinking more about it, I think adding 2 modes of operation to the > already slightly complicated -bCl options is too confusing. > Since this is a separate mode of operation; i.e. one would be > specifying a particular number of files for a different reason > than a particular size, it would be better as a separate option. > So I changed -n to operate as follows. This is more general if > we want to add new split methods in future, and also compatible with > the existing BSD -n without needing a redundant option. I like it. > -n N split into N files based on size of input > -n K/N output K of N to stdout s/K/Kth/ removes an ambiguity (and two more, below) > -n l/N split into N files while maintaining lines s/while maintaining/without splitting/ (and below) > -n l/K/N output K of N to stdout while maintaining lines > -n r/N like `l' but use round robin distribution instead of size > -n r/K/N likewise but only output K of N to stdout > > Other changes I made in the attached version are: > > Removed the -t option as that's separate. > Removed erroneous 'c' from getopt() parameters. > Use K/N in code rather than M/N to match user instructions. > Added suffix len setter/checker based on N so that > we fail immediately if the wrong -a is specified, or > if -a is not specified we auto set it. > Flagged 0/N as an error, rather than treating like /N. > Changed r/K/N to buffer using stdio for much better performance (see below) > Fixed up the errno on some errors() > Normalized all "write error" messages so that all of these commands > output a single translated error message, of the form: > "split: write error: No space left on device" > split -n 1/10 $(which split) >/dev/full > stdbuf -o0 split -n 1/10 $(which split) >/dev/full > seq 10 | split -n r/1/10 >/dev/full > seq 10 | stdbuf -o0 split -n r/1/10 >/dev/full Nice. > Re the performance of the round robin implementation; > using stdio helps a LOT as can be seen with: > --- > $ time yes | head -n1000 | ./split-fwrite -n r/1/1 | wc -l > 1000 > > real0m1.568s > user0m1.486s > sys 0m0.072s > > $ time yes | head -n1000 | ./split-write -n r/1/1 | wc -l > 1000 > > real0m50.988s > user0m7.548s > sys 0m43.250s Indeed. We don't see a 32-x speed-up (esp. non-algorithmic) very often. > I still need to look at the round robin implementation when > outputting to file rather than stdout. I may default to using > stdio, but give an option to flush each line. I'm testing > with this currently which is performing badly just doing write() > --- > #create fifos > yes | head -n4 | ../split -n r/4 fifo > for f in x*; do rm $f && mkfifo $f; done > > #consumer > (for f in x*; do md5sum $f& done) > md5sum.out > > #producer > seq 10 | split -n r/4 > --- > > BTW, other modes perform well with write() > --- > $ yes | head -n1000 > 10m.txt > $ time ./split -n l/1/1 <10m.txt | wc -l > 1000 > > real0m0.201s > user0m0.145s > sys 0m0.043s > > $ time ./split -n 1/1 <10m.txt | wc -l > 1000 > > real0m0.199s > user0m0.154s > sys 0m0.041s > > $ time ./split -n 1 <10m.txt > > real0m0.088s > user0m0.000s > sys 0m0.081s > --- > > Here is stuff I intend TODO before checking in: > s/pread()/dd::skip()/ or at least add pread to bootstrap.conf > fix info docs for reworked interface > try
Re: modify chmod
Ubuntu stty
I'm using Ubuntu 9.10 (32-bit). I have an ASR-33 Teletype on ttyS0 which can only output uppercase characters that I'm trying to use as a serial console. My getty specifies -U to allow for uppercase terminals. stty -F/dev/ttyS0 shows this: speed 110 baud; line = 0; lnext = ; min = 1; time = 0; -brkint -icrnl ixoff iuclc -imaxbel olcuc -icanon -iexten -echo xcase As you can see, iuclc is clearly in effect but the conversion from uppercase to lowercase isn't being done. No commands work, and I had to create a user with an uppercase-only password just to be able to log in at all. I tried the following to see what I'm getting from the Teletype: cp /dev/ttyS0 test When I terminate the command, the file test has all uppercase in it. Shouldn't stty have converted the uppercase input to lowercase? If I set iuclc on tty0 (the normal console) it works as expected and I can turn on Caps Lock and still execute commands and cp tty0 test does put lowercase characters in test regardless of the state of Caps Lock. Is this a bug in stty? If so, I'd like to get a fix so I can use my terminals in a local antique computer museum. Thank you for your time. Tom Lake
Re: modify chmod
Re: modify chmod
According to jeff.liu on 2/5/2010 7:44 AM:
> +++ b/gnulib
> @@ -1 +1 @@
> -Subproject commit 2eb5a8a0ff8348149a9ca985e2ccbfb03bc8de53
> +Subproject commit 4b93a2579fb567b9fbbeb24439770d814dac95cd
Why are you modifying the gnulib submodule?
>
> +/* Some systems only have , other systems also
> + * have , where the former includes the latter.
> + * So it seems including the former is the best choice. */
> +#include "fs.h"
> +#if HAVE_SYS_VFS_H
> +# include
> +#else
> +# include
> +#endif
Hmm. POSIX standardized , not . Maybe a better
approach would be to propose a patch to gnulib that guarantees that
provides everything we need, so that we don't have to ever
worry about in coreutils.
>
> +/* Return true if the file resides on NFS filesystem.
> + * limit this optimization to systems that provide statfs. */
This comment formatting is not consistent with other comments in the file.
> +
> +static bool
> +may_have_nfsacl(const char *file)
Missing a space between the function name and the (. And coreutils
prefers 'char const *' over the synonymous 'const char *'.
> +{
> +# if HAVE_SYS_VFS_H && HAVE_SYS_STATFS_H && HAVE_STRUCT_STATFS_F_TYPE
> + struct statfs buf;
> +
> + /* If statfs fails, assume we can't use the optimization. */
> + if (statfs (file, &buf) < 0)
> + return true;
> +
> + return buf.f_type == S_MAGIC_NFS;
> +#endif
> +
> + return true;
Rather than have #ifdefs inside the function body, this seems like it
would be better to do:
#ifdef ...
function
#else
# define may_have_nfsacl(ignored) true
#else
> +}
> +
> /* Change the mode of FILE.
> Return true if successful. This function is called
> once for every file system object that fts encounters. */
> @@ -257,18 +290,38 @@ process_file (FTS *fts, FTSENT *ent)
> old_mode = file_stats->st_mode;
> new_mode = mode_adjust (old_mode, S_ISDIR (old_mode) != 0, umask_value,
> change, NULL);
> -
> - if (! S_ISLNK (old_mode))
> +
> + /* Do not touch the inode if the new file mode is the same as it was
> before;
> + * This is an optimization for some cases. However, the new behavior
> must be
> + * disabled for filesystems that support NFSv4 ACLs.
> + * The idea suggested by Paul Eggert refer to FreeBSD chmod(1).
> + * it uses pathconf(2)/lpathconf(2) to determine whether this is the case.
> + * but on linux, we lack of them. Instead, calling statfs(2) and
> compare the
> + * f_type we can still do optimization at least its not NFS. */
Again, inconsistent comment formatting.
> + if (file_stats->st_uid != euid && euid != 0)
> + error (0, 0, _("changing permissions of %s: Operation not permitted"),
> + quote (file_full_name));
Indentation in this patch looks wrong; did it get munged? Also, typically
the "Operation not permitted" is provided for free by passing the
appropriate errno to error(), rather than open-coding it into the error
string.
--
Don't work too hard, make some time for fun as well!
Eric Blake e...@byu.net
signature.asc
Description: OpenPGP digital signature
Re: modify chmod
A tiny patch, make chmod do not touch the inode if the new file
permission mode is same as it was before.
>From c7bc43a05e4f021db8047ecd390b4fb75368f558 Mon Sep 17 00:00:00 2001
From: Jeff Liu
Date: Fri, 5 Feb 2010 20:45:41 +0800
Subject: [PATCH 1/1] Modify chmod v1
do not touch the inode if the new file permission bits is identical to
it was before.
Signed-off-by: Jie Liu
---
gnulib | 2 +-
src/chmod.c | 75 +++
2 files changed, 66 insertions(+), 11 deletions(-)
diff --git a/gnulib b/gnulib
index 2eb5a8a..4b93a25 16
--- a/gnulib
+++ b/gnulib
@@ -1 +1 @@
-Subproject commit 2eb5a8a0ff8348149a9ca985e2ccbfb03bc8de53
+Subproject commit 4b93a2579fb567b9fbbeb24439770d814dac95cd
diff --git a/src/chmod.c b/src/chmod.c
index 3dab92e..30802d7 100644
--- a/src/chmod.c
+++ b/src/chmod.c
@@ -20,6 +20,7 @@
#include
#include
#include
+#include
#include "system.h"
#include "dev-ino.h"
@@ -32,6 +33,16 @@
#include "root-dev-ino.h"
#include "xfts.h"
+/* Some systems only have , other systems also
+ * have , where the former includes the latter.
+ * So it seems including the former is the best choice. */
+#include "fs.h"
+#if HAVE_SYS_VFS_H
+# include
+#else
+# include
+#endif
+
/* The official name of this program (e.g., no `g' prefix). */
#define PROGRAM_NAME "chmod"
@@ -83,6 +94,9 @@ static enum Verbosity verbosity = V_off;
Otherwise NULL. */
static struct dev_ino *root_dev_ino;
+/* The effective user ID of the caller process. */
+static uid_t euid;
+
/* For long options that have no equivalent short option, use a
non-character as a pseudo short option, starting with CHAR_MAX + 1. */
enum
@@ -170,6 +184,25 @@ describe_change (const char *file, mode_t mode,
(unsigned long int) (mode & CHMOD_MODE_BITS), &perms[1]);
}
+/* Return true if the file resides on NFS filesystem.
+ * limit this optimization to systems that provide statfs. */
+
+static bool
+may_have_nfsacl(const char *file)
+{
+# if HAVE_SYS_VFS_H && HAVE_SYS_STATFS_H && HAVE_STRUCT_STATFS_F_TYPE
+ struct statfs buf;
+
+ /* If statfs fails, assume we can't use the optimization. */
+ if (statfs (file, &buf) < 0)
+ return true;
+
+ return buf.f_type == S_MAGIC_NFS;
+#endif
+
+ return true;
+}
+
/* Change the mode of FILE.
Return true if successful. This function is called
once for every file system object that fts encounters. */
@@ -257,18 +290,38 @@ process_file (FTS *fts, FTSENT *ent)
old_mode = file_stats->st_mode;
new_mode = mode_adjust (old_mode, S_ISDIR (old_mode) != 0, umask_value,
change, NULL);
-
- if (! S_ISLNK (old_mode))
+
+ /* Do not touch the inode if the new file mode is the same as it was
before;
+ * This is an optimization for some cases. However, the new behavior
must be
+ * disabled for filesystems that support NFSv4 ACLs.
+ * The idea suggested by Paul Eggert refer to FreeBSD chmod(1).
+ * it uses pathconf(2)/lpathconf(2) to determine whether this is the case.
+ * but on linux, we lack of them. Instead, calling statfs(2) and
compare the
+ * f_type we can still do optimization at least its not NFS. */
+
+ if ((old_mode & CHMOD_MODE_BITS) == new_mode &&
+ !may_have_nfsacl (file_full_name))
{
- if (chmodat (fts->fts_cwd_fd, file, new_mode) == 0)
- chmod_succeeded = true;
+ if (file_stats->st_uid != euid && euid != 0)
+ error (0, 0, _("changing permissions of %s: Operation not permitted"),
+ quote (file_full_name));
else
- {
- if (! force_silent)
- error (0, errno, _("changing permissions of %s"),
- quote (file_full_name));
- ok = false;
- }
+ chmod_succeeded = true;
+ }
+ else
+ {
+ if (! S_ISLNK (old_mode))
+ {
+ if (chmodat (fts->fts_cwd_fd, file, new_mode) == 0)
+ chmod_succeeded = true;
+ else
+ {
+ if (! force_silent)
+ error (0, errno, _("changing permissions of %s"),
+ quote (file_full_name));
+ ok = false;
+ }
+ }
}
}
@@ -545,6 +598,8 @@ main (int argc, char **argv)
root_dev_ino = NULL;
}
+ euid = geteuid ();
+
ok = process_files (argv + optind,
FTS_COMFOLLOW | FTS_PHYSICAL | FTS_DEFER_STAT);
--
1.5.4.3
Re: modify chmod
Re: modify chmod
Sorry for the inconvenient, looks my email body was filtered somehow, :-(
Re: modify chmod
>From c7bc43a05e4f021db8047ecd390b4fb75368f558 Mon Sep 17 00:00:00 2001
From: Jeff Liu
Date: Fri, 5 Feb 2010 20:45:41 +0800
Subject: [PATCH 1/1] Modify chmod v1
do not touch the inode if the new file permission bits is identical to it was before.
Signed-off-by: Jie Liu
---
gnulib |2 +-
src/chmod.c | 75 +++
2 files changed, 66 insertions(+), 11 deletions(-)
diff --git a/gnulib b/gnulib
index 2eb5a8a..4b93a25 16
--- a/gnulib
+++ b/gnulib
@@ -1 +1 @@
-Subproject commit 2eb5a8a0ff8348149a9ca985e2ccbfb03bc8de53
+Subproject commit 4b93a2579fb567b9fbbeb24439770d814dac95cd
diff --git a/src/chmod.c b/src/chmod.c
index 3dab92e..30802d7 100644
--- a/src/chmod.c
+++ b/src/chmod.c
@@ -20,6 +20,7 @@
#include
#include
#include
+#include
#include "system.h"
#include "dev-ino.h"
@@ -32,6 +33,16 @@
#include "root-dev-ino.h"
#include "xfts.h"
+/* Some systems only have , other systems also
+ * have , where the former includes the latter.
+ * So it seems including the former is the best choice. */
+#include "fs.h"
+#if HAVE_SYS_VFS_H
+# include
+#else
+# include
+#endif
+
/* The official name of this program (e.g., no `g' prefix). */
#define PROGRAM_NAME "chmod"
@@ -83,6 +94,9 @@ static enum Verbosity verbosity = V_off;
Otherwise NULL. */
static struct dev_ino *root_dev_ino;
+/* The effective user ID of the caller process. */
+static uid_t euid;
+
/* For long options that have no equivalent short option, use a
non-character as a pseudo short option, starting with CHAR_MAX + 1. */
enum
@@ -170,6 +184,25 @@ describe_change (const char *file, mode_t mode,
(unsigned long int) (mode & CHMOD_MODE_BITS), &perms[1]);
}
+/* Return true if the file resides on NFS filesystem.
+ * limit this optimization to systems that provide statfs. */
+
+static bool
+may_have_nfsacl(const char *file)
+{
+# if HAVE_SYS_VFS_H && HAVE_SYS_STATFS_H && HAVE_STRUCT_STATFS_F_TYPE
+ struct statfs buf;
+
+ /* If statfs fails, assume we can't use the optimization. */
+ if (statfs (file, &buf) < 0)
+ return true;
+
+ return buf.f_type == S_MAGIC_NFS;
+#endif
+
+ return true;
+}
+
/* Change the mode of FILE.
Return true if successful. This function is called
once for every file system object that fts encounters. */
@@ -257,18 +290,38 @@ process_file (FTS *fts, FTSENT *ent)
old_mode = file_stats->st_mode;
new_mode = mode_adjust (old_mode, S_ISDIR (old_mode) != 0, umask_value,
change, NULL);
-
- if (! S_ISLNK (old_mode))
+
+ /* Do not touch the inode if the new file mode is the same as it was before;
+ * This is an optimization for some cases. However, the new behavior must be
+ * disabled for filesystems that support NFSv4 ACLs.
+ * The idea suggested by Paul Eggert refer to FreeBSD chmod(1).
+ * it uses pathconf(2)/lpathconf(2) to determine whether this is the case.
+ * but on linux, we lack of them. Instead, calling statfs(2) and compare the
+ * f_type we can still do optimization at least its not NFS. */
+
+ if ((old_mode & CHMOD_MODE_BITS) == new_mode &&
+ !may_have_nfsacl (file_full_name))
{
- if (chmodat (fts->fts_cwd_fd, file, new_mode) == 0)
-chmod_succeeded = true;
+ if (file_stats->st_uid != euid && euid != 0)
+error (0, 0, _("changing permissions of %s: Operation not permitted"),
+ quote (file_full_name));
else
-{
- if (! force_silent)
-error (0, errno, _("changing permissions of %s"),
- quote (file_full_name));
- ok = false;
-}
+chmod_succeeded = true;
+}
+ else
+{
+ if (! S_ISLNK (old_mode))
+ {
+if (chmodat (fts->fts_cwd_fd, file, new_mode) == 0)
+ chmod_succeeded = true;
+else
+ {
+if (! force_silent)
+ error (0, errno, _("changing permissions of %s"),
+ quote (file_full_name));
+ok = false;
+ }
+ }
}
}
@@ -545,6 +598,8 @@ main (int argc, char **argv)
root_dev_ino = NULL;
}
+ euid = geteuid ();
+
ok = process_files (argv + optind,
FTS_COMFOLLOW | FTS_PHYSICAL | FTS_DEFER_STAT);
--
1.5.4.3
Re: [PATCH] split: --chunks option
I got a bit of time for the review last night... This was your last interface change for this: -b, --bytes=SIZEput SIZE bytes per output file\n\ + -b, --bytes=/N generate N output files\n\ + -b, --bytes=K/N print Kth of N chunks of file\n\ -C, --line-bytes=SIZE put at most SIZE bytes of lines per output file\n\ -d, --numeric-suffixes use numeric suffixes instead of alphabetic\n\ -l, --lines=NUMBER put NUMBER lines per output file\n\ + -l, --lines=/N generate N eol delineated output files\n\ + -l, --lines=K/N print Kth of N eol delineated chunks\n\ + -n, --number=N same as --bytes=/N\n\ + -n, --number=K/Nsame as --bytes=K/N\n\ + -r, --round-robin=N generate N eol delineated output files using\n\ +round-robin style distribution.\n\ + -r. --round-robin=K/N print Kth of N eol delineated chunk as -rN would\n\ +have generated.\n\ + -t, --term=CHAR specify CHAR as eol. This will also convert\n\ +-b to its line delineated equivalent (-C if\n\ +splitting normally, -l if splitting by\n\ +chunks). C escape sequences are accepted.\n\ Thinking more about it, I think adding 2 modes of operation to the already slightly complicated -bCl options is too confusing. Since this is a separate mode of operation; i.e. one would be specifying a particular number of files for a different reason than a particular size, it would be better as a separate option. So I changed -n to operate as follows. This is more general if we want to add new split methods in future, and also compatible with the existing BSD -n without needing a redundant option. -n N split into N files based on size of input -n K/N output K of N to stdout -n l/N split into N files while maintaining lines -n l/K/N output K of N to stdout while maintaining lines -n r/N like `l' but use round robin distribution instead of size -n r/K/N likewise but only output K of N to stdout Other changes I made in the attached version are: Removed the -t option as that's separate. Removed erroneous 'c' from getopt() parameters. Use K/N in code rather than M/N to match user instructions. Added suffix len setter/checker based on N so that we fail immediately if the wrong -a is specified, or if -a is not specified we auto set it. Flagged 0/N as an error, rather than treating like /N. Changed r/K/N to buffer using stdio for much better performance (see below) Fixed up the errno on some errors() Normalized all "write error" messages so that all of these commands output a single translated error message, of the form: "split: write error: No space left on device" split -n 1/10 $(which split) >/dev/full stdbuf -o0 split -n 1/10 $(which split) >/dev/full seq 10 | split -n r/1/10 >/dev/full seq 10 | stdbuf -o0 split -n r/1/10 >/dev/full Re the performance of the round robin implementation; using stdio helps a LOT as can be seen with: --- $ time yes | head -n1000 | ./split-fwrite -n r/1/1 | wc -l 1000 real0m1.568s user0m1.486s sys 0m0.072s $ time yes | head -n1000 | ./split-write -n r/1/1 | wc -l 1000 real0m50.988s user0m7.548s sys 0m43.250s --- I still need to look at the round robin implementation when outputting to file rather than stdout. I may default to using stdio, but give an option to flush each line. I'm testing with this currently which is performing badly just doing write() --- #create fifos yes | head -n4 | ../split -n r/4 fifo for f in x*; do rm $f && mkfifo $f; done #consumer (for f in x*; do md5sum $f& done) > md5sum.out #producer seq 10 | split -n r/4 --- BTW, other modes perform well with write() --- $ yes | head -n1000 > 10m.txt $ time ./split -n l/1/1 <10m.txt | wc -l 1000 real0m0.201s user0m0.145s sys 0m0.043s $ time ./split -n 1/1 <10m.txt | wc -l 1000 real0m0.199s user0m0.154s sys 0m0.041s $ time ./split -n 1 <10m.txt real0m0.088s user0m0.000s sys 0m0.081s --- Here is stuff I intend TODO before checking in: s/pread()/dd::skip()/ or at least add pread to bootstrap.conf fix info docs for reworked interface try to refactor duplicated code cheers, Pádraig. >From 8a7fe06170ad8bb3050c4b6a43c9e51eb0ec22a7 Mon Sep 17 00:00:00 2001 From: Chen Guo Date: Fri, 8 Jan 2010 03:42:27 -0800 Subject: [PATCH] split: add --number to generate a particular number of files * doc/coreutils.texi: update documentation of split. * src/split.c (usage, long_options, main): New options --number. (set_suff
diffutils release imminent
For those not subscribed to the bug-diffutils mailing list, you may be interested in this snapshot: http://thread.gmane.org/gmane.comp.gnu.diffutils.bugs/11 I expect to make a release next week, so any patches, testing or feedback would be most welcome.
[PATCH] sync with gnulib
This syncs with gnulib for two changes:
newer regcomp.c and avoid a warning in test-gettimeofday.c.
The former required an adjustment to the patch we use in gl/lib:
It lets me removes two chunks.
>From 69cc8d5cb711823fd97992076494a5888c20da10 Mon Sep 17 00:00:00 2001
From: Jim Meyering
Date: Wed, 3 Feb 2010 21:57:24 +0100
Subject: [PATCH] sync with gnulib
* gl/lib/regcomp.c.diff: Update to apply to changed version in gnulib.
* gnulib: Update submodule to latest.
---
gl/lib/regcomp.c.diff | 39 ++-
gnulib|2 +-
2 files changed, 11 insertions(+), 30 deletions(-)
diff --git a/gl/lib/regcomp.c.diff b/gl/lib/regcomp.c.diff
index a31c39f..63fc187 100644
--- a/gl/lib/regcomp.c.diff
+++ b/gl/lib/regcomp.c.diff
@@ -1,17 +1,8 @@
-diff --git c/lib/regcomp.c i/lib/regcomp.c
-index 6472ff6..665b2ab 100644
c/lib/regcomp.c
-+++ i/lib/regcomp.c
-@@ -18,6 +18,8 @@
-with this program; if not, write to the Free Software Foundation,
-Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. */
-
-+#include "intprops.h"
-+#include "verify.h"
- static reg_errcode_t re_compile_internal (regex_t *preg, const char * pattern,
- size_t length, reg_syntax_t syntax);
- static void re_compile_fastmap_iter (regex_t *bufp,
-@@ -541,7 +543,7 @@ regerror (errcode, preg, errbuf, errbuf_size)
+diff --git a/lib/regcomp.c b/lib/regcomp.c
+index d5968bd..4926676 100644
+--- a/lib/regcomp.c
b/lib/regcomp.c
+@@ -541,7 +541,7 @@ regerror (errcode, preg, errbuf, errbuf_size)
size_t errbuf_size;
#else /* size_t might promote */
size_t
@@ -20,7 +11,7 @@ index 6472ff6..665b2ab 100644
char *_Restrict_ errbuf, size_t errbuf_size)
#endif
{
-@@ -1375,7 +1377,7 @@ calc_first (void *extra, bin_tree_t *node)
+@@ -1383,7 +1383,7 @@ calc_first (void *extra, bin_tree_t *node)
/* Pass 2: compute NEXT on the tree. Preorder visit. */
static reg_errcode_t
@@ -29,17 +20,7 @@ index 6472ff6..665b2ab 100644
{
switch (node->token.type)
{
-@@ -2571,7 +2573,8 @@ parse_dup_op (bin_tree_t *elem, re_string_t *regexp,
re_dfa_t *dfa,
- /* This loop is actually executed only when end != REG_MISSING,
- to rewrite {0,n} as ((...?)?)?... We have
- already created the start+1-th copy. */
-- if ((Idx) -1 < 0 || end != REG_MISSING)
-+ verify (! TYPE_SIGNED (Idx));
-+ if (end != REG_MISSING)
- for (i = start + 2; i <= end; ++i)
- {
- elem = duplicate_tree (elem, dfa);
-@@ -2731,7 +2734,8 @@ static reg_errcode_t
+@@ -2744,7 +2744,8 @@ static reg_errcode_t
internal_function
build_collating_symbol (bitset_t sbcset,
# ifdef RE_ENABLE_I18N
@@ -49,7 +30,7 @@ index 6472ff6..665b2ab 100644
# endif
const unsigned char *name)
{
-@@ -3309,7 +3313,8 @@ parse_bracket_exp (re_string_t *regexp, re_dfa_t *dfa,
re_token_t *token,
+@@ -3323,7 +3324,8 @@ parse_bracket_exp (re_string_t *regexp, re_dfa_t *dfa,
re_token_t *token,
static reg_errcode_t
parse_bracket_element (bracket_elem_t *elem, re_string_t *regexp,
@@ -59,7 +40,7 @@ index 6472ff6..665b2ab 100644
reg_syntax_t syntax, bool accept_hyphen)
{
#ifdef RE_ENABLE_I18N
-@@ -3396,8 +3401,9 @@ parse_bracket_symbol (bracket_elem_t *elem, re_string_t
*regexp,
+@@ -3410,8 +3412,9 @@ parse_bracket_symbol (bracket_elem_t *elem, re_string_t
*regexp,
static reg_errcode_t
#ifdef RE_ENABLE_I18N
@@ -71,7 +52,7 @@ index 6472ff6..665b2ab 100644
#else /* not RE_ENABLE_I18N */
build_equiv_class (bitset_t sbcset, const unsigned char *name)
#endif /* not RE_ENABLE_I18N */
-@@ -3798,7 +3804,7 @@ free_token (re_token_t *node)
+@@ -3816,7 +3819,7 @@ free_token (re_token_t *node)
and its children. */
static reg_errcode_t
diff --git a/gnulib b/gnulib
index 2eb5a8a..de4d0a3 16
--- a/gnulib
+++ b/gnulib
@@ -1 +1 @@
-Subproject commit 2eb5a8a0ff8348149a9ca985e2ccbfb03bc8de53
+Subproject commit de4d0a383be20eb6cbeb09ecbd66a5514779b914
--
1.7.0.rc1.204.gb96e
Re: BUG: date 7.1 - reading date string
Dear Sven, Bob, Philip, thank a lot for your hints! The daylight saving time (DST) is indeed the problem, so I have to use the '-u' option of date for proper working of my script. Sorry for bothering, I should have thought myself about this. Maybe an additional hint in the error message to the DST problem would help other users (like (DST gap))? thanks again Klaus On Thu, 4 Feb 2010, Sven Joachim wrote: Am 04.02.2010 um 14:45 schrieb Klaus Bramstedt: I used the command date -d "$DATE" for almost all dates between Aug 2002 end Jan 2010 in a script. Date has a format like this "2003-03-30 02:08:17" However, 7 of 42048 randomly distributed dates in the list failed: date: ungültiges Datum „2003-03-30 02:08:17“ date: ungültiges Datum „2003-03-30 02:25:58“ date: ungültiges Datum „2005-03-27 02:08:41“ date: ungültiges Datum „2006-03-26 02:41:44“ date: ungültiges Datum „2008-03-30 02:08:09“ date: ungültiges Datum „2009-03-29 02:09:26“ date: ungültiges Datum „2007-03-25 02:03:47“ For me, this seems to be a bug in the date reading part. The critical combination seems to be end of March and '02' as hour. Yes, it's the non-existent hour during DST adjustment on the last Sunday in March. These dates are indeed invalid in the German timezone. Sven
