In ref_remove_duplicates, when we encounter a duplicate and remove it
from the list we need to make sure that the prev pointer stays
pointing at the last entry and also skip over adding the just freed
entry to the string_list.
Previously fetch could crash with:
*** glibc detected *** git: corrupted double-linked list: ...
Also add a test to try and catch problems with duplicate removal in
the future.
Acked-by: Nicolas Pitre <nico@fluxnic.net>
Signed-off-by: Julian Phillips <julian@quantumfyre.co.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The kernel.org hosts where the packages are built are now using Fedora
11, which defaults to sha256 for file digests instead of md5. Older
versions of rpm can not handle these packages. Tell rpmbuild to use md5
file digests for better compatibility.
Signed-off-by: Todd Zullinger <tmz@pobox.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Make git_history use git_log_generic, passing git_history_body as one
of its paramaters. This required changes to git_log_generic, in
particular passing more things as parameters.
While refactoring common code of 'log', 'shortlog' and 'history' view,
we did unify pagination, using always the form used by 'history' view,
namely
first * prev * next
in place of
HEAD * prev * next
used by 'log' and 'shortlog' views.
The 'history' view now supports commit limiting via 'hpb' parameter,
similarly to 'shortlog' (and 'log') view. Performance of 'history'
view got improved a bit, as it doesn't run git_get_hash_by_path for
"current" version in a loop. Error detection and reporting for
'history' view changed a bit.
Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Put the common parts of git_log and git_shortlog into git_log_generic
subroutine: git_log and git_shortlog are now thin wrappers calling
git_log_generic with appropriate arguments.
The unification of code responsible for 'log' and 'shorlog' actions
lead to the following changes in gitweb output
* 'tree' link in page_nav now uses $hash parameter, as was the case
for 'shortlog' but not for 'log'
* 'log' view now respect $hash_parent limiting, like 'shortlog' did
* 'log' view doesn't have special case for empty list anymore, and it
always uses page_header linking to summary view, like 'shortlog'
did.
Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Put the main part of 'log' view generation into git_log_body,
similarly how it is done for 'shortlog' and 'history' views (and
also for 'tags' and 'heads' views).
This is preparation for extracting common code between 'log',
'shortlog' and 'history' actions.
Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The commit for:
git svn: read global+system config for clone+init
Initially lacked a test case because the author was unable to
reproduce it under his test environment, this adds it.
Signed-off-by: Eric Wong <normalperson@yhbt.net>
Since $GIT_DIR does not exist when initializing new repositories,
we can follow back to the global and system config files for
git.
The logic for this was originally introduced when
$GIT_DIR/config was the only config file git could read (back
when "git config" was "git repo-config"), so the function is
renamed to "read_git_config" instead of "read_repo_config".
Signed-off-by: Eric Wong <normalperson@yhbt.net>
When recording the revisions that it has merged, SVN sets the top
revision to be the latest revision in the repository, which is not
necessarily a revision on the branch that is being merged from. When
it is not on the branch, git-svn fails to add the extra parent to
represent the merge because it relies on finding the commit on the
branch that corresponds to the top of the SVN merge range.
In order to correctly handle this case, we look for the maximum
revision less than or equal to the top of the SVN merge range that is
actually on the branch being merged from.
[ew: This includes the following (squashed) commit to prevent
errors during bisect:]
Author: Toby Allsopp <toby.allsopp@navman.co.nz>
Date: Fri Nov 13 09:48:39 2009 +1300
git-svn: add (failing) test for SVN 1.5+ merge with intervening commit
This test exposes a bug in git-svn's handling of SVN 1.5+ mergeinfo
properties. The problematic case is when there is some commit on an
unrelated branch after the last commit on the merged-from branch.
When SVN records the mergeinfo property, it records the latest
revision in the whole repository, which, in the problematic case, is
not on the branch it is merging from.
To trigger the git-svn bug, we modify t9151 to include two SVN merges,
the second of which has an intervening commit. The SVN dump was
generated using SVN 1.6.6 (on Debian squeeze amd64).
Signed-off-by: Toby Allsopp <toby.allsopp@navman.co.nz>
Acked-by: Eric Wong <normalperson@yhbt.net>
Extra paragraphs should be prefixed with a plus sign.
Signed-off-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Our Content-Length needs to report an off_t, which could be larger
precision than size_t on this system (e.g. 32 bit binary built with
64 bit large file support).
We also shouldn't be passing a size_t parameter to printf when
we've used PRIuMAX as the format specifier.
Fix both issues by using uintmax_t for the hdr_int() routine,
allowing strbuf's size_t to automatically upcast, and off_t to
always fit.
Also fixed the copy loop we use inside of send_local_file(), we never
actually updated the size variable so we might as well not use it.
Reported-by: Tarmigan <tarmigan+git@gmail.com>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In theory it is possible for sideband channel #2 to be delayed if
pack data is quick to come up for sideband channel #1. And because
data for channel #2 is read only 128 bytes at a time while pack data
is read 8192 bytes at a time, it is possible for many pack blocks to
be sent to the client before the progress message fifo is emptied,
making the situation even worse. This would result in totally garbled
progress display on the client's console as local progress gets mixed
with partial remote progress lines.
Let's prevent such situations by giving transmission priority to
progress messages over pack data at all times.
Signed-off-by: Nicolas Pitre <nico@fluxnic.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Provide a DEFAULT_PAGER knob so packagers can set the fallback
pager to something appropriate during the build.
Examples:
On (old) solaris systems, /usr/bin/less (typically the first less
found) doesn't understand the default arguments (FXRS), which
forces users to alter their environment (PATH, GIT_PAGER, LESS,
etc) or have a local or global gitconfig before paging works as
expected.
On Debian systems, by policy packages must fall back to the
'pager' command, so that changing the target of the
/usr/bin/pager symlink changes the default pager for all packages
at once.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Ben Walton <bwalton@artsci.utoronto.ca>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Provide a DEFAULT_EDITOR knob to allow setting the fallback
editor to use instead of vi (when VISUAL, EDITOR, and GIT_EDITOR
are unset). The value can be set at build time according to a
system’s policy. For example, on Debian systems, the default
editor should be the 'editor' command.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Ben Walton <bwalton@artsci.utoronto.ca>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use the new "git var GIT_PAGER" command to ask what pager to use.
Without this change, the core.pager configuration is ignored by
these commands.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use the new "git var GIT_EDITOR" feature to decide what editor to
use, instead of duplicating its logic elsewhere. This should make
the behavior of commands in edge cases (e.g., editor names with
spaces) a little more consistent.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Expose the command found by setup_pager() for scripts to use.
Scripts can use this to avoid repeating the logic to look for a
proper pager in each command.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Expose the command used by launch_editor() for scripts to use.
This should allow one to avoid searching for a proper editor
separately in each command.
git_editor(void) uses the logic to decide which editor to use
that used to live in launch_editor(). The function returns NULL
if there is no suitable editor; the caller is expected to issue
an error message when appropriate.
launch_editor() uses git_editor() and gives the error message the
same way as before when EDITOR is not set.
"git var GIT_EDITOR" gives the editor name, or an error message
when there is no appropriate one.
"git var -l" gives GIT_EDITOR=name only if there is an
appropriate editor.
Originally-submitted-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
For scripts using "git var -l" to read all logical variables at
once, not all per-variable warnings will be relevant. So suppress
them.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Refuse to use $VISUAL and fall back to $EDITOR if TERM is unset
or set to "dumb". Traditionally, VISUAL is set to a screen
editor and EDITOR to a line-based editor, which should be more
useful in that situation.
vim, for example, is happy to assume a terminal supports ANSI
sequences even if TERM is dumb (e.g., when running from a text
editor like Acme). git already refuses to fall back to vi on a
dumb terminal if GIT_EDITOR, core.editor, VISUAL, and EDITOR are
unset, but without this patch, that check is suppressed by
VISUAL=vi.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The current technical documentation for the packfile protocol is both
sparse and incorrect. This documents the fetch-pack/upload-pack and
send-pack/ receive-pack protocols much more fully.
Add documentation from Shawn's upcoming http-protocol docs that is
shared by the packfile protocol. protocol-common.txt describes ABNF
notation amendments, refname rules and the packet line format.
Add documentation on the various capabilities supported by the
upload-pack and receive-pack protocols. protocol-capabilities.txt
describes multi-ack, thin-pack, side-band[-64k], shallow, no-progress,
include-tag, ofs-delta, delete-refs and report-status.
Signed-off-by: Scott Chacon <schacon@gmail.com>
Signed-off-by: Nanako Shiraishi <nanako3@lavabit.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since a0e4639 (filter-branch: fix ref rewriting with
--subdirectory-filter, 2008-08-12) git-filter-branch has done
nearest-ancestor rewriting when using a --subdirectory-filter.
However, that rewriting strategy is also a useful building block in
other tasks. For example, if you want to split out a subset of files
from your history, you would typically call
git filter-branch -- <refs> -- <files>
But this fails for all refs that do not point directly to a commit
that affects <files>, because their referenced commit will not be
rewritten and the ref remains untouched.
The code was already there for the --subdirectory-filter case, so just
introduce an option that enables it independently.
Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Handling $filter_subdir in the usual way requires a separate case at
every use, because the variable is empty when unused.
Furthermore, --subdirectory-filter supplies its own '--', and if the user
provided one himself, such as in
git filter-branch --subdirectory-filter subdir -- --all -- subdir/file
an extra '--' was used as path filter in the call to git-rev-list that
determines the commits that shall be rewritten.
To keep the argument handling sane, we filter $@ to contain only the
non-revision arguments, and store all revisions in $ref_args. The
$ref_args are easy to handle since only the SHA1s are needed; the
actual branch names have already been stored in $tempdir/heads at this
point.
An extra separating -- is only required if the user did not provide
any non-revision arguments, as the latter disambiguate the
$filter_subdir following after them (or fail earlier because they are
ambiguous themselves).
Thanks to Johannes Sixt for suggesting this solution.
Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Start the diff --no-index usage string with "usage:" instead of
"fatal:".
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Usage messages (for example, from "git merge-recursive -h") are
friendlier when not preceded by "fatal".
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Some new callers would want to use printf-like formatting, when issuing
their usage messages. An option is to change usage() itself also be like
printf(), which would make it similar to die() and warn().
But usage() is typically fixed, as opposed to die() and warn() that gives
diagnostics depending on the situation. Indeed, the majority of strings
given by existing callsites to usage() are fixed strings. If we were to
make usage() take printf-style format, they all need to be changed to have
"%s" as their first argument.
So instead, introduce usagef() so that limited number of callers can use
it.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There is no need for "git <command> -h" to depend on being inside
a repository.
Reported by Gerfried Fuchs through http://bugs.debian.org/462557
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
http-push already knows how to dump usage if it is given no options, but
it interprets '-h' as the URL to a remote repository:
$ git http-push -h
error: Cannot access URL -h/, return code 6
Dump usage instead. Humans wanting to pass the URL -h/ to curl for some
reason can use 'git http-push -h/' explicitly. Scripts expecting to
access an HTTP repository at URL '-h' will break, though.
Also delay finding a git directory until after option parsing, so
"http-push -h" can be used outside any git repository.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Acked-by: Tay Ray Chuan <rctay89@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"-p" means "generate patch" in 'git log' and 'git diff', so it's
quite surprising that it means "suppress diffstat" in
'git format-patch'.
Keep the "-p" option for backward compatibility, but add
"--no-stat" as a more intuitive synonym. For backward compatibility
with scripts, we must allow combinations of --stat and --no-stat.
Signed-off-by: Björn Gustavsson <bgustavsson@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Format git commands and options consistently using back quotes
(i.e. a fixed font in the resulting HTML document).
Signed-off-by: Björn Gustavsson <bgustavsson@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To simplify reading the documentation for format-patch, remove the
description of common diff options that are not useful for the
purpose of the command (i.e. "Prepare patches for e-mail submission").
Specifically, this removes the description of the following options:
--raw
-z
--color
--no-color
--color-words
--diff-filter
-S
--pickaxe-all
--pickaxe-regex
-R
--relative
--exit-code
--quiet
Signed-off-by: Björn Gustavsson <bgustavsson@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Jeff King recently reinstated -p to suppress the default diffstat
(as -p used to work before 68daa64, about 14 months ago).
However, -p is also needed in combination with certain options
(e.g. --stat or --numstat) in order to produce any patch at all.
The documentation does not mention this.
Since the purpose of format-patch is to produce a patch that
can be emailed, it does not make sense that certain combination
of options will suppress the generation of the patch itself.
Therefore:
* Update 'git format-patch' to always generate a patch.
* Since the --name-only, --name-status, and --check suppresses
the generation of the patch, disallow those options,
and remove the description of them in the documentation.
* Remove the reference to -p in the description of -U.
* Remove the descriptions of the options that are synonyms for -p
plus another option (--patch-with-raw and --patch-with-stat).
* While at it, slightly tweak the description of -p itself
to say that it generates "plain patches", so that you can
think of -p as "plain patch" as an mnemonic aid.
Signed-off-by: Björn Gustavsson <bgustavsson@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Delay search for a git directory until option parsing has finished.
None of the functions used in option parsing look for or read any
files other than stdin, so this is safe.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"unpack-file -h" could be asking to save the contents of a blob
named "-h". Strictly speaking, such a pathological ref name is
possible, but the user would have to had said something like
"tags/-h" to name such a pathological ref already. When used in
scripts, unpack-file is typically not passed a user-supplied tag
name directly.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
