pager: find out the terminal width before spawning the pager
term_columns() checks for terminal width via ioctl(2) on the standard
output, but we spawn the pager too early for this check to be useful.
The effect of this buglet can be observed by opening a wide terminal and
running "git -p help --all", which still shows 80-column output, while
"git help --all" uses the full terminal width. Run the check before we
spawn the pager to fix this.
While at it, move term_columns() to pager.c and export it from cache.h so
that callers other than the help subsystem can use it.
Signed-off-by: Zbigniew Jędrzejewski-Szmek <zbyszek@in.waw.pl> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Receive runs rev-list --verify-objects in order to detect missing
objects. However, such errors are ignored and overridden later.
Instead, consequently ignore all update commands for which an error has
already been detected.
Some tests in t5504 are obsoleted by this change, because invalid
objects are detected even if fsck is not enabled. Instead, they now test
for different error messages depending on whether or not fsck is turned
on. A better fix would be to force a corruption that will be detected by
fsck but not by rev-list.
Signed-off-by: Clemens Buchacher <drizzd@aon.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
By default, progress output is disabled if stderr is not a terminal.
The --progress option can be used to force progress output anyways.
Conversely, --no-progress does not force progress output. In particular,
if stderr is a terminal, progress output is enabled.
This is unintuitive. Change --no-progress to force output off.
Signed-off-by: Clemens Buchacher <drizzd@aon.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git rev-list passes rev_list_info, not rev_list objects. Without this
fix, rev-list enables or disables the --verify-objects option depending
on a read from an undefined memory location.
Signed-off-by: Clemens Buchacher <drizzd@aon.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
I'm no longer running the Git smoke testing service at
smoke.git.nix.is due to Smolder being a fragile piece of software not
having time to follow through on making it easy for third parties to
run and submit their own smoke tests.
So remove the support in Git for sending smoke tests to
smoke.git.nix.is, it's still easy to modify the test suite to submit
smokes somewhere else.
This reverts the following commits:
Revert "t/README: Add SMOKE_{COMMENT,TAGS}= to smoke_report target" -- e38efac87d
Revert "t/README: Document the Smoke testing" -- d15e9ebc5c
Revert "t/Makefile: Create test-results dir for smoke target" -- 617344d77b
Revert "tests: Infrastructure for Git smoke testing" -- b6b84d1b74
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Makefile: Change the default compiler from "gcc" to "cc"
Ever since the very first commit to git.git we've been setting CC to
"gcc". Presumably this is behavior that Linus copied from the Linux
Makefile.
However unlike Linux Git is written in ANSI C and supports a multitude
of compilers, including Clang, Sun Studio, xlc etc. On my Linux box
"cc" is a symlink to clang, and on a Solaris box I have access to "cc"
is Sun Studio's CC.
Both of these are perfectly capable of compiling Git, and it's
annoying to have to specify CC=cc on the command-line when compiling
Git when that's the default behavior of most other portable programs.
So change the default to "cc". Users who want to compile with GCC can
still add "CC=gcc" to the make(1) command-line, but those users who
don't have GCC as their "cc" will see expected behavior, and as a
bonus we'll be more likely to smoke out new compilation warnings from
our distributors since they'll me using a more varied set of compilers
by default.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Makefile: introduce CHARSET_LIB to link with -lcharset
On some systems, the function locale_charset() may not be exported from
libiconv but is available from libcharset, and we need -lcharset when
linking.
Introduce a make variable CHARSET_LIB that can be set to -lcharsetlib
on such systems. Also autodetect this in the configure script by first
looking for the symbol in libiconv, and then libcharset.
mergetools/meld: Use --help output to detect --output support
In v1.7.7-rc0~3^2 (2011-08-19), git mergetool's "meld" support learned
to use the --output option when calling versions of meld that are
detected to support it (1.5.0 and newer, hopefully).
Alas, it misdetects old versions (before 1.1.5, 2006-06-11) of meld as
supporting the option, so on systems with such meld, instead of
getting a nice merge helper, the operator gets a dialog box with the
text "Wrong number of arguments (Got 5)". (Version 1.1.5 is when meld
switched to using optparse. One consequence of that change was that
errors in usage are detected and signalled through the exit status
even when --help was passed.)
Luckily there is a simpler check that is more reliable: the usage
string printed by "meld --help" reliably reflects whether --output is
supported in a given version. Use it.
Reported-by: Jeff Epler <jepler@unpythonic.net> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Explicitly set X to avoid potential build breakage
$X is appended to binary names for Windows builds (ie. git.exe).
Pollution from the environment can inadvertently trigger this behaviour,
resulting in 'git' turning into 'gitwhatever' without warning.
Signed-off-by: Michael Palimaka <kensington@astralcloak.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
merge: do not launch an editor on "--no-edit $tag"
When the user explicitly asked us not to, don't launch an editor.
But do everything else the same way as the "edit" case, i.e. leave the
comment with verification result in the log template and record the
mergesig in the resulting merge commit for later inspection.
It is necessary to write the else branch as a nested conditional. Also,
write the conditions with parentheses because we use them throughout the
Makefile.
Signed-off-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git tag -n" did not check the type of the object it is reading the top n
lines from. At least, avoid showing the beginning of trees and blobs when
dealing with lightweight tags that point at them.
As the payload of a tag and a commit look similar in that they both start
with a header block, which is skipped for the purpose of "-n" output,
followed by human readable text, allow the message of commit objects to be
shown just like the contents of tag objects. This avoids regression for
people who have been using "tag -n" to show the log messages of commits
that are pointed at by lightweight tags.
commit: ignore intent-to-add entries instead of refusing
Originally, "git add -N" was introduced to help users from forgetting to
add new files to the index before they ran "git commit -a". As an attempt
to help them further so that they do not forget to say "-a", "git commit"
to commit the index as-is was taught to error out, reminding the user that
they may have forgotten to add the final contents of the paths before
running the command.
This turned out to be a false "safety" that is useless. If the user made
changes to already tracked paths and paths added with "git add -N", and
then ran "git add" to register the final contents of the paths added with
"git add -N", "git commit" will happily create a commit out of the index,
without including the local changes made to the already tracked paths. It
was not a useful "safety" measure to prevent "forgetful" mistakes from
happening.
It turns out that this behaviour is not just a useless false "safety", but
actively hurts use cases of "git add -N" that were discovered later and
have become popular, namely, to tell Git to be aware of these paths added
by "git add -N", so that commands like "git status" and "git diff" would
include them in their output, even though the user is not interested in
including them in the next commit they are going to make.
Fix this ancient UI mistake, and instead make a commit from the index
ignoring the paths added by "git add -N" without adding real contents.
Based on the work by Nguyễn Thái Ngọc Duy, and helped by injection of
sanity from Jonathan Nieder and others on the Git mailing list.
add -e: do not show difference in a submodule that is merely dirty
When the HEAD of the submodule matches what is recorded in the index of
the superproject, and it has local changes or untracked files, the patch
offered by "git add -e" for editing shows a diff like this:
Because applying such a patch has no effect to the index, this is a
useless noise. Generate the patch with IGNORE_DIRTY_SUBMODULES flag to
prevent such a change from getting reported.
This patch also loses the "-dirty" suffix from the output when the HEAD of
the submodule is different from what is in the index of the superproject.
As such dirtiness expressed by the suffix does not affect the result of
the patch application at all, there is no information lost if we remove
it. The user could still run "git status" before "git add -e" if s/he
cares about the dirtiness.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git checkout -b: allow switching out of an unborn branch
Running "git checkout -b another" immediately after "git init" when you do
not even have a commit on 'master' fails with:
$ git checkout -b another
fatal: You are on a branch yet to be born
This is unnecessary, if we redefine "git checkout -b $name" that does not
take any $start_point (which has to be a commit) as "I want to check out a
new branch $name from the state I am in".
completion: simplify __gitcomp and __gitcomp_nl implementations
These shell functions are written in an unnecessarily verbose way;
simplify their "conditionally use $<number> after checking $# against
<number>" logic by using shell's built-in conditional substitution
facilities.
Also remove the first of the two assignments to IFS in __gitcomp_nl
that does not have any effect.
Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: use ls -1 instead of rolling a loop to do that ourselves
This simplifies the code a great deal. In particular, it allows us to
get rid of __git_shopt, which is used only in this fuction to enable
'nullglob' in zsh.
[jn: squashed with a patch that actually gets rid of __git_shopt]
Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: work around zsh option propagation bug
When listing commands in zsh (git <TAB><TAB>), all of them will show up,
instead of only porcelain ones.
The root cause of this is because zsh versions from 4.3.0 to present
(4.3.15) do not correctly propagate the SH_WORD_SPLIT option into the
subshell in ${foo:=$(bar)} expressions. Because of this bug, the list of
all commands was treated as a single word in __git_list_porcelain_commands
and did not match any of the patterns that would usually cause plumbing to
be excluded.
With problematic versions of zsh, after running
emulate sh
fn () {
var='one two'
for v in $var; do echo $v; done
}
x=$(fn)
: ${y=$(fn)}
printing "$x" results in two lines as expected, but printing "$y" results
in a single line because $var is expanded as a single word when evaluating
fn to compute y.
So avoid the construct, and use an explicit 'test -n "$foo" || foo=$(bar)'
instead.
Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
mailmap: always return a plain mail address from map_user()
The callers of map_user() give email and name to it, and expect to get the
up-to-date email and/or name to be used in their output. The function
rewrites the given buffers in place. To optimize the majority of cases,
the function returns 0 when it did not do anything, and it returns 1 when
the caller should use the updated contents.
The 'email' input to the function is terminated by '>' or a NUL (whichever
comes first) for historical reasons, but when a rewrite happens, the value
is replaced with the mailbox inside the <> pair. However, it failed to
meet this expectation when it only rewrote the name part without rewriting
the email part, and the email in the input was terminated by '>'.
This causes an extra '>' to appear in the output of "blame -e", because the
caller does send in '>'-terminated email, and when the function returned 1
to tell it that rewriting happened, it appends '>' that is necessary when
the email part was rewritten.
The patch looks bigger than it actually is, because this change makes a
variable that points at the end of the email part in the input 'p' live
much longer than it used to, deserving a more descriptive name.
Noticed and diagnosed by Felipe Contreras and Jeff King.
fsck: give accurate error message on empty loose object files
Since 3ba7a065527a (A loose object is not corrupt if it
cannot be read due to EMFILE), "git fsck" on a repository with an empty
loose object file complains with the error message
fatal: failed to read object <sha1>: Invalid argument
This comes from a failure of mmap on this empty file, which sets errno to
EINVAL. Instead of calling xmmap on empty file, we display a clean error
message ourselves, and return a NULL pointer. The new message is
error: object file .git/objects/09/<rest-of-sha1> is empty
fatal: loose object <sha1> (stored in .git/objects/09/<rest-of-sha1>) is corrupt
The second line was already there before the regression in 3ba7a065527a,
and the first is an additional message, that should help diagnosing the
problem for the user.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We don't usually bother looking at tagged objects at all
when listing. However, if "-n" is specified, we open the
objects to read the annotations of the tags. If we fail to
read an object, or if the object has zero length, we simply
silently return.
The first case is an indication of a broken or corrupt repo,
and we should notify the user of the error.
The second case is OK to silently ignore; however, the
existing code leaked the buffer returned by read_sha1_file.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When "git tag" is instructed to print lines from annotated
tags via "-n", it first prints the tag name, then attempts
to parse and print the lines of the tag object, and then
finally adds a trailing newline.
If an error occurs, we return early from the function and
never print the newline, screwing up the output for the next
tag. Let's factor the line-printing into its own function so
we can manage the early returns better, and make sure that
we always terminate the line.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix build problems related to profile-directed optimization
There was a number of problems I ran into when trying the
profile-directed optimizations added by Andi Kleen in git commit 7ddc2710b9. (This was using gcc 4.4 found on many enterprise
distros.)
1) The -fprofile-generate and -fprofile-use commands are incompatible
with ccache; the code ends up looking in the wrong place for the gcda
files based on the ccache object names.
2) If the makefile notices that CFLAGS are different, it will rebuild
all of the binaries. Hence the recipe originally specified by the
INSTALL file ("make profile-all" followed by "make install") doesn't
work. It will appear to work, but the binaries will end up getting
built with no optimization.
This patch fixes this by using an explicit set of options passed via
the PROFILE variable then using this to directly manipulate CFLAGS and
EXTLIBS.
The developer can run "make PROFILE=BUILD all ; sudo make
PROFILE=BUILD install" automatically run a two-pass build with the
test suite run in between as the sample workload for the purpose of
recording profiling information to do the profile-directed
optimization.
Alternatively, the profiling version of binaries can be built using:
make PROFILE=GEN PROFILE_DIR=/var/cache/profile all
make PROFILE=GEN install
and then after git has been used for a while, the optimized version of
the binary can be built as follows:
make PROFILE=USE PROFILE_DIR=/var/cache/profile all
make PROFILE=USE install
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu> Cc: Andi Kleen <ak@linux.intel.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* cb/push-quiet:
t5541: avoid TAP test miscounting
fix push --quiet: add 'quiet' capability to receive-pack
server_supports(): parse feature list more carefully
The imap-send code was adapted from another project, and
still contains many unused bits of code. One of these bits
contains a type "struct string_list" which bears no
resemblence to the "struct string_list" we use elsewhere in
git. This causes the compiler to complain if git's
string_list ever becomes part of cache.h.
Let's just drop the dead code.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
branch --edit-description: protect against mistyped branch name
It is very easy to mistype the branch name when editing its description,
e.g.
$ git checkout -b my-topic master
: work work work
: now we are at a good point to switch working something else
$ git checkout master
: ah, let's write it down before we forget what we were doing
$ git branch --edit-description my-tpoic
The command does not notice that branch 'my-tpoic' does not exist. It is
not lost (it becomes description of an unborn my-tpoic branch), but is not
very useful. So detect such a case and error out to reduce the grief
factor from this common mistake.
This incidentally also errors out --edit-description when the HEAD points
at an unborn branch (immediately after "init", or "checkout --orphan"),
because at that point, you do not even have any commit that is part of
your history and there is no point in describing how this particular
branch is different from the branch it forked off of, which is the useful
bit of information the branch description is designed to capture.
We may want to special case the unborn case later, but that is outside the
scope of this patch to prevent more common mistakes before 1.7.9 series
gains too much widespread use.
Drop system includes from inet_pton/inet_ntop compatibility wrappers
As both of these compatibility wrappers include git-compat-utils.h,
all of the system includes were redundant.
Dropping these system includes also makes git-compat-utils.h the first
include which avoids a compiler warning on Solaris due to the
redefinition of _FILE_OFFSET_BITS.
Signed-off-by: Ben Walton <bwalton@artsci.utoronto.ca> Signed-off-by: Junio C Hamano <gitster@pobox.com>
merge: do not create a signed tag merge under --ff-only option
Starting at release v1.7.9, if you ask to merge a signed tag, "git merge"
always creates a merge commit, even when the tag points at a commit that
happens to be a descendant of your current commit.
Unfortunately, this interacts rather badly for people who use --ff-only to
make sure that their branch is free of local developments. It used to be
possible to say:
and expect that the resulting tip of frotz branch matches v1.7.9^0 (aka
the commit tagged as v1.7.9), but this fails with the updated Git with:
fatal: Not possible to fast-forward, aborting.
because a merge that merges v1.7.9 tag to v1.7.9~30 cannot be created by
fast forwarding.
We could teach users that now they have to do
$ git merge --ff-only v1.7.9^0
but it is far more pleasant for users if we DWIMmed this ourselves.
When an integrator pulls in a topic from a lieutenant via a signed tag,
even when the work done by the lieutenant happens to fast-forward, the
integrator wants to have a merge record, so the integrator will not be
asking for --ff-only when running "git pull" in such a case. Therefore,
this change should not regress the support for the use case v1.7.9 wanted
to add.
"git diff --stat" and "git apply --stat" now learn to print the line
"%d files changed, %d insertions(+), %d deletions(-)" in singular form
whenever applicable. "0 insertions" and "0 deletions" are also omitted
unless they are both zero.
This matches how versions of "diffstat" that are not prehistoric produced
their output, and also makes this line translatable.
[jc: with help from Thomas Dickey in archaeology of "diffstat"]
[jc: squashed Jonathan's updates to illustrations in tutorials and a test]
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The only place that the issue this series addresses was observed
where we read "cat-file commit" output and put it in GIT_AUTHOR_DATE
in order to replay a commit with an ancient timestamp.
With the previous patch alone, "git commit --date='20100917 +0900'"
can be misinterpreted to mean an ancient timestamp, not September in
year 2010. Guard this codepath by requring an extra '@' in front of
the raw git timestamp on the parsing side. This of course needs to
be compensated by updating get_author_ident_from_commit and the code
for "git commit --amend" to prepend '@' to the string read from the
existing commit in the GIT_AUTHOR_DATE environment variable.
The date-time parser parses out a human-readble datestring piece by
piece, so that it could even parse a string in a rather strange
notation like 'noon november 11, 2005', but restricts itself from
parsing strings in "<seconds since epoch> <timezone>" format only
for reasonably new timestamps (like 1974 or newer) with 10 or more
digits. This is to prevent a string like "20100917" from getting
interpreted as seconds since epoch (we want to treat it as September
17, 2010 instead) while doing so.
The same codepath is used to read back the timestamp that we have
already recorded in the headers of commit and tag objects; because
of this, such a commit with timestamp "0 +0000" cannot be rebased or
amended very easily.
Teach parse_date() codepath to special case a string of the form
"<digits> +<4-digits>" to work this issue around, but require that
there is no other cruft around the string when parsing a timestamp
of this format for safety.
Note that this has a slight backward incompatibility implications.
If somebody writes "git commit --date='20100917 +0900'" and wants it
to mean a timestamp in September 2010 in Japan, this change will
break such a use case.
This is caused by the fact that localized messages do not have their
place in some RPM package. Let's postpone decision where they should
be put (be it git-i18n-Icelandic, or git-i18n, or git package itself)
for later by removing locale files at the end of install phase.
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Many of the scripts in the test suite write small helper
shell scripts to disk. It's best if these shell scripts
start with "#!$SHELL_PATH" rather than "#!/bin/sh", because
/bin/sh on some platforms is too buggy to be used.
However, it can be cumbersome to expand $SHELL_PATH, because
the usual recipe for writing a script is:
cat >foo.sh <<-\EOF
#!/bin/sh
echo my arguments are "$@"
EOF
To expand $SHELL_PATH, you have to either interpolate the
here-doc (which would require quoting "\$@"), or split the
creation into two commands (interpolating the $SHELL_PATH
line, but not the rest of the script). Let's provide a
helper function that makes that less syntactically painful.
While we're at it, this helper can also take care of the
"chmod +x" that typically comes after the creation of such a
script, saving the caller a line.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The current askpass code simply dies if calling an askpass
helper fails. Worse, in some failure modes it doesn't even
print an error (if start_command fails, then it prints its
own error; if reading fails, we print an error; but if the
command exits non-zero, finish_command fails and we print
nothing!).
Let's be more kind to the user by printing an error message
when askpass doesn't work out, and then falling back to the
terminal (which also may fail, of course, but we die already
there with a nice message).
While we're at it, let's clean up the existing error
messages a bit. Now that our prompts are very long and
contain quotes and colons themselves, our error messages are
hard to read.
So the new failure modes look like:
[before, with a terminal]
$ GIT_ASKPASS=false git push
$ echo $?
128
[before, with no terminal, and we must give up]
$ setsid git push
fatal: could not read 'Password for 'https://peff@github.com': ': No such device or address
[after, with a terminal]
$ GIT_ASKPASS=false git push
error: unable to read askpass response from 'false'
Password for 'https://peff@github.com':
[after, with no terminal, and we must give up]
$ GIT_ASKPASS=false setsid git push
error: unable to read askpass response from 'false'
fatal: could not read Password for 'https://peff@github.com': No such device or address
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The do_askpass function inherited a few bad habits from the
original git_getpass. One, there's no need to strbuf_reset a
buffer which was just initialized. And two, it's a good
habit to use strbuf_detach to claim ownership of a buffer's
string (even though in this case the owning buffer goes out
of scope, so it's effectively the same thing).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
gitweb: Allow UTF-8 encoded CGI query parameters and path_info
Gitweb forgot to turn query parameters into UTF-8. This results in a bug
that one cannot search for a string with characters outside US-ASCII. For
example searching for "Michał Kiedrowicz" (containing letter 'ł' - LATIN
SMALL LETTER L WITH STROKE, with Unicode codepoint U+0142, represented
with 0xc5 0x82 bytes in UTF-8 and percent-encoded as %C5%82) result in the
following incorrect data in search field
MichaÅ\202 Kiedrowicz
This is caused by CGI by default treating '0xc5 0x82' bytes as two
characters in Perl legacy encoding latin-1 (iso-8859-1), because 's'
query parameter is not processed explicitly as UTF-8 encoded string.
The solution used here follows "Using Unicode in a Perl CGI script"
article on http://www.lemoda.net/cgi/perl-unicode/index.html:
use CGI;
use Encode 'decode_utf8;
my $value = params('input');
$value = decode_utf8($value);
Decoding UTF-8 is done when filling %input_params hash and $path_info
variable; the former requires to move from explicit $cgi->param(<label>)
to $input_params{<name>} in a few places, which is a good idea anyway.
Also add -override=>1 parameter to $cgi->textfield() invocation in search
form. Otherwise CGI would use values from query string if it is present,
filling value from $cgi->param... without decode_utf8(). As we are using
value of appropriate parameter anyway, -override=>1 doesn't change the
situation but makes gitweb fill search field correctly.
We could simply use the '-utf8' pragma (via "use CGI '-utf8';") to solve
this, but according to CGI.pm documentation, it may cause problems with
POST requests containing binary files, and it requires CGI 3.31 (I think),
released with perl v5.8.9.
Reported-by: Michał Kiedrowicz <michal.kiedrowicz@gmail.com> Signed-off-by: Jakub Narębski <jnareb@gmail.com> Tested-by: Michał Kiedrowicz <michal.kiedrowicz@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
standardize and improve lookup rules for external local repos
When you specify a local repository on the command line of
clone, ls-remote, upload-pack, receive-pack, or upload-archive,
or in a request to git-daemon, we perform a little bit of
lookup magic, doing things like looking in working trees for
.git directories and appending ".git" for bare repos.
For clone, this magic happens in get_repo_path. For
everything else, it happens in enter_repo. In both cases,
there are some ambiguous or confusing cases that aren't
handled well, and there is one case that is not handled the
same by both methods.
This patch tries to provide (and test!) standard, sensible
lookup rules for both code paths. The intended changes are:
1. When looking up "foo", we have always preferred
a working tree "foo" (containing "foo/.git" over the
bare "foo.git". But we did not prefer a bare "foo" over
"foo.git". With this patch, we do so.
2. We would select directories that existed but didn't
actually look like git repositories. With this patch,
we make sure a selected directory looks like a git
repo. Not only is this more sensible in general, but it
will help anybody who is negatively affected by change
(1) negatively (e.g., if they had "foo.git" next to its
separate work tree "foo", and expect to keep finding
"foo.git" when they reference "foo").
3. The enter_repo code path would, given "foo", look for
"foo.git/.git" (i.e., do the ".git" append magic even
for a repo with working tree). The clone code path did
not; with this patch, they now behave the same.
In the unlikely case of a working tree overlaying a bare
repo (i.e., a ".git" directory _inside_ a bare repo), we
continue to treat it as a working tree (prefering the
"inner" .git over the bare repo). This is mainly because the
combination seems nonsensical, and I'd rather stick with
existing behavior on the off chance that somebody is relying
on it.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The low-level grep_source code will automatically load the
userdiff driver to see whether a file is binary. However,
when we are threaded, it will load the drivers in a
non-deterministic order, handling each one as its assigned
thread happens to be scheduled.
Meanwhile, the attribute lookup code (which underlies the
userdiff driver lookup) is optimized to handle paths in
sequential order (because they tend to share the same
gitattributes files). Multi-threading the lookups destroys
the locality and makes this optimization less effective.
We can fix this by pre-loading the userdiff driver in the
main thread, before we hand off the file to a worker thread.
My best-of-five for "git grep foo" on the linux-2.6
repository went from:
real 0m0.391s
user 0m1.708s
sys 0m0.584s
to:
real 0m0.360s
user 0m1.576s
sys 0m0.572s
Not a huge speedup, but it's quite easy to do. The only
trick is that we shouldn't perform this optimization if "-a"
was used, in which case we won't bother checking whether
the files are binary at all.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Usually we load each file to grep into memory, check whether
it's binary, and then either grep it (the default) or not
(if "-I" was given).
In the "-I" case, we can skip loading the file entirely if
it is marked as binary via gitattributes. On my giant
3-gigabyte media repository, doing "git grep -I foo" went
from:
real 0m0.712s
user 0m0.044s
sys 0m4.780s
to:
real 0m0.026s
user 0m0.016s
sys 0m0.020s
Obviously this is an extreme example. The repo is almost
entirely binary files, and you can see that we spent all of
our time asking the kernel to read() the data. However, with
a cold disk cache, even avoiding a few binary files can have
an impact.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
There is currently no way for users to tell git-grep that a
particular path is or is not a binary file; instead, grep
always relies on its auto-detection (or the user specifying
"-a" to treat all binary-looking files like text).
This patch teaches git-grep to use the same attribute lookup
that is used by git-diff. We could add a new "grep" flag,
but that is unnecessarily complex and unlikely to be useful.
Despite the name, the "-diff" attribute (or "diff=foo" and
the associated diff.foo.binary config option) are really
about describing the contents of the path. It's simply
historical that diff was the only thing that cared about
these attributes in the past.
And if this simple approach turns out to be insufficient, we
still have a backwards-compatible path forward: we can add a
separate "grep" attribute, and fall back to respecting
"diff" if it is unset.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Right now, grep only uses the userdiff_driver for one thing:
looking up funcname patterns for "-p" and "-W". As new uses
for userdiff drivers are added to the grep code, we want to
minimize attribute lookups, which can be expensive.
It might seem at first that this would also optimize multiple
lookups when the funcname pattern for a file is needed
multiple times. However, the compiled funcname pattern is
already cached in struct grep_opt's "priv" member, so
multiple lookups are already suppressed.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before the grep_source interface existed, grep_buffer was
used by two types of callers:
1. Ones which pulled a file into a buffer, and then wanted
to supply the file's name for the output (i.e.,
git grep).
2. Ones which really just wanted to grep a buffer (i.e.,
git log --grep).
Callers in set (1) should now be using grep_source. Callers
in set (2) always pass NULL for the "name" parameter of
grep_buffer. We can therefore get rid of this now-useless
parameter.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The grep_source interface (as opposed to grep_buffer) will
eventually gives us a richer interface for telling the
low-level grep code about our buffers. Eventually this will
lead to things like better binary-file handling. For now, it
lets us drop a lot of now-redundant code.
The conversion is mostly straight-forward. One thing to note
is that the memory ownership rules for "struct grep_source"
are different than the "struct work_item" found here (the
former will copy things like the filename, rather than
taking ownership). Therefore you will also see some slight
tweaking of when filename buffers are released.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
grep: refactor the concept of "grep source" into an object
The main interface to the low-level grep code is
grep_buffer, which takes a pointer to a buffer and a size.
This is convenient and flexible (we use it to grep commit
bodies, files on disk, and blobs by sha1), but it makes it
hard to pass extra information about what we are grepping
(either for correctness, like overriding binary
auto-detection, or for optimizations, like lazily loading
blob contents).
Instead, let's encapsulate the idea of a "grep source",
including the buffer, its size, and where the data is coming
from. This is similar to the diff_filespec structure used by
the diff code (unsurprising, since future patches will
implement some of the same optimizations found there).
The diffstat is slightly scarier than the actual patch
content. Most of the modified lines are simply replacing
access to raw variables with their counterparts that are now
in a "struct grep_source". Most of the added lines were
taken from builtin/grep.c, which partially abstracted the
idea of grep sources (for file vs sha1 sources).
Instead of dropping the now-redundant code, this patch
leaves builtin/grep.c using the traditional grep_buffer
interface (which now wraps the grep_source interface). That
makes it easy to test that there is no change of behavior
(yet).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The multi-threaded git-grep code needs to serialize access
to the thread-unsafe read_sha1_file call. It does this with
a mutex that is local to builtin/grep.c.
Let's instead push this down into grep.c, where it can be
used by both builtin/grep.c and grep.c. This will let us
safely teach the low-level grep.c code tricks that involve
reading from the object db.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The low-level grep code traditionally didn't care about
threading, as it doesn't do any threading itself and didn't
call out to other non-thread-safe code. That changed with 0579f91 (grep: enable threading with -p and -W using lazy
attribute lookup, 2011-12-12), which pushed the lookup of
funcname attributes (which is not thread-safe) into the
low-level grep code.
As a result, the low-level code learned about a new global
"grep_attr_mutex" to serialize access to the attribute code.
A multi-threaded caller (e.g., builtin/grep.c) is expected
to initialize the mutex and set "use_threads" in the
grep_opt structure. The low-level code only uses the lock if
use_threads is set.
However, putting the use_threads flag into the grep_opt
struct is not the most logical place. Whether threading is
in use is not something that matters for each call to
grep_buffer, but is instead global to the whole program
(i.e., if any thread is doing multi-threaded grep, every
other thread, even if it thinks it is doing its own
single-threaded grep, would need to use the locking). In
practice, this distinction isn't a problem for us, because
the only user of multi-threaded grep is "git-grep", which
does nothing except call grep.
This patch turns the opt->use_threads flag into a global
flag. More important than the nit-picking semantic argument
above is that this means that the locking functions don't
need to actually have access to a grep_opt to know whether
to lock. Which in turn can make adding new locks simpler, as
we don't need to pass around a grep_opt.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
i18n: format_tracking_info "Your branch is behind" message
Function format_tracking_info in remote.c is called by
wt_status_print_tracking in wt-status.c, which will print
branch tracking message in git-status. git-checkout also
show these messages through it's report_tracking function.
Signed-off-by: Jiang Xin <worldhello.net@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Mark the "merge/cherry-pick" messages in whence_s for translation.
These messages returned from whence_s function are used as argument
to build other messages.
Signed-off-by: Jiang Xin <worldhello.net@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
find_pack_entry(): do not keep packed_git pointer locally
Commit f7c22cc (always start looking up objects in the last used pack
first - 2007-05-30) introduce a static packed_git* pointer as an
optimization. The kept pointer however may become invalid if
free_pack_by_name() happens to free that particular pack.
Current code base does not access packs after calling
free_pack_by_name() so it should not be a problem. Anyway, move the
pointer out so that free_pack_by_name() can reset it to avoid running
into troubles in future.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Acked-by: Nicolas Pitre <nico@fluxnic.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
sha1_file.c: move the core logic of find_pack_entry() into fill_pack_entry()
The new helper function implements the logic to find the offset for the
object in one pack and fill a pack_entry structure. The next patch will
restructure the loop and will call the helper from two places.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Acked-by: Nicolas Pitre <nico@fluxnic.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
request-pull: explicitly ask tags/$name to be pulled
When asking for a tag to be pulled, disambiguate by leaving tags/ prefix
in front of the name of the tag. E.g.
... in the git repository at:
git://example.com/git/git.git/ tags/v1.2.3
for you to fetch changes up to 123456...
This way, older versions of "git pull" can be used to respond to such a
request more easily, as "git pull $URL v1.2.3" did not DWIM to fetch
v1.2.3 tag in older versions. Also this makes it clearer for humans that
the pull request is made for a tag and he should anticipate a signed one.
merge: add instructions to the commit message when editing
Before f824628 (merge: use editor by default in interactive sessions,
2012-01-10), git-merge only started an editor if the user explicitly
asked for it with --edit. Thus it seemed unlikely that the user would
need extra guidance.
After f824628 the _normal_ thing is to start an editor. Give at least
an indication of why we are doing it.
The sentence about justification is one of the few things about
standard git that are not agnostic to the workflow that the user
chose. However, f824628 was proposed by Linus specifically to
discourage users from merging unrelated upstream progress into topic
branches. So we may as well take another step in the same direction.
Signed-off-by: Thomas Rast <trast@student.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Recent releases of Redhat/Fedora are reported to ship Perl binary package
with some core modules stripped away (see http://lwn.net/Articles/477234/)
against the upstream Perl5 people's wishes. The Time::HiRes module used by
gitweb one of them.
zsh adds a backslash (foo\ ) for each item in the COMPREPLY array if IFS
doesn't contain spaces. This issue has been reported[1], but there is no
solution yet.
This wasn't a problem due to another bug[2], which was fixed in zsh
version 4.3.12. After this change, 'git checkout ma<tab>' would resolve
to 'git checkout master\ '.
Aditionally, the introduction of __gitcomp_nl in commit a31e626
(completion: optimize refs completion) in git also made the problem
apparent, as Matthieu Moy reported.
The simplest and most generic solution is to hide all the changes we do
to IFS, so that "foo \nbar " is recognized by zsh as "foo bar". This
works on versions of git before and after the introduction of
__gitcomp_nl (a31e626), and versions of zsh before and after 4.3.12.
Once zsh is fixed, we should conditionally disable this workaround to
have the same benefits as bash users.
submodule add: fix breakage when re-adding a deep submodule
Since recently a submodule with name <name> has its git directory in the
.git/modules/<name> directory of the superproject while the work tree
contains a gitfile pointing there.
When the same submodule is added on a branch where it wasn't present so
far (it is not found in the .gitmodules file), the name is not initialized
from the path as it should. This leads to a wrong path entered in the
gitfile when the .git/modules/<name> directory is found, as this happily
uses the - now empty - name. It then always points only a single directory
up, even if we have a path deeper in the directory hierarchy.
Fix that by initializing the name of the submodule early in module_clone()
if module_name() returned an empty name and add a test to catch that bug.
Reported-by: Jehan Bing <jehan@orb.com> Signed-off-by: Jens Lehmann <Jens.Lehmann@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Some merge tools cannot cope when $LOCAL, $BASE, or $REMOTE are missing.
$BASE can be missing when two branches independently add the same
filename.
Provide an empty file to make these tools happy.
When a delete/modify conflict occurs, $LOCAL and $REMOTE can also be
missing. We have special case code to handle such case so this change
may not affect that codepath, but try to be consistent and create an
empty file for them anyway.
Reported-by: Jason Wenger <jcwenger@gmail.com> Signed-off-by: David Aguilar <davvid@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
In threaded mode, git-grep emits file breaks (enabled with context, -W
and --break) into the accumulation buffers even if they are not
required. The output collection thread then uses skip_first_line to
skip the first such line in the output, which would otherwise be at
the very top.
This is wrong when the user also specified -l/-L/-c, in which case
every line is relevant. While arguably giving these options together
doesn't make any sense, git-grep has always quietly accepted it. So
do not skip anything in these cases.
Signed-off-by: Albert Yale <surfingalbert@gmail.com> Signed-off-by: Thomas Rast <trast@student.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
remote-curl: Fix push status report when all branches fail
The protocol between transport-helper.c and remote-curl requires
remote-curl to always print a blank line after the push command
has run. If the blank line is ommitted, transport-helper kills its
container process (the git push the user started) with exit(128)
and no message indicating a problem, assuming the helper already
printed reasonable error text to the console.
However if the remote rejects all branches with "ng" commands in the
report-status reply, send-pack terminates with non-zero status, and
in turn remote-curl exited with non-zero status before outputting
the blank line after the helper status printed by send-pack. No
error messages reach the user.
This caused users to see the following from git push over HTTP
when the remote side's update hook rejected the branch:
$ git push http://... master
Counting objects: 4, done.
Delta compression using up to 6 threads.
Compressing objects: 100% (2/2), done.
Writing objects: 100% (3/3), 301 bytes, done.
Total 3 (delta 0), reused 0 (delta 0)
$
Always print a blank line after the send-pack process terminates,
ensuring the helper status report (if it was output) will be
correctly parsed by the calling transport-helper.c. This ensures
the helper doesn't abort before the status report can be shown to
the user.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Making pathspec limited log play nicer with --first-parent
In a topic branch workflow, you often want to find the latest commit that
merged a side branch that touched a particular area of the system, so that
a new topic branch to work on that area can be forked from that commit.
For example, I wanted to find an appropriate fork-point to queue Luke's
changes related to git-p4 in contrib/fast-import/.
"git log --first-parent" traverses the first-parent chain, and "-m --stat"
shows the list of paths touched by commits including merge commits. We
could ask the question this way:
# What is the latest commit that touched that path?
$ git log --first-parent --oneline -m --stat master |
sed -e '/^ contrib\/fast-import\/git-p4 /q' | tail
The above finds that 8cbfc11 (Merge branch 'pw/p4-view-updates',
2012-01-06) was such a commit.
But a more natural way to spell this question is this:
Unfortunately, this does not work. It finds ecb7cf9 (git-p4: rewrite view
handling, 2012-01-02). This commit is a part of the merged topic branch
and is _not_ on the first-parent path from the 'master':
The problem is caused by the merge simplification logic when it inspects
the merge commit 8cbfc11. In this case, the history leading to the tip of
'master' did not touch git-p4 since 'pw/p4-view-updates' topic forked, and
the result of the merge is simply a copy from the tip of the topic branch
in the view limited by the given pathspec. The merge simplification logic
discards the history on the mainline side of the merge, and pretends as if
the sole parent of the merge is its second parent, i.e. the tip of the
topic. While this simplification is correct in the general case, it is at
least surprising if not outright wrong when the user explicitly asked to
show the first-parent history.
Here is an attempt to fix this issue, by not allowing us to compare the
merge result with anything but the first parent when --first-parent is in
effect, to avoid the history traversal veering off to the side branch.
diff-index: enable recursive pathspec matching in unpack_trees
The pathspec structure has a few bits of data to drive various operation
modes after we unified the pathspec matching logic in various codepaths.
For example, max_depth field is there so that "git grep" can limit the
output for files found in limited depth of tree traversal. Also in order
to show just the surface level differences in "git diff-tree", recursive
field stops us from descending into deeper level of the tree structure
when it is set to false, and this also affects pathspec matching when
we have wildcards in the pathspec.
The diff-index has always wanted the recursive behaviour, and wanted to
match pathspecs without any depth limit. But we forgot to do so when we
updated tree_entry_interesting() logic to unify the pathspec matching
logic.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* jk/credentials:
credential-cache: ignore "connection refused" errors
unix-socket: do not let close() or chdir() clobber errno during cleanup
credential-cache: report more daemon connection errors
unix-socket: handle long socket pathnames