This was originally 'pull request #330 from ethomson/poll_inftim' in
msysgit/git.
poll: honor the timeout on Win32
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
These fixes were necessary for Sverre Rabbelier's remote-hg to work,
but for some magic reason they are not necessary for the current
remote-hg. Makes you wonder how that one gets away with it.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
When calculating hashes from pointers, it actually makes sense to cut
off the most significant bits. In that case, said warning does not make
a whole lot of sense.
So let's just work around it.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
When prefixing a Git call in the test suite with 'TEST_GDB_GIT=1 ', it
will now be run with GDB, allowing the developer to debug test failures
more conveniently.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
With this patch we properly support SOCKS proxies, configured e.g. like
this:
git config http.proxy socks5://192.168.67.1:32767
Without this patch, Git mistakenly tries to use SOCKS proxies as if they
were HTTP proxies, resulting in a error message like:
fatal: unable to access 'http://.../': Proxy CONNECT aborted
This patch was required to work behind a faulty AP and scraped from
http://stackoverflow.com/questions/15227130/#15228479 and guarded with
an appropriate cURL version check by Johannes Schindelin.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This fixes the compilation on an older Linux that was used to debug
test failures when upgrading Git for Windows to Git v2.3.0.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Again, avoid using echo (which issues DOS line endings on msysGit) to not mix
with Unix line-endings issued by git built-ins, even if this is at the cost of
calling an external executable (cat) instead of a shell built-in (echo).
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This happens to shut up t7602 on Windows which would otherwise take
the different line endings for a sign that the merge failed.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Along the lines of 05d0e3b and f33946d, use cat instead of echo to avoid
line ending mismatches in the test result of "am empty-file does not
infloop" which make the test fail.
Signed-off-by: Sebastian Schuberth <sschuberth@gmail.com>
HANDLE is defined internally as a void *, but in many cases it is
actually guaranteed to be a 32-bit integer. In these cases, GCC should
not warn about a cast of a pointer to an integer of a different type
because we know exactly what we are doing.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
When the result of a (a, 0) expression is not used, GCC now finds it
necessary to complain with a warning:
right-hand operand of comma expression has no effect
Let's just pretend to use the 0 value and have a peaceful and quiet life
again.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
The environment is modified in most surprising circumstances, and not
all of them are under Git's control. For example, calling
curl_global_init() on Windows will ensure that the CHARSET variable is
set, adding one if necessary.
While the previous commit worked around crashes triggered by such
outside changes of the environment by relaxing the requirement that the
environment be terminated by a NULL pointer, the other assumption made
by `mingw_getenv()` and `mingw_putenv()` is that the environment is
sorted, for efficient lookup via binary search.
Let's make real sure that our environment is intact before querying or
modifying it, and reinitialize our idea of the environment if necessary.
With this commit, before working on the environment we look briefly for
indicators that the environment was modified outside of our control, and
to ensure that it is terminated with a NULL pointer and sorted again in
that case.
Note: the indicators are maybe not sufficient. For example, when a
variable is removed, it will not be noticed. It might also be a problem
if outside changes to the environment result in a modified `environ`
pointer: it is unclear whether such a modification could result in a
problem when `mingw_putenv()` needs to `realloc()` the environment
buffer.
For the moment, however, the current fix works well enough, so let's
only face the potential problems when (and if!) they occur.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Outside of our Windows-specific code, the end of the environment can be
marked also by a pointer to a NUL character, not only by a NULL pointer
as our code assumed so far.
That led to a buffer overrun in `make_environment_block()` when running
`git-remote-https` in `mintty` (because `curl_global_init()` added the
`CHARSET` environment variable *outside* of `mingw_putenv()`, ending the
environment in a pointer to an empty string).
Side note for future debugging on Windows: when running programs in
`mintty`, the standard input/output/error is not connected to a Win32
Console, but instead is pipe()d. That means that even stderr may not be
written completely before a crash, but has to be fflush()ed explicitly.
For example, when debugging crashes, the developer should insert an
`fflush(stderr);` at the end of the `error()` function defined in
usage.c.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
It is unlikely that we have an empty environment, ever, but *if* we do,
when `environ_size - 1` is passed to `bsearchenv()` it is misinterpreted
as a real large integer.
To make the code truly defensive, refuse to do anything at all if the
size is negative (which should not happen, of course).
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Ensure that when passing a pipe, the gnulib poll replacement will not
return 0 before the timeout has passed.
Not obeying the timeout (and merely returning 0) causes pathological
behavior when preparing a packfile for a repository and taking a
long time to do so. If poll were to return 0 immediately, this would
cause keep-alives to get sent as quickly as possible until the packfile
was created. Such deviance from the standard would cause megabytes (or
more) of keep-alive packets to be sent.
GetTickCount is used as it is efficient, stable and monotonically
increasing. (Neither GetSystemTime nor QueryPerformanceCounter have
all three of these properties.)
Windows does not necessarily mean Cygwin, it could also be MSYS. The
latter ships with a version of "kill" that does not understand "-f". In
msysgit this was addressed shipping Cygwin's version of kill.
Properly fix this by using the stock Windows "taskkill" command instead,
which is available since Windows XP Professional.
Signed-off-by: Sebastian Schuberth <sschuberth@gmail.com>
Our current version of bash 3.1.17(5) can not parse the following snippet
correctly
p=abcd
abspath=/$p
subdir="x$(echo "$p" | tail -c $((253 - ${#abspath})))"
as it returns
tail: cannot open `253' for reading: No such file or directory
This is fixed in bash 3.1.20(4), I did not check earlier versions.
Signed-off-by: Thomas Braun <thomas.braun@byte-physics.de>
Windows paths are typically limited to MAX_PATH = 260 characters, even
though the underlying NTFS file system supports paths up to 32,767 chars.
This limitation is also evident in Windows Explorer, cmd.exe and many
other applications (including IDEs).
Particularly annoying is that most Windows APIs return bogus error codes
if a relative path only barely exceeds MAX_PATH in conjunction with the
current directory, e.g. ERROR_PATH_NOT_FOUND / ENOENT instead of the
infinitely more helpful ERROR_FILENAME_EXCED_RANGE / ENAMETOOLONG.
Many Windows wide char APIs support longer than MAX_PATH paths through the
file namespace prefix ('\\?\' or '\\?\UNC\') followed by an absolute path.
Notable exceptions include functions dealing with executables and the
current directory (CreateProcess, LoadLibrary, Get/SetCurrentDirectory) as
well as the entire shell API (ShellExecute, SHGetSpecialFolderPath...).
Introduce a handle_long_path function to check the length of a specified
path properly (and fail with ENAMETOOLONG), and to optionally expand long
paths using the '\\?\' file namespace prefix. Short paths will not be
modified, so we don't need to worry about device names (NUL, CON, AUX).
Contrary to MSDN docs, the GetFullPathNameW function doesn't seem to be
limited to MAX_PATH (at least not on Win7), so we can use it to do the
heavy lifting of the conversion (translate '/' to '\', eliminate '.' and
'..', and make an absolute path).
Add long path error checking to xutftowcs_path for APIs with hard MAX_PATH
limit.
Add a new MAX_LONG_PATH constant and xutftowcs_long_path function for APIs
that support long paths.
While improved error checking is always active, long paths support must be
explicitly enabled via 'core.longpaths' option. This is to prevent end
users to shoot themselves in the foot by checking out files that Windows
Explorer, cmd/bash or their favorite IDE cannot handle.
Test suite:
Test the case is when the full pathname length of a dir is close
to 260 (MAX_PATH).
Bug report and an original reproducer by Andrey Rogozhnikov:
https://github.com/msysgit/git/pull/122#issuecomment-43604199
Thanks-to: Martin W. Kirst <maki@bitkings.de>
Thanks-to: Doug Kelly <dougk.ff7@gmail.com>
Signed-off-by: Karsten Blees <blees@dcon.de>
Original-test-by: Andrey Rogozhnikov <rogozhnikov.andrey@gmail.com>
Signed-off-by: Stepan Kasal <kasal@ucw.cz>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
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>
Since commit 0c499ea60f the send-pack builtin uses the side-band-64k
capability if advertised by the server.
Unfortunately this breaks pushing over the dump git protocol if used
over a network connection.
The detailed reasons for this breakage are (by courtesy of Jeff Preshing,
quoted from ttps://groups.google.com/d/msg/msysgit/at8D7J-h7mw/eaLujILGUWoJ):
----------------------------------------------------------------------------
MinGW wraps Windows sockets in CRT file descriptors in order to mimic the
functionality of POSIX sockets. This causes msvcrt.dll to treat sockets as
Installable File System (IFS) handles, calling ReadFile, WriteFile,
DuplicateHandle and CloseHandle on them. This approach works well in simple
cases on recent versions of Windows, but does not support all usage patterns.
In particular, using this approach, any attempt to read & write concurrently
on the same socket (from one or more processes) will deadlock in a scenario
where the read waits for a response from the server which is only invoked after
the write. This is what send_pack currently attempts to do in the use_sideband
codepath.
----------------------------------------------------------------------------
The new config option "sendpack.sideband" allows to override the side-band-64k
capability of the server, and thus makes the dump git protocol work.
Other transportation methods like ssh and http/https still benefit from
the sideband channel, therefore the default value of "sendpack.sideband"
is still true.
Signed-off-by: Thomas Braun <thomas.braun@byte-physics.de>
The previous implementation said that the filesystem information on
Windows is not reliable to determine whether a file is executable.
To find gather this information it was peeking into the first two bytes
of a file to see whether it looks executable.
Apart from the fact that on Windows executables are usually defined as
such by their extension it lead to slow opening of help file in some
situations.
When you have virus scanner running calling open on an executable file
is a potentially expensive operation. See the following measurements (in
seconds) for example.
With virus scanner running (coldcache):
$ ./a.exe /libexec/git-core/
before open (git-add.exe): 0.000000
after open (git-add.exe): 0.412873
before open (git-annotate.exe): 0.000175
after open (git-annotate.exe): 0.397925
before open (git-apply.exe): 0.000243
after open (git-apply.exe): 0.399996
before open (git-archive.exe): 0.000147
after open (git-archive.exe): 0.397783
before open (git-bisect--helper.exe): 0.000160
after open (git-bisect--helper.exe): 0.397700
before open (git-blame.exe): 0.000160
after open (git-blame.exe): 0.399136
...
With virus scanner running (hotcache):
$ ./a.exe /libexec/git-core/
before open (git-add.exe): 0.000000
after open (git-add.exe): 0.000325
before open (git-annotate.exe): 0.000229
after open (git-annotate.exe): 0.000177
before open (git-apply.exe): 0.000167
after open (git-apply.exe): 0.000150
before open (git-archive.exe): 0.000154
after open (git-archive.exe): 0.000156
before open (git-bisect--helper.exe): 0.000132
after open (git-bisect--helper.exe): 0.000180
before open (git-blame.exe): 0.000718
after open (git-blame.exe): 0.000724
...
This test did just list the given directory and open() each file in it.
With this patch I get:
$ time git help git
Launching default browser to display HTML ...
real 0m8.723s
user 0m0.000s
sys 0m0.000s
and without
$ time git help git
Launching default browser to display HTML ...
real 1m37.734s
user 0m0.000s
sys 0m0.031s
both tests with cold cache and giving the machine some time to settle
down after restart.
Signed-off-by: Heiko Voigt <heiko.voigt@mahr.de>
