So I was looking at the disgusting size of the git binary, and even with
the debugging removed, and using -Os instead of -O2, the size of the text
section was pretty high. In this day and age I guess almost a megabyte of
text isn't really all that surprising, but it still doesn't exactly make
me think "lean and mean".
With -Os, a surprising amount of text space is wasted on inline functions
that end up just being replicated multiple times, and where performance
really isn't a valid reason to inline them. In particular, the trivial
wrapper functions like "xmalloc()" are used _everywhere_, and making them
inline just duplicates the text (and the string we use to 'die()' on
failure) unnecessarily.
So this just moves them into a "wrapper.c" file, getting rid of a tiny bit
of unnecessary bloat. The following numbers are both with "CFLAGS=-Os":
Before:
[torvalds@woody git]$ size git
text data bss dec hex filename
700460 15160 292184 1007804 f60bc git
After:
[torvalds@woody git]$ size git
text data bss dec hex filename
670540 15160 292184 977884 eebdc git
so it saves almost 30k of text-space (it actually saves more than that
with the default -O2, but I don't think that's necessarily a very relevant
number from a "try to shrink git" standpoint).
It might conceivably have a performance impact, but none of this should be
_that_ performance critical. The real cost is not generally in the wrapper
anyway, but in the code it wraps (ie the cost of "xread()" is all in the
read itself, not in the trivial wrapping of it).
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This patch adds support to compile and run git on 12 additional platforms.
The platforms are based on UNIX Systems Labs (USL)/Novell/SYS V code base.
The most common are Novell UnixWare 2.X.X, SCO UnixWare 7.X.X,
OpenServer 5.0.X, OpenServer 6.0.X, and SCO pre OSR 5 platforms.
Looking at the the various platform headers, I find:
#if defined(_KERNEL) || !defined(_POSIX_SOURCE) \
&& !defined(_POSIX_C_SOURCE) && !defined(_XOPEN_SOURCE)
which hides u_short and other typedefs that other header files on these
platforms depend on. WIth _XOPEN_SOURCE defined, sources that include
system header files that depend on the typedefs such as u_short cannot be
compiled on these platforms.
__USLC__ indicates UNIX System Labs Corperation (USLC), or a Novell-derived
compiler and/or some SysV based OS's.
__M_UNIX indicates XENIX/SCO UNIX/OpenServer 5.0.7 and prior releases
of the SCO OS's. It is used just like Apple and BSD, both of these
shouldn't have _XOPEN_SOURCE defined.
This is with suggestions and modifications from
Daniel Barkalow, Junio C Hamano, Thomas Harning, and Jeremy Maitin-Shepard.
Signed-off-by: Boyd Lynn Gerber <gerberb@zenez.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Runtime pack access is done in the pack file mtime order since recent
packs are more likely to contain frequently used objects than old packs.
However the --max-pack-size option can produce multiple packs with mtime
in the reversed order as newer objects are always written first.
Let's modify mtime of later pack files (when any) so they appear older
than preceding ones when a repack creates multiple packs.
Signed-off-by: Nicolas Pitre <nico@cam.org>
Some systems (namely HPUX and Windows) return -1 when maxsize in snprintf()
and in vsnprintf() is reached. So replace snprintf() and vsnprintf()
functions with our own ones that return correct value upon overflow.
[jc: verified that review comments by J6t have been incorporated, and
tightened the check to verify the resulting buffer contents, suggested
by Wayne Davison]
Signed-off-by: Michal Rokos <michal.rokos@nextsoft.cz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It does not allow changing the bit to a non-root user.
This fixes t1301-shared-repo.sh on the platform.
Signed-off-by: Alex Riesen <raa.lkml@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Some systems do not fail as expected when fread et al. are called on
a directory stream. Replace fopen on such systems which will fail
when the supplied path is a directory.
Signed-off-by: Brandon Casey <casey@nrlssc.navy.mil>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
qsort in Windows 2000 (and various other C libraries) is a Quicksort
with the usual O(n^2) worst case. Unfortunately, sorting Git trees
seems to get very close to that worst case quite often:
$ /git/gitbad runstatus
# On branch master
qsort, nmemb = 30842
done, 237838087 comparisons.
This patch adds a simplified version of the merge sort that is glibc's
qsort(3). As a merge sort, this needs a temporary array equal in size
to the array that is to be sorted, but has a worst-case performance of
O(n log n).
The complexity that was removed is:
* Doing direct stores for word-size and -aligned data.
* Falling back to quicksort if the allocation required to perform the
merge sort would likely push the machine into swap.
Even with these simplifications, this seems to outperform the Windows
qsort(3) implementation, even in Windows XP (where it is "fixed" and
doesn't trigger O(n^2) complexity on trees).
[jes: moved into compat/qsort.c, as per Johannes Sixt's suggestion]
[bcd: removed gcc-ism, thanks to Edgar Toernig. renamed make variable
per Junio's comment.]
Signed-off-by: Brian Downing <bdowning@lavos.net>
Signed-off-by: Steffen Prohaska <prohaska@zib.de>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
POSIX.1-2001 has declaration of select(2) in <sys/select.h>, but
in the previous version of SUS, it was declared in <sys/time.h>
(which is already included in git-compat-util.h).
This introduces NO_SYS_SELECT_H macro in the Makefile to be set
on older systems, to skip inclusion of <sys/select.h> that does
not exist on them.
We could check _POSIX_VERSION with 200112L and do this
automatically, but earlier it was reported that the approach
does not work well on some vintage of HP-UX. Other systems may
get _POSIX_VERSION itself wrong. At least for now, this manual
configuration is safer.
Signed-off-by: Robert Schiele <rschiele@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Now the routine is an open-coded loop that avoids an extra
strlen() in the previous implementation, it got a bit too big to
be inlined. Uninlining it makes code footprint smaller but the
result still retains the avoidance of strlen() cost.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Certain codepaths (notably "git log --pretty=format...") use
prefixcmp() extensively, with very short prefixes. In those cases,
calling strlen() is a wasteful operation, so avoid it.
Initial patch by Marco Costalba.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When declaring a structure with a flexible array member, instead
of defaulting to the c99 syntax for non-gnu compilers (which
burned people with older compilers), default to the traditional
and more portable "member[1]; /* more */" syntax.
At the same time, other c99 compilers should be able to take
advantage of the modern syntax to flexible array members without
being gcc. Check __STDC_VERSION__ for that.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
... since all system headers are pulled in via git-compat-util.h
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Apart from the error in the condition (&& should actually be ||), the
construct
#if !defined(A) || !A
leads to a syntax error in the C preprocessor if A is indeed not defined.
Tested-by: David Symonds <dsymonds@gmail.com>
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
options flags:
~~~~~~~~~~~~~
PARSE_OPT_NONEG allow the caller to disallow the negated option to exists.
option types:
~~~~~~~~~~~~
OPTION_BIT: ORs (or NANDs) a mask.
OPTION_SET_INT: force the value to be set to this integer.
OPTION_SET_PTR: force the value to be set to this pointer.
helper:
~~~~~~
HAS_MULTI_BITS (in git-compat-util.h) is a bit-hack to check if an
unsigned integer has more than one bit set, useful to check if conflicting
options have been used.
Signed-off-by: Pierre Habouzit <madcoder@debian.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
strchrnul() was introduced in glibc in April 1999 and included in
glibc-2.1. Checking for that version means the majority of all git
users would get to use the optimized version in glibc. Of the
remaining few some might get to use a slightly slower version
than necessary but probably not slower than what we have today.
Unfortunately, __GLIBC_PREREQ() macro was not available in glibc 2.1.1
which was short lived but already supported strchrnul(). Odd minority
users of that library needs to live with our compatibility inline version.
Rediffed-against-next-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
As suggested by Pierre Habouzit, add strchrnul(). It's a useful GNU
extension and can simplify string parser code. There are several
places in git that can be converted to strchrnul(); as a trivial
example, this patch introduces its usage to builtin-fetch--tool.c.
Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Usually you cannot revert a merge because you do not know which
side of the merge should be considered the mainline (iow, what
change to reverse).
With this patch, cherry-pick and revert learn -m (--mainline)
option that lets you specify the parent number (starting from 1)
of the mainline, so that you can:
git revert -m 1 $merge
to reverse the changes introduced by the $merge commit relative
to its first parent, and:
git cherry-pick -m 2 $merge
to replay the changes introduced by the $merge commit relative
to its second parent.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Solaris 9 doesn't have mkdtemp() so we need to emulate it for the
rsync transport implementation. Since Solaris 9 is lacking this
function we can also reasonably assume it is not available on
Solaris 8 either. The new Makfile definition NO_MKDTEMP can be
set to enable the git compat version.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
A lot of places in git's code use code like:
char *res;
len = ... find length of an interesting segment in src ...;
res = xmalloc(len + 1);
memcpy(res, src, len);
res[len] = '\0';
return res;
A new function xmemdupz() captures the allocation, copy and NUL
termination. Existing xstrndup() is reimplemented in terms of
this new function.
Signed-off-by: Pierre Habouzit <madcoder@debian.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
memmem() is a nice GNU extension for searching a length limited string
in another one.
This compat version is based on the version found in glibc 2.2 (GPL 2);
I only removed the optimization of checking the first char by hand, and
generally tried to keep the code simple. We can add it back if memcmp
shows up high in a profile, but for now I prefer to keep it (almost
trivially) simple.
Since I don't really know which platforms beside those with a glibc
have their own memmem(), I used a heuristic: if NO_STRCASESTR is set,
then NO_MEMMEM is set, too.
Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is a wrapper for mkstemp() that performs error checking and
calls die() when an error occur.
Signed-off-by: Luiz Fernando N. Capitulino <lcapitulino@mandriva.com.br>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This defines xdup() and xfdopen() in git-compat-util.h to give
us error-catching variants of them without cluttering the code
too much.
Signed-off-by: Jim Meyering <jim@meyering.net>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The function converts the value of h_errno (last error of name
resolver library, see netdb.h).
One of systems which supposedly do not have the function is SunOS.
POSIX does not mandate its presence.
Signed-off-by: Alex Riesen <raa.lkml@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When using git name-rev on my kernel tree I triggered a malloc()
corruption warning from glibc.
apw@pinky$ git log --pretty=one $N/base.. | git name-rev --stdin
*** glibc detected *** malloc(): memory corruption: 0x0bff8950 ***
Aborted
This comes from name_rev() which is building the name of the revision
in a malloc'd string, which it sprintf's into:
char *new_name = xmalloc(len + 8);
[...]
sprintf(new_name, "%.*s~%d^%d", len, tip_name,
generation, parent_number);
This allocation is only sufficient if the generation number is
less than 5 digits, in my case generation was 13432. In reality
parent_number can be up to 16 so that also can require two digits,
reducing us to 3 digits before we are at risk of blowing this
allocation.
This patch introduces a decimal_length() which approximates the
number of digits a type may hold, it produces the following:
Type Longest Value Len Est
---- ------------- --- ---
unsigned char 256 3 4
unsigned short 65536 5 6
unsigned long 4294967296 10 11
unsigned long long 18446744073709551616 20 21
char -128 4 4
short -32768 6 6
long -2147483648 11 11
long long -9223372036854775808 20 21
This is then used to size the new_name.
Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This also improves the implementation to match how strndup is
specified (by GNU): if the length given is longer than the string,
only the string's length is allocated and copied, but the string need
not be null-terminated if it is at least as long as the given length.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Tim Ansell discovered his Debian server didn't permit git-daemon to
use as much memory as it needed to handle cloning a project with
a 128 MiB packfile. Filtering the strace provided by Tim of the
rev-list child showed this gem of a sequence:
open("./objects/pack/pack-*.pack", O_RDONLY|O_LARGEFILE <unfinished ...>
<... open resumed> ) = 5
OK, so the packfile is fd 5...
mmap2(NULL, 33554432, PROT_READ, MAP_PRIVATE, 5, 0 <unfinished ...>
<... mmap2 resumed> ) = 0xb5e2d000
and we mapped one 32 MiB window from it at position 0...
mmap2(NULL, 31020635, PROT_READ, MAP_PRIVATE, 5, 0x6000 <unfinished ...>
<... mmap2 resumed> ) = -1 ENOMEM (Cannot allocate memory)
And we asked for another window further into the file. But got
denied. In Tim's case this was due to a resource limit on the
git-daemon process, and its children.
Now where are we in the code? We're down inside use_pack(),
after we have called unuse_one_window() enough times to make sure
we stay within our allowed maximum window size. However since we
didn't unmap the prior window at 0xb5e2d000 we aren't exceeding
the current limit (which probably was just the defaults).
But we're actually down inside xmmap()...
So we release the window we do have (by calling release_pack_memory),
assuming there is some memory pressure...
munmap(0xb5e2d000, 33554432 <unfinished ...>
<... munmap resumed> ) = 0
close(5 <unfinished ...>
<... close resumed> ) = 0
And that was the last window in this packfile. So we closed it.
Way to go us. Our xmmap did not expect release_pack_memory to
close the fd its about to map...
mmap2(NULL, 31020635, PROT_READ, MAP_PRIVATE, 5, 0x6000 <unfinished ...>
<... mmap2 resumed> ) = -1 EBADF (Bad file descriptor)
And so the Linux kernel happily tells us f' off.
write(2, "fatal: ", 7 <unfinished ...>
<... write resumed> ) = 7
write(2, "Out of memory? mmap failed: Bad "..., 47 <unfinished ...>
<... write resumed> ) = 47
And we report the bad file descriptor error, and not the ENOMEM,
and die, claiming we are out of memory. But actually that mmap
should have succeeded, as we had enough memory for that window,
seeing as how we released the prior one.
Originally when I developed the sliding window mmap feature I had
this exact same bug in fast-import, and I dealt with it by handing
in the struct packed_git* we want to open the new window for, as the
caller wasn't prepared to reopen the packfile if unuse_one_window
closed it. The same is true here from xmmap, but the caller doesn't
have the struct packed_git* handy. So I'm using the file descriptor
instead to perform the same test.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
* builtin-grep.c (strtoul_ui): Move function definition from here, to...
* git-compat-util.h (strtoul_ui): ...here, with an added "base" parameter.
* builtin-grep.c (cmd_grep): Update use of strtoul_ui to include base, "10".
* builtin-update-index.c (read_index_info): Diagnose an invalid mode integer
that is out of range or merely larger than INT_MAX.
(cmd_update_index): Use strtoul_ui, not sscanf.
* convert-objects.c (write_subdirectory): Likewise.
Signed-off-by: Jim Meyering <jim@meyering.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
* builtin-grep.c (strtoul_ui): Move function definition from here, to...
* git-compat-util.h (strtoul_ui): ...here, with an added "base" parameter.
* builtin-grep.c (cmd_grep): Update use of strtoul_ui to include base, "10".
* builtin-update-index.c (read_index_info): Diagnose an invalid mode integer
that is out of range or merely larger than INT_MAX.
(cmd_update_index): Use strtoul_ui, not sscanf.
* convert-objects.c (write_subdirectory): Likewise.
Signed-off-by: Jim Meyering <jim@meyering.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This patch introduces the MSB() macro to obtain the desired number of
most significant bits from a given variable independently of the variable
type.
It is then used to better implement the overflow test on the OBJ_OFS_DELTA
base offset variable with the property of always working correctly
regardless of the type/size of that variable.
Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This fixes a problem reported by Randal Schwartz:
>I finally tracked down all the (albeit inconsequential) errors I was getting
>on both OpenBSD and OSX. It's the warn() function in usage.c. There's
>warn(3) in BSD-style distros. It'd take a "great rename" to change it, but if
>someone with better C skills than I have could do that, my linker and I would
>appreciate it.
It was annoying to me, too, when I was doing some mergetool testing on
Mac OS X, so here's a fix.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: "Randal L. Schwartz" <merlyn@stonehenge.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Some systems have sizeof(off_t) == 8 while sizeof(size_t) == 4.
This implies that we are able to access and work on files whose
maximum length is around 2^63-1 bytes, but we can only malloc or
mmap somewhat less than 2^32-1 bytes of memory.
On such a system an implicit conversion of off_t to size_t can cause
the size_t to wrap, resulting in unexpected and exciting behavior.
Right now we are working around all gcc warnings generated by the
-Wshorten-64-to-32 option by passing the off_t through xsize_t().
In the future we should make xsize_t on such problematic platforms
detect the wrapping and die if such a file is accessed.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Not all platforms have declared 'unsigned long' to be a 64 bit value,
but we want to support a 64 bit packfile (or close enough anyway)
in the near future as some projects are getting large enough that
their packed size exceeds 4 GiB.
By using off_t, the POSIX type that is declared to mean an offset
within a file, we support whatever maximum file size the underlying
operating system will handle. For most modern systems this is up
around 2^60 or higher.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
The function at issue being initgroups() from the <grp.h> header
file. On Cygwin, setting _XOPEN_SOURCE suppresses the definition
of initgroups(), which causes the warning while compiling daemon.c.
Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Glibc uses the same size for int and off_t by default.
In order to support large pack sizes (>2GB) we force Glibc to a 64bit off_t.
Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Solaris 8 was pre-c99, and they weren't willing to commit to
the strtoumax definition according to /usr/include/inttypes.h.
This adds NO_STRTOUMAX and NO_STRTOULL for ancient systems.
If NO_STRTOUMAX is defined, the routine in compat/strtoumax.c
will be used instead. That routine passes its arguments to
strtoull unless NO_STRTOULL is defined. If NO_STRTOULL, then
the routine uses strtoul (unsigned long).
Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu>
Acked-by: Shawn O Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Older Solaris machines lack stdint.h but have inttypes.h.
The standard has inttypes.h including stdint.h, so at worst
this pollutes the namespace a bit.
Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Plain integer types without a fixed size can vary between platforms. Even
though all common platforms use 32-bit ints, there is no guarantee that
this won't change at some point. Furthermore, specifying an integer type
with explicit size makes the definition of structures more obvious.
Signed-off-by: Simon 'corecode' Schubert <corecode@fs.ei.tum.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Currently the pack .idx file format uses 32-bit unsigned integers
for the fan-out table and the object offsets. We had previously
defined these as 'unsigned int', but not every system will define
that type to be a 32 bit value. To ensure maximum portability we
should always use 'uint32_t'.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
AIX 5.3 seems to need _ALL_SOURCE for struct addrinfo, but that
introduces a struct list in grp.h.
Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
If a frontend wants to use a mark per file revision and per commit
and is doing a truly huge import (such as a 32 GiB SVN repository)
we may need more than 2**32 unique mark values, especially if the
frontend is unable (or unwilling) to recycle mark values. For mark
idnums we should use the largest unsigned integer type available,
hoping that will be at least 64 bits when we are compiled as a 64
bit executable. This way we may consume huge amounts of memory
storing our mark table, but we'll at least be able to process
the entire import without failing.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This fixes another problem that Andy's case showed: git-fsck-objects
reports nonsensical results for corrupt objects.
There were actually two independent and confusing problems:
- when we had a zero-sized file and used map_sha1_file, mmap() would
return EINVAL, and git-fsck-objects would report that as an insane and
confusing error. I don't know when this was introduced, it might have
been there forever.
- when "parse_object()" returned NULL, fsck would say "object not found",
which can be very confusing, since obviously the object might "exist",
it's just unparseable because it's totally corrupt.
So this just makes "xmmap()" return NULL for a zero-sized object (which is
a valid thing pointer, exactly the same way "malloc()" can return NULL for
a zero-sized allocation). That fixes the first problem (but we could have
fixed it in the caller too - I don't personally much care whichever way it
goes, but maybe somebody should check that the NO_MMAP case does
something sane in this case too?).
And the second problem is solved by just making the error message slightly
clearer - the failure to parse an object may be because it's missing or
corrupt, not necessarily because it's not "found".
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Using cygwin with cygwin.dll before 1.5.22 the system call pread() is buggy.
This patch introduces NO_PREAD. If NO_PREAD is set git uses a sequence of
lseek()/xread()/lseek() to emulate pread.
Signed-off-by: Stefan-W. Hahn <stefan.hahn@s-hahn.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This is shorter and easier to read, and also makes sure the
constant expression does not overflow integer range.
Signed-off-by: Junio C Hamano <junkio@cox.net>
If we have a 64 bit address space we can easily afford to commit
a larger amount of virtual address space to pack file access.
So on these platforms we should increase the default settings of
core.packedGit{Limit,WindowSize} to something that will better
handle very large projects.
Thanks to Andy Whitcroft for pointing out that we can safely
increase these defaults on such systems.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
In some cases we did not even bother to check the return value of
mmap() and just assume it worked. This is bad, because if we are
out of virtual address space the kernel returned MAP_FAILED and we
would attempt to dereference that address, segfaulting without any
real error output to the user.
We are replacing all calls to mmap() with xmmap() and moving all
MAP_FAILED checking into that single location. If a mmap call
fails we try to release enough least-recently-used pack windows
to possibly succeed, then retry the mmap() attempt. If we cannot
mmap even after releasing pack memory then we die() as none of our
callers have any reasonable recovery strategy for a failed mmap.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
If we are about to fail because this process has run out of memory we
should first try to automatically control our appetite for address
space by releasing enough least-recently-used pack windows to gain
back enough memory such that we might actually be able to meet the
current allocation request.
This should help users who have fairly large repositories but are
working on systems with relatively small virtual address space.
Many times we see reports on the mailing list of these users running
out of memory during various Git operations. Dynamically decreasing
the amount of pack memory used when the demand for heap memory is
increasing is an intelligent solution to this problem.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Linus Torvalds