If multiple threads access a directory that is not yet in the cache, the
directory will be loaded by each thread. Only one of the results is added
to the cache, all others are leaked. This wastes performance and memory.
On cache miss, add a future object to the cache to indicate that the
directory is currently being loaded. Subsequent threads register themselves
with the future object and wait. When the first thread has loaded the
directory, it replaces the future object with the result and notifies
waiting threads.
Signed-off-by: Karsten Blees <blees@dcon.de>
Checking the work tree status is quite slow on Windows, due to slow lstat
emulation (git calls lstat once for each file in the index). Windows
operating system APIs seem to be much better at scanning the status
of entire directories than checking single files.
Add an lstat implementation that uses a cache for lstat data. Cache misses
read the entire parent directory and add it to the cache. Subsequent lstat
calls for the same directory are served directly from the cache.
Also implement opendir / readdir / closedir so that they create and use
directory listings in the cache.
The cache doesn't track file system changes and doesn't plug into any
modifying file APIs, so it has to be explicitly enabled for git functions
that don't modify the working copy.
Note: in an earlier version of this patch, the cache was always active and
tracked file system changes via ReadDirectoryChangesW. However, this was
much more complex and had negative impact on the performance of modifying
git commands such as 'git checkout'.
Signed-off-by: Karsten Blees <blees@dcon.de>
Add a macro to mark code sections that only read from the file system,
along with a config option and documentation.
This facilitates implementation of relatively simple file system level
caches without the need to synchronize with the file system.
Enable read-only sections for 'git status' and preload_index.
Signed-off-by: Karsten Blees <blees@dcon.de>
Emulating the POSIX lstat API on Windows via GetFileAttributes[Ex] is quite
slow. Windows operating system APIs seem to be much better at scanning the
status of entire directories than checking single files. A caching
implementation may improve performance by bulk-reading entire directories
or reusing data obtained via opendir / readdir.
Make the lstat implementation pluggable so that it can be switched at
runtime, e.g. based on a config option.
Signed-off-by: Karsten Blees <blees@dcon.de>
Emulating the POSIX dirent API on Windows via FindFirstFile/FindNextFile is
pretty staightforward, however, most of the information provided in the
WIN32_FIND_DATA structure is thrown away in the process. A more
sophisticated implementation may cache this data, e.g. for later reuse in
calls to lstat.
Make the dirent implementation pluggable so that it can be switched at
runtime, e.g. based on a config option.
Define a base DIR structure with pointers to readdir/closedir that match
the opendir implementation (i.e. similar to vtable pointers in OOP).
Define readdir/closedir so that they call the function pointers in the DIR
structure. This allows to choose the opendir implementation on a
call-by-call basis.
Move the fixed sized dirent.d_name buffer to the dirent-specific DIR
structure, as d_name may be implementation specific (e.g. a caching
implementation may just set d_name to point into the cache instead of
copying the entire file name string).
Signed-off-by: Karsten Blees <blees@dcon.de>
Git for Windows ships with its own Perl interpreter, and insists on
using it, so it will most likely wreak havoc if PERL5LIB is set before
launching Git.
Let's just unset that environment variables when spawning processes.
To make this feature extensible (and overrideable), there is a new
config setting `core.unsetenvvars` that allows specifying a
comma-separated list of names to unset before spawning processes.
Reported by Gabriel Fuhrmann.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
In the Git for Windows project, we have ample precendent for config
settings that apply to Windows, and to Windows only.
Let's formalize this concept by introducing a platform_core_config()
function that can be #define'd in a platform-specific manner.
This will allow us to contain platform-specific code better, as the
corresponding variables no longer need to be exported so that they can
be defined in environment.c and be set in config.c
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
The interaction of "url.*.insteadOf" and custom URL scheme's
whitelisting is now documented better.
* jk/url-insteadof-config:
docs/config: mention protocol implications of url.insteadOf
The "collision detecting" SHA-1 implementation shipped with 2.13
was quite broken on some big-endian platforms and/or platforms that
do not like unaligned fetches. Update to the upstream code which
has already fixed these issues.
* ab/sha1dc-maint:
sha1dc: update from upstream
"foo\bar\baz" in "git fetch foo\bar\baz", even though there is no
slashes in it, cannot be a nickname for a remote on Windows, as
that is likely to be a pathname on a local filesystem.
* js/bs-is-a-dir-sep-on-windows:
Windows: do not treat a path with backslashes as a remote's nick name
mingw.h: permit arguments with side effects for is_dir_sep
A test allowed both "git push" and "git receive-pack" on the other
end write their traces into the same file. This is OK on platforms
that allows atomically appending to a file opened with O_APPEND,
but on other platforms led to a mangled output, causing
intermittent test failures. This has been fixed by disabling
traces from "receive-pack" in the test.
* jk/alternate-ref-optim:
t5400: avoid concurrent writes into a trace file
"git interpret-trailers", when used as GIT_EDITOR for "git commit
-v", looked for and appended to a trailer block at the very end,
i.e. at the end of the "diff" output. The command has been
corrected to pay attention to the cut-mark line "commit -v" adds to
the buffer---the real trailer block should appear just before it.
* bm/interpret-trailers-cut-line-is-eom:
interpret-trailers: honor the cut line
"git for-each-ref --format=..." with %(HEAD) in the format used to
resolve the HEAD symref as many times as it had processed refs,
which was wasteful, and "git branch" shared the same problem.
* kn/ref-filter-branch-list:
ref-filter: resolve HEAD when parsing %(HEAD) atom
A few codepaths in "checkout" and "am" working on an unborn branch
tried to access an uninitialized piece of memory.
* rs/checkout-am-fix-unborn:
am: check return value of resolve_refdup before using hash
checkout: check return value of resolve_refdup before using hash
"git clone --config var=val" is a way to populate the
per-repository configuration file of the new repository, but it did
not work well when val is an empty string. This has been fixed.
* jn/clone-add-empty-config-from-command-line:
clone: handle empty config values in -c
Update the C style recommendation for notes for translators, as
recent versions of gettext tools can work with our style of
multi-line comments.
* ab/c-translators-comment-style:
C style: use standard style for "TRANSLATORS" comments
Travis CI gained a task to format the documentation with both
AsciiDoc and AsciiDoctor.
* ls/travis-doc-asciidoctor:
travis-ci: check AsciiDoc/AsciiDoctor stderr output
travis-ci: unset compiler for jobs that do not need one
travis-ci: parallelize documentation build
travis-ci: build documentation with AsciiDoc and Asciidoctor
"git cherry-pick" and other uses of the sequencer machinery
mishandled a trailer block whose last line is an incomplete line.
This has been fixed so that an additional sign-off etc. are added
after completing the existing incomplete line.
* jt/use-trailer-api-in-commands:
sequencer: add newline before adding footers
The receive-pack program now makes sure that the push certificate
records the same set of push options used for pushing.
* jt/push-options-doc:
receive-pack: verify push options in cert
docs: correct receive.advertisePushOptions default
Make sure our tests would pass when the sources are checked out
with "platform native" line ending convention by default on
Windows. Some "text" files out tests use and the test scripts
themselves that are meant to be run with /bin/sh, ought to be
checked out with eol=LF even on Windows.
* js/eol-on-ourselves:
t4051: mark supporting files as requiring LF-only line endings
Fix the remaining tests that failed with core.autocrlf=true
t3901: move supporting files into t/t3901/
completion: mark bash script as LF-only
git-new-workdir: mark script as LF-only
Fix build with core.autocrlf=true
A few http:// links that are redirected to https:// in the
documentation have been updated to https:// links.
* jk/update-links-in-docs:
doc: use https links to Wikipedia to avoid http redirects
Tag objects, which are not reachable from any ref, that point at
missing objects were mishandled by "git gc" and friends (they
should silently be ignored instead)
* jk/ignore-broken-tags-when-ignoring-missing-links:
revision.c: ignore broken tags with ignore_missing_links
Clarify documentation for include.path and includeIf.<condition>.path
configuration variables.
* jk/doc-config-include:
docs/config: consistify include.path examples
docs/config: avoid the term "expand" for includes
docs/config: give a relative includeIf example
docs/config: clarify include/includeIf relationship
"pack-objects" can stream a slice of an existing packfile out when
the pack bitmap can tell that the reachable objects are all needed
in the output, without inspecting individual objects. This
strategy however would not work well when "--local" and other
options are in use, and need to be disabled.
* jk/disable-pack-reuse-when-broken:
t5310: fix "; do" style
pack-objects: disable pack reuse for object-selection options
The codepath in "git am" that is used when running "git rebase"
leaked memory held for the log message of the commits being rebased.
* jk/am-leakfix:
am: shorten ident_split variable name in get_commit_info()
am: simplify allocations in get_commit_info()
am: fix commit buffer leak in get_commit_info()
"git read-tree -m" (no tree-ish) gave a nonsense suggestion "use
--empty if you want to clear the index". With "-m", such a request
will still fail anyway, as you'd need to name at least one tree-ish
to be merged.
* jc/read-tree-empty-with-m:
read-tree: "read-tree -m --empty" does not make sense
Git sometimes gives an advice in a rhetorical question that does
not require an answer, which can confuse new users and non native
speakers. Attempt to rephrase them.
* ja/do-not-ask-needless-questions:
git-filter-branch: be more direct in an error message
read-tree -m: make error message for merging 0 trees less smart aleck
usability: don't ask questions if no reply is required
Setting "log.decorate=false" in the configuration file did not take
effect in v2.13, which has been corrected.
* ah/log-decorate-default-to-auto:
builtin/log: honor log.decorate
