The mergesort removal from qsort implementation (commit 03bf8357e8)
had the side-effect of making sorting nonstable. Although neither
POSIX nor C standard specify that qsort should be stable, it seems
that it has become an instance of Hyrum's law where multiple programs
expect it.
Also, the resulting introsort implementation is not faster than
the previous mergesort (which makes the change even less appealing).
This patch restores the previous mergesort implementation, with the
exception of machinery that checks the resulting allocation against
the _SC_PHYS_PAGES (it only adds complexity and the heuristic not
always make sense depending on the system configuration and load).
The alloca usage was replaced with a fixed-size buffer.
For the fallback mechanism, the implementation uses heapsort. It is
simpler than quicksort, and it does not suffer from adversarial
inputs. With memory overcommit, it should be rarely triggered.
The drawback is mergesort requires O(n) extra space, and since it is
allocated with malloc the function is AS-signal-unsafe. It should be
feasible to change it to use mmap, although I am not sure how urgent
it is. The heapsort is also nonstable, so programs that require a
stable sort would still be subject to this latent issue.
The tst-qsort5 is removed since it will not create quicksort adversarial
inputs with the current qsort_r implementation.
Checked on x86_64-linux-gnu and aarch64-linux-gnu.
Reviewed-by: Florian Weimer <fweimer@redhat.com>
This patch removes the mergesort optimization on qsort implementation
and uses the introsort instead. The mergesort implementation has some
issues:
- It is as-safe only for certain types sizes (if total size is less
than 1 KB with large element sizes also forcing memory allocation)
which contradicts the function documentation. Although not required
by the C standard, it is preferable and doable to have an O(1) space
implementation.
- The malloc for certain element size and element number adds
arbitrary latency (might even be worse if malloc is interposed).
- To avoid trigger swap from memory allocation the implementation
relies on system information that might be virtualized (for instance
VMs with overcommit memory) which might lead to potentially use of
swap even if system advertise more memory than actually has. The
check also have the downside of issuing syscalls where none is
expected (although only once per execution).
- The mergesort is suboptimal on an already sorted array (BZ#21719).
The introsort implementation is already optimized to use constant extra
space (due to the limit of total number of elements from maximum VM
size) and thus can be used to avoid the malloc usage issues.
Resulting performance is slower due the usage of qsort, specially in the
worst-case scenario (partialy or sorted arrays) and due the fact
mergesort uses a slight improved swap operations.
This change also renders the BZ#21719 fix unrequired (since it is meant
to fix the sorted input performance degradation for mergesort). The
manual is also updated to indicate the function is now async-cancel
safe.
Checked on x86_64-linux-gnu.
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
On platforms where long double used to have the same format as double,
but later switched to a different format (alpha, s390, sparc, and
powerpc), accessing the older behavior is possible and it happens via
__nldbl_* functions (not on the API, but accessible from header
redirection and from compat symbols). These functions write to the
global flag __ldbl_is_dbl, which tells other functions that long double
variables should be handled as double. This patch takes the first step
towards removing this global flag and creates __vstrfmon_l_internal,
which takes an explicit flags parameter.
This change arguably makes the generated code slightly worse on
architectures where __ldbl_is_dbl is never true; right now, on those
architectures, it's a compile-time constant; after this change, the
compiler could theoretically prove that __vstrfmon_l_internal was
never called with a nonzero flags argument, but it would probably need
LTO to do it. This is not performance critical code and I tend to
think that the maintainability benefits of removing action at a
distance are worth it. However, we _could_ wrap the runtime flag
check with a macro that was defined to ignore its argument and always
return false on architectures where __ldbl_is_dbl is never true, if
people think the codegen benefits are important.
Tested for powerpc and powerpc64le.
"%OB" is considered a conversion specifier ("B" is the format
specifier), and the list of format specifiers for months in the
description of the optional "O" modifier was incomplete. A
cross-reference from the ALTMON_* constants to the strftime section
is also provided. Lastly, some grammatical fixes (commas) are made
and paragraphs refactored (rewrapped).
* manual/locale.texi (ALTMON_1, ALTMON_2, ALTMON_3, ALTMON_4,
ALTMON_5, ALTMON_6, ALTMON_7, ALTMON_8, ALTMON_9, ALTMON_10,
ALTMON_11, ALTMON_12): Improve documentation.
* manual/time.texi (strftime): Likewise.
[BZ #10871]
* manual/locale.texi: Document ALTMON_1..12 constants for
nl_langinfo. Specify when to use ALTMON instead of MON.
* manual/time.texi (strftime, strptime): Document GNU extension
permitting O modifier with %B and %b. Specify when to use
%OB instead of %B.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
The Summary is now generated from @standards, and syntax-checking is
performed. If invalid @standards syntax is detected, summary.pl will
fail, reporting all errors. Failure and error reporting is disabled
for now, however, since much of the manual is still incomplete
wrt. header and standards annotations.
Note that the sorting order of the Summary has changed; summary.pl
respects the locale, like summary.awk did, but the use of LC_ALL=C is
introduced in the Makefile. Other notable deviations are improved
detection of the annotated elements' names, which are used for
sorting, and improved detection of the @node used to reference into
the manual. The most noticeable difference in the rendered Summary is
that entries may now contain multiple lines, one for each header and
standard combination.
summary.pl accepts a `--help' option, which details the expected
syntax of @standards. If errors are reported, the user is directed to
this feature for further information.
* manual/Makefile: Generate summary.texi with summary.pl.
Force use of the C locale. Update Perl dependency comment.
* manual/header.texi: Update reference to summary.awk.
* manual/macros.texi: Refer authors to `summary.pl --help'.
* manual/summary.awk: Remove file.
* manual/summary.pl: New file. Generate summary.texi, and
check for @standards-related syntax errors.
* manual/argp.texi: Convert header and standards @comments to
@standards.
* manual/arith.texi: Likewise.
* manual/charset.texi: Likewise.
* manual/conf.texi: Likewise.
* manual/creature.texi: Likewise.
* manual/crypt.texi: Likewise.
* manual/ctype.texi: Likewise.
* manual/debug.texi: Likewise.
* manual/errno.texi: Likewise.
* manual/filesys.texi: Likewise.
* manual/getopt.texi: Likewise.
* manual/job.texi: Likewise.
* manual/lang.texi: Likewise.
* manual/llio.texi: Likewise.
* manual/locale.texi: Likewise.
* manual/math.texi: Likewise.
* manual/memory.texi: Likewise.
* manual/message.texi: Likewise.
* manual/pattern.texi: Likewise.
* manual/pipe.texi: Likewise.
* manual/process.texi: Likewise.
* manual/resource.texi: Likewise.
* manual/search.texi: Likewise.
* manual/setjmp.texi: Likewise.
* manual/signal.texi: Likewise.
* manual/socket.texi: Likewise.
* manual/startup.texi: Likewise.
* manual/stdio.texi: Likewise.
* manual/string.texi: Likewise.
* manual/sysinfo.texi: Likewise.
* manual/syslog.texi: Likewise.
* manual/terminal.texi: Likewise.
* manual/threads.texi: Likewise.
* manual/time.texi: Likewise.
* manual/users.texi: Likewise.
This commit handles exceptional cases of invalid syntax for the
@standards conversion script.
* manual/crypt.texi: Move a comment out of an @*x list.
* manual/filesys.texi: Refactor some comments, one of which
looks like a standard. Fix incorrectly separated standards.
* manual/locale.texi: Invert an annotation.
* manual/resource.texi: Fix incorrectly separated standards.
* manual/time.texi: Refactor a @vtable that obscures an
annotation.
* manual/users.texi: Refactor multiple headers to occupy a
single @comment.
* manual/examples/strncat.c: Remove.
This example was misleading, as the code would have undefined
behavior if "hello" was longer than SIZE. Anyway, the manual
shouldn't encourage strncpy+strncat for this sort of thing.
* manual/string.texi (Copying Strings and Arrays): Split into
three sections Copying Strings and Arrays, Concatenating Strings,
and Truncating Strings, as this section was way too long. All
cross-referenced changed. Add advice about string-truncation
functions. Remove misleading strncat example.
1999-08-28 Ulrich Drepper <drepper@cygnus.com>
* malloc/malloc.c (ptmalloc_init): Don't use variables to set
thresholds for SUID binaries. [PR libc/1277]
1999-08-28 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* manual/install.texi (Running make install): Give examples for
timezone and locale installation.
1999-08-28 Zack Weinberg <zack@bitmover.com>
* glibcbug.in: Get CCVERSION from autoconf substitution.
* configure.in: Calculate and substitute CCVERSION.
1999-08-28 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* sysdeps/unix/sysv/linux/mips/bits/sigaction.h: Fix sa_flags,
partially reverting a patch from 1998-12-29. We just can't change
the flags, kernels with different flags lead to
incompatibilities.
1999-01-11 Ulrich Drepper <drepper@cygnus.com>
* ctype/Versions [GLIBC_2.0]: Export __ctype32_b.
* include/wctype.h: Declare __iswctype.
* stdio-common/vfscanf.c (__vfscanf): Use __iswspace instead of
iswspace.
* wctype/Makefile (routines): Add wcextra_l.
* wctype/wcextra.c (iswblank): Implement function here and don't use
__iswctype.
(__iswblank_l): Move definition to...
* wctype/wcextra_l.c: ...here. New file.
* wctype/wcfuncs.c: Really implement functions and don't call
__iswctype or __towctrans.
* wctype/wctype.h: Change isw* and tow* macros. Don't call
__iswctype or __towctrans. Instead optimize constant argument case.
* iconv/gconv.h: Fix typos.
* iconv/skeleton.c: Fix typos. Optimize init function a bit.
Correctly emit escape sequence to return to initial state in
conversion function.
* iconvdata/iso-2022-jp.c (gconv_init): Correctly initialize
max_needed_to element.
* manual/mbyte.texi: Removed. This is now described in charset.texi.
* manual/charset.texi: New file.
* manual/Makefile (chapters): Replace mbyte by charset.
* manual/ctype.texi: Document wide character functions.
* manual/intro.texi: Fix reference to mbyte chapter.
* manual/lang.texi: Likewise.
* manual/locale.texi: Likewise.
* manual/stdio.texi: Likewise.
* manual/string.texi: Fix @node line for new charset chapter.
* manual/libc.texinfo (UPDATED): Updated. Also update copyright years.
* manual/memory.texi (savestring): Optimize code to give a good
example.
* manual/filesys.texi: Fix wording. Patches by Jim Meyering.
* nscd/nscd_getgr_r.c: Include stdint.h to get uintptr_t definition.
* nscd/nscd_getpw_r.c: Likewise.
* nscd/nscd_gethst_r.c: Likewise.
* stdlib/stdtold_l.c: Always include xlocale.h.
1999-01-11 Geoffrey Keating <geoffk@ozemail.com.au>
* stdlib/fpioconst.h (LDBL_MAX_10_EXP_LOG): Define to be same as
DBL_MAX_10_EXP_LOG if there is no long double.
(_fpioconst_pow10): Always use size as LDBL_MAX_10_EXP_LOG to match
printf_fp.c.
1999-01-10 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* timezone/Makefile ($(testdata)/GB): Changed to ...
($(testdata)/Europe/London): ... for tst-timezone test.
($(objpfx)tst-timezone.out): Change GB to Europe/London.
* timezone/tst-timezone.c (main): Enable DST switching test,
change GB to Europe/London.
1999-01-10 Philip Blundell <philb@gnu.org>
* socket/Makefile (headers): Remove bits/sockunion.h.
1999-01-09 Philip Blundell <philb@gnu.org>
* socket/sys/socket.h: Don't include <bits/sockunion.h>.
* sysdeps/generic/bits/sockunion.h: Deleted.
* sysdeps/unix/sysv/linux/bits/sockunion.h: Likewise.
1999-01-08 H.J. Lu <hjl@gnu.org>
* io/fts.c (fts_close): Don't access memory after having it freed.
1998-11-16 Ulrich Drepper <drepper@cygnus.com>
* sysdeps/unix/opendir.c: Use o_directory_works differently. Treat
zero value as uninitialized so that variable can go into .bss.
* dirent/opendir-tst1.c: Print error strings to stdout.
* sysdeps/unix/sysv/linux/mmap64.c: New file. Empty.
* sysdeps/unix/sysv/linux/syscalls.list: Add mmap with mmap64 alias.
* sysdeps/unix/sysv/linux/arm/mmap.S: Add mmap64 alias.
* sysdeps/unix/sysv/linux/i386/mmap.S: Likewise.
* sysdeps/unix/sysv/linux/m68k/mmap.S: Likewise.
1998-11-16 Andreas Schwab <schwab@issan.cs.uni-dortmund.de>
* time/tzfile.c (__tzfile_read): Check that types[i].isdst is in
range.
1998-11-16 Ulrich Drepper <drepper@cygnus.com>
* libio/bits/stdio.h (fread_unlocked): Do handle zero size and number.
(fwrite_unlocked): Likewise.
1998-11-16 Andreas Schwab <schwab@issan.cs.uni-dortmund.de>
* libio/bits/stdio.h (fread_unlocked): Don't evaluate non-constant
twice. Don't bother handling zero size.
(frwite_unlocked): Likewise.
1998-11-15 Andreas Schwab <schwab@issan.cs.uni-dortmund.de>
* configure.in: Fix regex for makeinfo version string.
1998-11-15 Andreas Schwab <schwab@issan.cs.uni-dortmund.de>
* sunrpc/rpc_sample.c (write_sample_server): Avoid -Wparentheses
warning.
* sunrpc/rpc_svcout.c (write_program): Likewise.
1998-11-15 Andreas Schwab <schwab@issan.cs.uni-dortmund.de>
* time/Makefile (CFLAGS-test_time.c): Add -Wno-format.
1998-11-16 Ulrich Drepper <drepper@cygnus.com
* misc/sys/mman.h: Pretty print prototypes.
1998-11-15 Ulrich Drepper <drepper@cygnus.com
* misc/error.c (error): Don't use result of strerror_r call directly
since at least DEC Unix has a wrong return type.
Patch by Johan Danielsson <joda@pdc.kth.se>.
1998-11-16 Andreas Schwab <schwab@issan.cs.uni-dortmund.de>
1998-10-16 10:07 Ulrich Drepper <drepper@cygnus.com>
* iconv/Makefile [! elf]: Define CFLAGS-gconv_simple.c to
-DSTATIC_GCONV.
* iconv/skeleton.c: Include ELF header only of STATIC?GCONV is not
defined. Avoid using DL_CLL_FCT if it is not defined.
1998-10-16 10:40 -0400 Zack Weinberg <zack@rabi.phys.columbia.edu>
* resolv/arpa/nameser.h: Include sys/types.h unconditionally.
Include endian.h unconditionally. Remove portability goop
depending on #ifdef linux, #ifdef BSD, #ifdef machine-type.
1998-10-16 11:39 -0400 Zack Weinberg <zack@rabi.phys.columbia.edu>
* sunrpc/rpc/types.h: Mark file so fixincludes won't modify it.
1998-09-01 10:34 Ulrich Drepper <drepper@cygnus.com>
* manual/locale.texi: Almost compelte rewrite. Document more functions
and functionality.
* manual/arith.texi: Correct reference.
* manual/string.texi: Pretty printing.
* manual/texinfo.tex: Update from last available version.
1998-08-31 22:44 Ulrich Drepper <drepper@cygnus.com>
* nis/nss_nis/nis-pwd.c (_nss_nis_getpwnam_r): Correct test for
invalid password.
(_nss_nis_getpwuid_r): Likewise.
Patch by Matthew Arnison <matthewa@physics.usyd.edu.au>.
* inet/rcmd.c: Implement netgroup support.
Patch by Dick Streefland <dick_streefland@tasking.com>.
Thu Dec 19 23:28:33 1996 Ulrich Drepper <drepper@cygnus.com>
* resolv/resolv.h: Update from BIND 4.9.5-P1.
* resolv/res_comp.c: Likewise.
* resolv/res_debug.c: Likewise.
* resolv/Banner: Update version number.
Thu Dec 19 20:58:53 1996 Ulrich Drepper <drepper@cygnus.com>
* elf/dlfcn.h: Add extern "C" wrapper.
* io/utime.h: Don't define NULL since this isn't allowed in POSIX.
* io/sys/stat.h: Declare `lstat' only if __USE_BSD ||
__USE_XOPEN_EXTENDED.
* locale/locale.h: Define NULL.
* math/math.c: Don't include <errno.h> to define math errors.
* stdlib/stdlib.h: Likewise.
* posix/unistd.h: Don't declare environ.
* posix/sys/utsname.h (struct utsname): Declare member domainname
as __domainname is !__USE_GNU.
* signal/signal.h: Declare size_t only if __USE_BSD ||
__USE_XOPEN_EXTENDED.
* stdio/stdio.h: Don't declare cuserid when __USE_POSIX, but
instead when __USE_XOPEN.
* string/string.h: Define strndup only if __USE_GNU.
* sysdeps/unix/sysv/linux/clock.c: New file.
* sysdeps/unix/sysv/linux/timebits.h: Define CLOCKS_PER_SEC as
1000000 per X/Open standard.
* features.h: Add code to recognize _POSIX_C_SOURCE value 199309.
Define __USE_POSIX199309.
* posix/unistd.h: Declare fdatasync only if __USE_POSIX199309.
* time/time.c: Declare nanosleep only if __USE_POSIX199309.
Patches by Rüdiger Helsch <rh@unifix.de>.
* locale/locale.h: Add declaration of newlocale and freelocale.
* new-malloc/Makefile (distibute): Add mtrace.awk.
(dist-routines): Add mcheck and mtrace.
(install-lib, non-lib.a): Define as libmcheck.a.
* new-malloc/malloc.h: Add declaration of __malloc_initialized.
* new-malloc/mcheck.c: New file.
* new-malloc/mcheck.h: New file.
* new-malloc/mtrace.c: New file.
* new-malloc/mtrace.awk: New file.
* posix/unistd.h: Correct prototype for usleep.
* sysdeps/unix/bsd/usleep.c: De-ANSI-declfy. Correct return type.
* sysdeps/unix/sysv/linux/usleep.c: Real implementation based on
nanosleep.
* signal/signal.h: Change protoype of __sigpause to take two
arguments. Remove prototype for sigpause. Add two different
macros named sigpause selected when __USE_BSD or __USE_XOPEN
are defined. This is necessary since the old BSD definition
of theis function collides with the X/Open definition.
* sysdeps/posix/sigpause.c: Change function definition to also
fit X/Open definition.
* sysdeps/libm-i387/e_exp.S: Make sure stack is empty when the
function is left.
* sysdeps/libm-i387/e_expl.S: Likewise.
Patch by HJ Lu.
1996-12-17 Paul Eggert <eggert@twinsun.com>
* many, many files: Spelling corrections.
* catgets/catgetsinfo.h (mmapped):
Renamed from mmaped (in struct catalog_info.status).
* mach/err_kern.sub (err_codes_unix), string/stratcliff.c (main):
Fix spelling in message.
* po/libc.pot: Fix spelling in message for `zic'; this anticipates
a fix in the tzcode distribution.
Wed Dec 18 15:48:02 1996 Ulrich Drepper <drepper@cygnus.com>
* time/strftime.c: Implement ^ flag to cause output be converted
to use upper case characters.
* time/zic.c: Update from ADO tzcode1996n.
Wed Dec 18 14:29:24 1996 Erik Naggum <erik@naggum.no>
* time/strftime.c (add): Don't change global `i' until all is over.
Define NULL is not already defined.
Tue Dec 17 09:49:03 1996 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de>
* libio/iovsprintf.c (_IO_vsprintf): Change `&sf' to `&sf._sbf._f'
to avoid the need for a cast.
* libio/iovsscanf.c (_IO_vsscanf): Likewise.
* sunrpc/rpc/xdr.h: Add prototype for xdr_free.
