Teach do_write_index() to close the index.lock file
before getting the mtime and updating the istate.timestamp
fields.
On Windows, a file's mtime is not updated until the file is
closed. On Linux, the mtime is set after the last flush.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Add check for the end of the entries for the thread partition.
Add test for lazy init name hash with specific directory structure
The lazy init hash name was causing a buffer overflow when the last
entry in the index was multiple folder deep with parent folders that
did not have any files in them.
This adds a test for the boundary condition of the thread partitions
with the folder structure that was triggering the buffer overflow.
The fix was to check if it is the last entry for the thread partition
in the handle_range_dir and not try to use the next entry in the cache.
Signed-off-by: Kevin Willford <kewillf@microsoft.com>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The name-hash used for detecting paths that are different only in
cases (which matter on case insensitive filesystems) has been
optimized to take advantage of multi-threading when it makes sense.
* jh/memihash-opt:
name-hash: add test-lazy-init-name-hash to .gitignore
name-hash: add perf test for lazy_init_name_hash
name-hash: add test-lazy-init-name-hash
name-hash: perf improvement for lazy_init_name_hash
hashmap: document memihash_cont, hashmap_disallow_rehash api
hashmap: add disallow_rehash setting
hashmap: allow memihash computation to be continued
name-hash: specify initial size for istate.dir_hash table
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This topic branch allows us to skip the gettext initialization
when the locale directory does not even exist.
This saves 150ms out of 210ms for a simply `git version` call on
Windows, and it most likely will help scripts that call out to
`git.exe` hundreds of times.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Created t/perf/p0004-lazy-init-name-hash.sh test
to demonstrate correctness and performance gains
with the multithreaded version of lazy_init_name_hash().
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add t/helper/test-lazy-init-name-hash.c test code
to demonstrate performance times for lazy_init_name_hash()
using the original single-threaded and the new multi-threaded
code paths.
Includes a --dump option to dump the created hashmaps to
stdout. You can use this to run both code paths and
confirm that they generate the same hashmaps.
Includes a --analyze option to analyze performance of both
code paths over a range of index sizes to help you find a
lower bound for the LAZY_THREAD_COST in name-hash.c.
For example, passing "-a 4000" will set "istate.cache_nr"
to 4000 and then try the multi-threaded code -- probably
giving 2 threads with 2000 entries each. It will then
run both the single-threaded (1x4000) and the multi-threaded
(2x2000) and compare the times. It will then repeat the
test with 8000, 12000, and etc. so that you can see the
cross over.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Improve performance of lazy_init_name_hash() when
ignore_case is set. Teach name-hash to build the
istate.name_hash and istate.dir_hash simultaneously
using a forward-diving technique on the pathname
of the index_entry, rather than adding name_hash
entries and then searching backwards in the pathname
for parent directories.
This borrows algorithm ideas from clear_ce_flags_{1,dir}.
Multiple threads are used with the new algorithm to
speed hashmap construction.
This new code path is only used when threads are present
(a compiler settings) and when the index is large enough
to warrant the pthread complexity.
The code in clear_ce_flags_dir() uses a linear search to
find the adjacent index entries with the same prefix; a
binary search is used here handle_range_dir() to further
speed things up.
The size of LAZY_THREAD_COST was determined from rough
analysis using:
t/helper/test-lazy-init-name-hash --analyze
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Document memihash_cont() and hashmap_disallow_rehash()
in Documentation/technical/api-hashmap.txt.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Teach hashmap to allow rehashes to be suppressed.
This is useful when hashmaps are accessed by multiple
threads. It still requires the caller to properly
manage their locking. This just prevents unexpected
rehashing during inserts and deletes.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add variant of memihash() to allow the hash computation to
be continued. There are times when we compute the hash on
a full path and then the hash on just the path to the parent
directory. This can be expensive on large repositories.
With this, we can hash the parent directory first. And then
continue the computation to include the "/filename".
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Specify an initial size for the istate.dir_hash HashMap matching
the size of the istate.name_hash.
Previously hashmap_init() was given 0, causing a 64 bucket
hashmap to be created. When working with very large
repositories, this would cause numerous rehash() calls to
realloc and rebalance the hashmap. This is especially true
when the worktree is deep, with many directories containing
a few files.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The runtime of a simple `git.exe version` call on Windows is currently
dominated by the gettext setup, adding a whopping ~150ms to the ~210ms
total.
Given that this cost is added to each and every git.exe invocation goes
through common-main's invocation of git_setup_gettext(), and given that
scripts have to call git.exe dozens, if not hundreds, of times, this is
a substantial performance penalty.
This is particularly pointless when considering that Git for Windows
ships without localization (to keep the installer's size to a bearable
~34MB): all that time setting up gettext is for naught.
So let's be smart about it and skip setting up gettext if the locale
directory is not even present.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
On Windows, we simply pass a POSIX path to bindtextdomain(), relying on
the current libintl-8.dll implementation to handle that gracefully by
resolving the path relative to the "root" directory inferred from the
location of the .dll file itself.
However, not only does this rely on the custom patches of the gettext
library as shipped with MSYS2 (gettext's own source code is not prepared
to handle POSIX paths on Windows), it also means that Git itself cannot
use the `podir` variable at all because it does not handle absolute
POSIX paths in system_path() correctly, leaving them as-is.
This patch fixes that behavior by always using a GIT_LOCALE_PATH
relative to the (runtime) prefix.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
On Windows, there is no single root directory. And what Git thinks is a
root directory is not a root directory at all: everything is relative to
the location where Git is installed.
To handle this situation better, let's just allow for GIT_TEXTDOMAINDIR
to be a path relative to the (runtime) prefix.
To that end, we have to switch the order in which common-main handles
argv0 and sets up gettext: in order to have access to the runtime
prefix, we need it to be inferred from argv0 already.
This patch also prepares for GIT_LOCALE_PATH to be relative to prefix,
which is the even more important fix.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
We newly handle isatty() by special-casing the stdin/stdout/stderr file
descriptors, caching the return value. However, we missed the case where
dup2() overrides the respective file descriptor.
That poses a problem e.g. where the `show` builtin asks for a pager very
early, the `setup_pager()` function sets the pager depending on the
return value of `isatty()` and then redirects stdout. Subsequently,
`cmd_log_init_finish()` calls `setup_pager()` *again*. What should
happen now is that `isatty()` reports that stdout is *not* a TTY and
consequently stdout should be left alone.
Let's override dup2() to handle this appropriately.
This fixes https://github.com/git-for-windows/git/issues/1077
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This does not (yet) resolve the problem that Strg+T with multiple
selected lines fails to (un)stage them all, but it addresses one of
the reported Ctrl+T issues.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
The gui.recentrepo list may be longer than the maxrecent setting.
Allow extra space to show any extra entries.
In an ideal world, the git gui would limit the number of entries
to the maxrecent setting, however the recentrepo config list may
have been extended outwith the gui, or the maxrecent setting changed
to a reduced value. Further, when testing the gui's recentrepo
logic it is useful to show these extra, but valid, entries.
Signed-off-by: Philip Oakley <philipoakley@iee.org>
When the gui/user selects a repo for display, that repo is brought to
the end of the recentrepo config list. The logic can fail if there are
duplicate old entries for the repo (you cannot unset a single config
entry when duplicates are present).
Similarly, the maxrecentrepo logic could fail if older duplicate entries
are present.
The first commit of this series ({this}~2) fixed the config unsetting
issue. Rather than manipulating a local copy of the $recent list (one
cannot know how many entries were removed), simply re-read it.
We must also catch the error when the attempt to remove the second copy
from the re-read list is performed.
Signed-off-by: Philip Oakley <philipoakley@iee.org>
_get_recentrepo will fail if duplicate invalid entries are present
in the recentrepo config list. The previous commit fixed the
'git config' limitations in _unset_recentrepo by unsetting all config
entries, however this code would fail on the second attempt to unset it.
Refactor the code to pre-sort and de-duplicate the recentrepo list to
avoid a potential second unset attempt.
Signed-off-by: Philip Oakley <philipoakley@iee.org>
The git gui's recent repo list may become contaminated with duplicate
entries. The git gui would barf when attempting to remove one entry.
Remove them all - there is no option within 'git config' to selectively
remove one of the entries.
This issue was reported on the 'Git User' list
(https://groups.google.com/forum/#!topic/git-users/msev4KsQGFc,
Warning: gui.recentrepo has multiply values while executing).
And also by zosrothko as a Git-for-Windows issue
https://github.com/git-for-windows/git/issues/1014.
On startup the gui checks that entries in the recentrepo list are still
valid repos and deletes thoses that are not. If duplicate entries are
present the 'git config --unset' will barf and this prevents the gui
from starting.
Subsequent patches fix other parts of recentrepo logic used for syncing
internal lists with the external .gitconfig.
Reported-by: Alexey Astakhov <asstv7@gmail.com>
Signed-off-by: Philip Oakley <philipoakley@iee.org>
Previously unstaged files can be staged by clicking on them and then
pressing Ctrl+T. Conveniently, the next unstaged file is selected
automatically so that the unstaged files can be staged by repeatedly
pressing Ctrl+T.
When a user hits Ctrl+T one time too many, though, Git GUI used to throw
this exception:
expected number but got ""
expected number but got ""
while executing
"expr {int([lindex [$w tag ranges in_diff] 0])}"
(procedure "toggle_or_diff" line 13)
invoked from within
"toggle_or_diff toggle .vpane.files.workdir.list "
(command bound to event)
Let's just avoid that by skipping the operation when there are no more
files to stage.
This fixes https://github.com/git-for-windows/git/issues/1060
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
We should not actually expect the first `attrib.exe` in the PATH to
be the one we are looking for. Or that it is in the PATH, for that
matter.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
A change between versions 2.4.1 and 2.6.0 of the MSYS2 runtime modified
how Cygwin's runtime (and hence Git for Windows' MSYS2 runtime
derivative) handles locales: d16a56306d (Consolidate wctomb/mbtowc calls
for POSIX-1.2008, 2016-07-20).
An unintended side-effect is that "cold-calling" into the POSIX
emulation will start with a locale based on the current code page,
something that Git for Windows is very ill-prepared for, as it expects
to be able to pass a command-line containing non-ASCII characters to the
shell without having those characters munged.
One symptom of this behavior: when `git clone` or `git fetch` shell out
to call `git-upload-pack` with a path that contains non-ASCII
characters, the shell tried to interpret the entire command-line
(including command-line parameters) as executable path, which obviously
must fail.
This fixes https://github.com/git-for-windows/git/issues/1036
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
The fix we introduced in Git for Windows will be made obsolete by a more
general fix that has been already accepted into upstream Git's `next`
branch.
But we still can introduce a regression test that verifies that this bug
will be caught very quickly, if reintroduced.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Teach FSCACHE to remember "not found" directories.
This is a performance optimization.
FSCACHE is a performance optimization available for Windows. It
intercepts Posix-style lstat() calls into an in-memory directory
using FindFirst/FindNext. It improves performance on Windows by
catching the first lstat() call in a directory, using FindFirst/
FindNext to read the list of files (and attribute data) for the
entire directory into the cache, and short-cut subsequent lstat()
calls in the same directory. This gives a major performance
boost on Windows.
However, it does not remember "not found" directories. When STATUS
runs and there are missing directories, the lstat() interception
fails to find the parent directory and simply return ENOENT for the
file -- it does not remember that the FindFirst on the directory
failed. Thus subsequent lstat() calls in the same directory, each
re-attempt the FindFirst. This completely defeats any performance
gains.
This can be seen by doing a sparse-checkout on a large repo and
then doing a read-tree to reset the skip-worktree bits and then
running status.
This change reduced status times for my very large repo by 60%.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
git-gui tries to temporary set GIT_DIR for starting gitk and restore
it back after they are started. But in case of GIT_DIR which was not set
prior to invocation it is not unset after it. This affects commands
which can be later started from that git gui, for example "Git Bash".
Fix it.
Signed-off-by: Max Kirillov <max@max630.net>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Just like the workaround we added for t9116, t9001.83 hangs sometimes --
but not always! -- when being run in the Git for Windows SDK.
The issue seems to be related to redirection via a pipe, but it is really
hard to diagnose, what with git.exe (a non-MSYS2 program) calling a Perl
script (which is executed by an MSYS2 Perl), piping into another MSYS2
program.
As hunting time is scarce these days, simply work around this for now and
leave the real diagnosis and resolution for later.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
On Windows, UNC paths are a very convenient way to share data, and
alternates are all about sharing data.
We fixed a bug where alternates specifying UNC paths were not handled
properly, and it is high time that we add a regression test to ensure
that this bug is not reintroduced.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
As of a couple of weeks ago, t9116 hangs sometimes -- but not always! --
when being run in the Git for Windows SDK.
The issue seems to be related to redirection via a pipe, but it is really
hard to diagnose, what with git.exe (a non-MSYS2 program) calling a Perl
script (which is executed by an MSYS2 Perl), piping into another MSYS2
program.
As hunting time is scarce these days, simply work around this for now and
leave the real diagnosis and resolution for later.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This series of branches introduces the git-rebase--helper, a builtin
helping to accelerate the interactive rebase dramatically.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This operation has quadratic complexity, which is especially painful
on Windows, where shell scripts are *already* slow (mainly due to the
overhead of the POSIX emulation layer).
Let's reimplement this with linear complexity (using a hash map to
match the commits' subject lines) for the common case; Sadly, the
fixup/squash feature's design neglected performance considerations,
allowing arbitrary prefixes (read: `fixup! hell` will match the
commit subject `hello world`), which means that we are stuck with
quadratic performance in the worst case.
The reimplemented logic also happens to fix a bug where commented-out
lines (representing empty patches) were dropped by the previous code.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
The `git commit --fixup` command unwraps wrapped onelines when
constructing the commit message, without wrapping the result.
We need to make sure that `git rebase --autosquash` keeps handling such
cases correctly, in particular since we are about to move the autosquash
handling into the rebase--helper.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
In particular on Windows, where shell scripts are even more expensive
than on MacOSX or Linux, it makes sense to move a loop that forks
Git at least once for every line in the todo list into a builtin.
Note: The original code did not try to skip unnecessary picks of root
commits but punts instead (probably --root was not considered common
enough of a use case to bother optimizing). We do the same, for now.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
In particular on Windows, where shell scripts are even more expensive
than on MacOSX or Linux, it makes sense to move a loop that forks
Git at least once for every line in the todo list into a builtin.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
These tests were a bit anal about the *exact* warning/error message
printed by git rebase. But those messages are intended for the *end
user*, therefore it does not make sense to test so rigidly for the
*exact* wording.
In the following, we will reimplement the missing commits check in
the sequencer, with slightly different words.
So let's just test for the parts in the warning/error message that
we *really* care about, nothing more, nothing less.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This is crucial to improve performance on Windows, as the speed is now
mostly dominated by the SHA-1 transformation (because it spawns a new
rev-parse process for *every* line, and spawning processes is pretty
slow from Git for Windows' MSYS2 Bash).
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
To avoid problems with short SHA-1s that become non-unique during the
rebase, we rewrite the todo script with short/long SHA-1s before and
after letting the user edit the script. Since SHA-1s are not intuitive
for humans, rebase -i also provides the onelines (commit message
subjects) in the script, purely for the user's convenience.
It is very possible to generate a todo script via different means than
rebase -i and then to let rebase -i run with it; In this case, these
onelines are not required.
And this is where the expand/collapse machinery has a bug: it *expects*
that oneline, and failing to find one reuses the previous SHA-1 as
"oneline".
It was most likely an oversight, and made implementation in the (quite
limiting) shell script language less convoluted. However, we are about
to reimplement performance-critical parts in C (and due to spawning a
git.exe process for every single line of the todo script, the
expansion/collapsing of the SHA-1s *is* performance-hampering on
Windows), therefore let's fix this bug to make cross-validation with the
C version of that functionality possible.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
