threaded pack-objects: Use condition variables for thread communication.
In the threaded pack-objects code the main thread and the worker threads
must mutually signal that they have assigned a new pack of work or have
completed their work, respectively. Previously, the code used mutexes that
were locked in one thread and unlocked from a different thread, which is
bogus (and happens to work on Linux).
Here we rectify the implementation by using condition variables: There is
one condition variable on which the main thread waits until a thread
requests new work; and each worker thread has its own condition variable
on which it waits until it is assigned new work or signaled to terminate.
As a cleanup, the worker threads are spawned only after the initial work
packages have been assigned.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Acked-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Because print_summary() forgot to call diff_setup_done() after futzing with
diff output options, it failed to activate recursive diff, which resulted in
an incorrect summary.
We need to be extra careful recovering the removed common section, so
that we do not break context nor the changed incomplete line (i.e. the
last line that does not end with LF).
whitespace: more accurate initial-indent highlighting
Instead of highlighting the entire initial indent, highlight only the
problematic spaces.
In the case of an indent like ' \t \t' there may be multiple problematic
ranges, so it's easiest to emit the highlighting as we go instead of
trying rember disjoint ranges and do it all at the end.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
After this patch, "written" counts the number of bytes up to and
including the most recently seen tab. This allows us to detect (and
count) spaces by comparing to "i".
This allows catching initial indents like '\t ' (a tab followed
by 8 spaces), while previously indent-with-non-tab caught only indents
that consisted entirely of spaces.
This also allows fixing an indent-with-non-tab regression, so we can
again detect indents like '\t \t'.
Also update tests to catch these cases.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The variable leading_space is initially used to represent the index of
the last space seen before a non-space. Then later it represents the
index of the first non-indent character.
It will prove simpler to replace it by a variable representing a number
of bytes. Eventually it will represent the number of bytes written so
far (in the stream != NULL case).
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The convention for helper scripts has been
git-$TOOL--$HELPER. Since this is a "browse" helper for the
"help" tool, git-help--browse is a more sensible name.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
gitweb: Make config_to_multi return [] instead of [undef]
This is important for the list of clone urls, where if there are
no per-repository clone URL configured, the default base URLs
are never used for URL construction without this patch.
Add tests for different ways of setting project URLs, just in case.
Note that those tests in current form wouldn't detect breakage fixed
by this patch, as it only checks for errors and not for expected
output.
Signed-off-by: Petr Baudis <pasky@suse.cz> Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
gitweb: Teach "a=blob" action to be more lenient about blob/file mime type
Since 930cf7dd7cc6b87d173f182230763e1f1913d319 'blob' action knows the
file type; if the file type is not "text/*" or one of common network
image formats/mimetypes (gif, png, jpeg) then the action "blob"
defaulted to "blob_plain". This caused the problem if mimetypes file
was not well suited for web, for example returning "application/x-sh"
for "*.sh" shell scripts, instead of "text/plain" (or other "text/*").
Now "blob" action defaults to "blob_plain" ('raw' view) only if file
is of type which is neither "text/*" nor "image/{gif,png,jpeg}"
AND it is binary file. Otherwise it assumes that it can be displayed
either in <img> tag ("image/*" mimetype), or can be displayed line by
line (otherwise).
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
gitweb: disambiguate heads and tags withs the same name
Avoid wrong disambiguation that would link logs/trees of tags and
heads which share the same name to the same page, leading to
a disambiguation that would prefer the tag, thus making it impossible
to access the corresponding head log and tree without hacking the url
by hand.
It does it by using full refname (with 'refs/heads/' or 'refs/tags/'
prefix) instead of shortened one in the URLs in 'heads' and 'tags'
tables. This makes URLs (and refs) provided by gitweb unambiguous.
Signed-off-by: Guillaume Seguin <guillaume@segu.in> Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
teach bash completion to treat commands with "--" as a helper
There is a convention that commands containing a double-dash
are implementation details and not to be used by mortals. We
should automatically remove them from the completion
suggestions as such.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The recovered context lines were not LF terminated due to off-by-one
error, which also caused the outer loop to count the number of recovered
lines to terminate after running only once.
Most of them are still stubs, but the procedure to build the HTML
documentation, maintaining the index and installing the end product are
there.
I placed names of people who are likely to know the most about the topic
in the stub files, so that volunteers will know whom to ask questions as
needed.
This config variable makes it possible to choose the default format
used to display help. This format will be used only if no option
like -a|--all|-i|--info|-m|--man|-w|--web is passed to "git-help".
The following values are possible for this variable:
- "man" --> "man" program is used
- "info" --> "info" program is used
- "web" --> "git-browse-help" is used
By default we still show help using "man".
This patch also adds -m|--man command line option to use "man"
to allow overriding the "help.format" configuration variable.
Note that this patch also revert some recent changes in
"git-browse-help" because they prevented to look for config
variables in the global configuration file.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* wc/diff:
Test interaction between diff --check and --exit-code
Use shorter error messages for whitespace problems
Add tests for "git diff --check" with core.whitespace options
Make "diff --check" output match "git apply"
Unify whitespace checking
diff --check: minor fixups
"diff --check" should affect exit status
Setting CURLOPT_HTTPHEADER doesn't add HTTP headers, but replaces whatever
set of headers was configured before, so setting to NULL doesn't have any
magic meaning, and is pretty much useless when setting to another list
right after.
Signed-off-by: Mike Hommey <mh@glandium.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix random sha1 in error message in http-fetch and http-push
When a downloaded ref doesn't contain a sha1, the error message displays
a random sha1 because of uninitialized memory. This happens when cloning
a repository that is already a clone of another one, in which case
refs/remotes/origin/HEAD is a symref.
Signed-off-by: Mike Hommey <mh@glandium.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The specialized pool allocator fast-import uses aligned objects on the
size of a pointer, which was not sufficient at least on Sparc. Instead,
make the alignment for objects of type unitmax_t.
Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
remote: Fix bogus make_branch() call in configuration reader.
The configuration reader to enumerate branches that have configuration
data were not careful enough and failed to skip "branch.<variable>"
entries (e.g. branch.autosetupmerge). This resulted in bogus attempt to
allocate huge memory.
commit: allow --amend to reuse message from another commit
After tentatively applying a patch from a contributor, you can get a
replacement patch with corrected code and unusable commit log message.
In such a case, this sequence ought to give you an editor based on the
message in the earlier commit, to let you describe an incremental
improvement:
git reset --hard HEAD^ ;# discard the earlier one
git am <corrected-patch
git commit --amend -c HEAD@{1}
Unfortunately, --amend insisted reusing the message from the commit
being amended, ignoring the -c option. This corrects it.
git-svn: handle our top-level path being deleted and later re-added
Previously, git-svn would ignore cases where the path we're
tracking is removed from the repository. This was to prevent
heads with follow-parent from ending up with a tree full of
empty revisions (and thus breaking rename detection).
The previous behavior is fine until the path we're tracking
is re-added later on, leading to the old files being merged
in with the new files in the directory (because the old
files were never marked as deleted)
We will now only remove all the old files locally that were
deleted remotely iff we detect the directory we're in is being
created from scratch.
Thanks for Marcus D. Hanwell for the bug report and
Peter Baumann for the analysis.
Signed-off-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Because we feed the changed filenames to CVS on the command
line, it was possible for massive commits to overflow the
system exec limits. Instead, we now do an xargs-like split
of the arguments.
This means that we lose some of the atomicity of calling CVS
in one shot. Since CVS commits are not atomic, but the CVS
protocol is, the possible effects of this are not clear;
however, since CVS doesn't provide a different interface,
this is our only option for large commits (short of writing
a CVS client library).
The argument size limit is arbitrarily set to 64kB. This
should be high enough to trigger only in rare cases where it
is necessary, so normal-sized commits are not affected by
the atomicity change.
For consistency, make the two tools report whitespace errors in the
same way (the output of "diff --check" has been tweaked to match
that of "git apply").
Note that although the textual content is basically the same only
"git diff --check" provides a colorized version of the problematic
lines; making "git apply" do colorization will require more extensive
changes (figuring out the diff colorization preferences of the user)
and so that will be a subject for another commit.
Signed-off-by: Wincent Colaiuta <win@wincent.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This commit unifies three separate places where whitespace checking was
performed:
- the whitespace checking previously done in builtin-apply.c is
extracted into a function in ws.c
- the equivalent logic in "git diff" is removed
- the emit_line_with_ws() function is also removed because that also
rechecks the whitespace, and its functionality is rolled into ws.c
The new function is called check_and_emit_line() and it does two things:
checks a line for whitespace errors and optionally emits it. The checking
is based on lines of content rather than patch lines (in other words, the
caller must strip the leading "+" or "-"); this was suggested by Junio on
the mailing list to allow for a future extension to "git show" to display
whitespace errors in blobs.
At the same time we teach it to report all classes of whitespace errors
found for a given line rather than reporting only the first found error.
Signed-off-by: Wincent Colaiuta <win@wincent.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
There is no reason --exit-code and --check-diff must be mutually
exclusive, so assign different bits to different results and allow them
to be returned from the command. Introduce diff_result_code() to factor
out the common code to decide final status code based on diffopt
settings and use it everywhere.
Update tests to match the above fix.
Turning pager off when "diff --check" is used is a regression.
This implements earlier Linus's optimization to trim common lines at the
end before passing them down to low level xdiff interface for all of our
xdiff users.
We could later enhance this to also trim common leading lines, but that
would need tweaking the output function to add the number of lines
trimmed at the beginning to line numbers that appear in the hunk
headers.
This inserts a new function xdi_diff() that currently does not
do anything other than calling the underlying xdl_diff() to the
callchain of current callers of xdl_diff() function.
Signed-off-by: Shawn Bohrer <shawn.bohrer@gmail.com> Acked-by: J. Bruce Fields <bfields@citi.umich.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This invocation of run_status() is asked to find out if there is a
committable change, but it unconditionally gave instructions such as
"use git-add" at the same time. When in merge, we do allow an empty
change to be recorded, so after showing this message the code still went
ahead and made a commit.
This introduces "nowarn" parameter to run_status() to avoid these
useless messages. If we are not allowed to create an empty commit, we
already call run_status() again in the original codepath, and the
message will be shown from that call anyway.
* cc/help:
RPM spec: Adjust htmldir
git-help -w: do not require to be in git repository
git.spec.in: remove python_path
Documentation: rename git.texi to user-manual.texi
Add git-browse-help to .gitignore
git-help -i: show info documentation from matching version of git
git-help -i: invoke info with document and node name
Documentation: add gitman.info target
Documentation: describe -w/--web option to "git-help".
Use {web,instaweb,help}.browser config options.
git-help: add -w|--web option to display html man page in a browser.
Documentation: describe -i/--info option to "git-help"
git-help: add -i|--info option to display info page.
git-help -w: do not require to be in git repository
The users used to be able to say "git help cat-file" from anywhere, but
the browse-help script insisted to be in a git repository, which caused
"git help -w cat-file" to barf outside. Correct it.
* jc/shortlog-e:
shortlog: default to HEAD when the standard input is a tty
Invert numbers and names in the git-shortlog summary mode.
shortlog: document -e option
git-shortlog -e: show e-mail address as well
"diff --check" would only detect spaces before tabs if a tab was the
last character in the leading indent. Fix that and add a test case to
make sure the bug doesn't regress in the future.
Signed-off-by: Wincent Colaiuta <win@wincent.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "-z" format is all about machine parsability, but showing renamed
paths as "common/{a => b}/suffix" makes it impossible. The scripts would
never have successfully parsed "--numstat -z -M" in the old format.
This fixes the output format in a (hopefully minimally) backward
incompatible way.
* The output without -z is not changed. This has given a good way for
humans to view added and deleted lines separately, and showing the
path in combined, shorter way would preserve readability.
* The output with -z is unchanged for paths that do not involve renames.
Existing scripts that do not pass -M/-C are not affected at all.
* The output with -z for a renamed path is shown in a format that can
easily be distinguished from an unrenamed path.
This is based on Jakub Narebski's patch. Bugs and documentation typos
are mine.
Fix clone not to ignore depth when performing a local clone
When git-clone detects that it can perform a local clone it
follows a path that silently ignores the depth parameter.
Presumably if the user explicitly requests a shallow clone they
have a reason to prefer a space efficient clone of just the recent
history so bypass the local magic if the user specifies the depth
parameter.
Signed-off-by: Charles Bailey <charles@hashpling.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
shortlog: default to HEAD when the standard input is a tty
Instead of warning the user that it is expecting git log output from
the standard input (and waiting for the user to type the log from
the keyboard, which is a silly thing to do), default to traverse from
HEAD when there is no rev parameter given and the standard input is
a tty.
This factors out a useful helper "add_head()" from builtin-diff.c to a
more appropriate place revision.c while renaming it to more descriptive
name add_head_to_pending(), as that is what the function is about.
blame: drop blob data after passing blame to the parent
We used to keep the blob data for each origin that has any remaining
line in the result, but this will get very costly with a huge file that
has a deep history. This patch releases the blob after we ran diff
between the child rev and its parents. When passing blame from a parent
to its parent (i.e. the grandparent), the blob data for the parent may
need to be read again, but it should be relatively cheap, thanks to
delta-base cache.
When deciding whether or not to turn on automatic color
support, git_config_colorbool checks whether stdout is a
tty. However, because we run a pager, if stdout is not a
tty, we must check whether it is because we started the
pager. This used to be done by checking the pager_in_use
variable.
This variable was set only when the git program being run
started the pager; there was no way for an external program
running git indicate that it had already started a pager.
This patch allows a program to set GIT_PAGER_IN_USE to a
true value to indicate that even though stdout is not a tty,
it is because a pager is being used.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
merge-recursive did not support merging trees that have conflicting
changes in submodules they contain, and died. Support it exactly the
same way as how it handles conflicting symbolic link changes --- mark it
as a conflict, take the tentative result from the current side, and
letting the caller resolve the conflict, without dying in merge_file()
function.
Also reword the error message issued when merge_file() has to die
because it sees a tree entry of type it does not support yet.
[jc: fixed up initial draft by Finn Arne Gangstad]
commit: do not add extra LF at the end of the summary.
The scripted version relied on the nice "auto-strip the terminating LF"
behaviour shell gives to "var=$(cmd)" construct, but we have to roll
that ourselves.
Commit 396865859918e9c7bf8ce74aae137c57da134610 broke signed tags using
the "-u" flag when it made builtin-tag.c use parse_options() to parse its
arguments (but it quite possibly was broken even before that, by the
builtin rewrite).
It used to be that passing the signing ID with the -u parameter also
(obviously!) implied that you wanted to sign and annotate the tag, but
that logic got dropped. It also totally ignored the actual key ID that was
passed in.
This reinstates it all.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
DESTDIR is supposed to be overridden on 'make install' after doing
'make'. Have the automatically generated perl/perl.mak not cache the
value of DESTDIR to support that for the perl/ subdirectory also.
Signed-off-by: Gerrit Pape <pape@smarden.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
autoconf: Check asciidoc version to automatically set ASCIIDOC8
Check for asciidoc, and if it exists check asciidoc version, setting
ASCIIDOC8 when needed. Currently it just runs asciidoc in asciidoc7
compatibility mode (see: Documentation/Makefile).
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-svn: reinstate old rev_db optimization in new rev_map
This reinstates an old optimization in .rev_db which
stored the highest revision number we scanned, allowing
us to avoid scanning the SVN log for those revisions
again in a subsequent invocation.
This means the last 24-byte record in a .rev_map file
can be a 4-byte SVN revision number with 20-bytes of
zeroes representing a non-existent commit. This record
can and will be overwritten when a new commit iff
the commit is all zeroes.
Signed-off-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
pack-objects: more threaded load balancing fix with often changed paths
The code that splits the object list amongst work threads tries to do so
on "path" boundaries not to prevent good delta matches. However, in
some cases, a few paths may largely dominate the hash distribution and
it is not possible to have good load balancing without ignoring those
boundaries.
Signed-off-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
If you have local changes that don't conflict with the
branch-switching changes, these should be kept, not cause errors even
without -m, and be reported afterwards in name-status format.
With -m, the changes carried across should be listed as well. And, for
now, include the merge-recursive output from this process.
Also test the detatched head message in at least one case.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>