This is following on from this change:
[02c3dc9d 2008-03-06 cat: use larger buffer sizes ...]
which increased the IO block size used by cat by 8 times,
but also capped it at 32KiB.
* NEWS: Mention the change in behavior.
* src/system.h: Add a new io_blksize() function that
returns the max of ST_BLKSIZE or 32KiB, as this was
seen as a good value for a minimum block size to use
to get good performance while minimizing system call overhead.
* src/cat.c: Use it.
* src/copy.c: ditto
* src/split.c: ditto
* src/pr.c (print_header): Honor chars_per_margin also for the header.
Inspired by a patch from C de-Avillez.
This change also adjusts two other %*s directives to be %*.*s instead,
(for the padding on either side of "file_text") to correct a bug that
could make pr print one space of padding instead of none.
* NEWS (Bug fixes): Mention it.
* tests/pr/o3Jml24f-lm-lo: Adjust for new expected output.
* tests/pr/o3a3Sl24f-tn: Likewise.
* tests/pr/o3a3Snl24f-tn: Likewise.
* tests/pr/o3a3l24f-tn: Likewise.
* tests/pr/o3b3Sl24f-tn: Likewise.
* tests/pr/o3b3Snl24f-tn: Likewise.
* tests/pr/o3b3l24f-tn: Likewise.
* tests/pr/o3mSl24f-bl-tn: Likewise.
* tests/pr/o3mSnl24fbltn: Likewise.
* tests/pr/o3ml24f-bl-tn: Likewise.
* src/cat.c (max): Remove definition. Use MAX from system.h instead.
(compute_buffer_size): New function to compute the input and output
buffer sizes, which are now set at 8 times st_blksize with a maximum
of 32KiB. Previously the typical block sizes used were 1KiB for pipes
and 4KiB for files, and now will be 8KiB and 32KiB respectively.
(main): Use it.
This change can double throughput on modern systems. For timings,
see http://article.gmane.org/gmane.comp.gnu.core-utils.bugs/16040
Suggestion from Tzvi Rotshtein.
* README-release: Fix typo. it's "maintainer-clean".
Refer to HACKING for full list of dependencies.
Use cvs add's -ko option to inhibit $Id$-style keyword expansion.
* maint.mk (web-manual): Factor out definition of $(manual_title).
(vc-dist): Set XZ_OPT=-9ev for the final "make dist".
* cfg.mk (manual_title): Define it here.
Suggestions from Eric Blake.
* tests/rm/ext3-perf: If set-up takes too long, use the set-up time
as the removal time limit. Otherwise, a parallel "make check" would
usually fail when run with RUN_VERY_EXPENSIVE_TESTS=yes.
* tests/mkdir/perm: Now that this test is no longer always skipped,
make it work independently of ls: use stat, not ls, to get perm string.
* test-lib.sh (rwx_to_mode_): Use stat, not ls to get perm string
This test was run only when RUN_VERY_EXPENSIVE_TESTS=yes was set.
* bootstrap.conf (gnulib_modules): Add manywarnings.
* configure.ac: Use gl_MANYWARN_ALL_GCC, and exclude options
I don't want or that provoke too many warnings.
(WARN_CFLAGS, WERROR_CFLAGS): Define.
(lint, GNULIB_PORTCHECK): Define.
(_FORTIFY_SOURCE): Define to 2.
Running "mkdir dir; cp -rl dir dir" would create dir/dir/dir/...
rather than diagnosing the "copy-into-self" failure.
The easy fix would have been to revert this part of the change
[3ece0355 2008-11-09 cp: use far less memory in some cases]
that introduced the bug:
- remember_copied (dst_name, dst_sb.st_ino, dst_sb.st_dev);
+ if (!x->hard_link)
+ remember_copied (dst_name, dst_sb.st_ino, dst_sb.st_dev);
However, that would have induced the failure of the new cp/link-heap
test, due to the added memory pressure of recording 10k dev/ino pairs.
And besides, I liked that improvement and wanted to keep it.
Now that it's obvious recording the just-created-directory dev/ino
needn't depend on the setting of hard_link, I realized it is necessary
to record the pair only for the first directory created for each
source command-line argument.
I made that change, then noticed the new test, cp -rl a d d, would pass
when run once, yet output the into-self diagnostic twice. Also note
the side effect: it creates d/a and d/d. However, running that same
command a second time, now with the modified directory, would fail.
That turned out to be due to the fact that although the first into-self
failure was detected in copy_dir, that function would continue copying
other entries regardless -- and that would make it fail (eventually)
with the unwanted recursion.
* src/copy.c (copy_internal): This function needed an indicator of
whether, for a give command line argument, it had already created its
first directory. If so, no more need to record dev/ino pairs. If this
is the first, then do record its pair. Hence, the new parameter.
(copy_dir, copy): Update callers.
(copy_dir): Upon any into-self failure, break out of the loop.
* tests/cp/into-self: Test for the above.
Reported by Mikael Magnusson.
* src/sort.c: When no specific number of chars to skip
is specified for the end field, always skip the whole field.
Also never include leading spaces from next field.
* tests/misc/sort: Add 2 new tests for these cases.
* NEWS: Mention this bug fix.
* THANKS: Add bug reporter.
Reported by Davide Canova.
* gl/lib/argv-iter.h (_ATTRIBUTE_NONNULL_): Don't assume C99 variadic
macros are portable yet; needed for at least IRIX/MIPSpro.
* THANKS: Update.
Reported by Stuart Shelton.
* tests/dd/skip-seek-past-file: Just check for
the non system specific part of the error strings.
This was causing an erroneous failure on NetBSD 1.6 at least.
Running "make check &" would hang due to SIGTTOU and SIGTTIN signals.
* tests/misc/stty: Ignore SIGTTOU.
* tests/misc/stty-invalid: Likewise.
* tests/mv/i-3: Ignore SIGTTIN.
* src/ls.c (usage): Make the description of -s mention "allocated size",
so that it's less likely to be confused with an apparent byte-count.
Suggested by Vito Caputo.
