When refactoring code to split one iteration of a too deeply
nested loop into a separate function, it inevitably makes the
indentation levels shallower (that's the sole point of such a
refactoring). With "git blame -w", you can ignore such
re-indentation and pass blame for such moved lines to the
parent.
When two branches are merged that modify a subdirectory (possibly in
different intermediate steps) such that both end up identical, then
rev-list chooses only one branch. But when we filter history, we want to
keep both branches. Therefore, we must use --full-history.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
With git-filter-branch --subdirectory-filter <subdirectory> you can
get at the history, as seen by a certain subdirectory. The history
of the rewritten branch will only contain commits that touched that
subdirectory, and the subdirectory will be rewritten to be the new
project root.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
remote.c: "git-push frotz" should update what matches at the source.
Earlier, when the local repository has a branch "frotz" and the
remote repository has a tag "frotz" (but not branch "frotz"),
"git-push frotz" mistakenly updated the tag at the remote side.
This was because the partial refname matching code was applied
independently on both source and destination side.
With this fix, when a colon-less refspec is given to git-push,
we first match it with the refs in the source repository, and
update the matching ref in the destination repository.
remote.c: fix "git push" weak match disambiguation
When "git push A:B" is given, and A (or B) is not a full refname
that begins with refs/, we require an unambiguous match with an
existing ref. For this purpose, a match with a local branch or
a tag (i.e. refs/heads/A and refs/tags/A) is called a "strong
match", and any other match is called a "weak match". A partial
refname is unambiguous when there is only one strong match with
any number of weak matches, or when there is only one weak match
and no other match.
However, as reported by Sparse with Ramsay Jones recently,
count_refspec_match() function had a bug where a variable in an
inner block masked a different variable of the same name, which
caused the weak matches to be ignored.
When checking what ref the source refspec matches, we have no
business setting the default for the destination, so move that
code lower. Also simplify the result from the code block that
matches the source side by making it set matched_src only upon
unambiguous match.
Make command description imperative statement, not third-person present.
In several of the text messages, the tense of the verb is inconsistent.
For example, "Add" vs "Creates". It is customary to use imperative for
command description.
Make git-cvsserver understand some options inspired by
git-daemon, namely --base-path, --export-all, --strict-paths.
Also allow the caller to specify a whitelist of allowed
directories, again similar to git-daemon.
While already adding option parsing also support the common
--help and --version options.
Rationale:
While the gitcvs.enabled configuration option already
offers means to limit git-cvsserver access to a repository,
there are some use cases where other methods of access
control prove to be more useful.
E.g. if setting up a pserver for a collection of public
repositories one might want limit the exported repositories
to exactly the directory this collection is located whithout
having to worry about other repositories that might lie around
with the configuration variable set (never trust your users ;)
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
cvsserver: Make req_Root more critical of its input data
The path submitted with the Root request has to be absolute
(cvs does it this way and it may save us some sanity checks
later)
If multiple roots are specified (e.g. because we use
pserver authentication which will already include the
root), ensure that they say all the same.
Probably neither is a security risk, and neither should ever
be triggered by a sane client, but when validating
input data, it's better to be save than sorry.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Define __slots__ for the Commit class. This reserves space in each Commit
object for only the defined variables. On my system this reduces heap usage
when viewing a kernel repo by 12% ~= 55868 KB.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This changes the Commit class to use new-style class, which has
been available since Python 2.2 (Dec 2001). This is a necessary
step in order to use __slots__[] declaration, so that we can
reduce the memory footprint in the next patch.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Junio C Hamano <gitster@pobox.com>
checkout: do not get confused with ambiguous tag/branch names
Although it is not advisable, we have always allowed a branch
and a tag to have the same basename (i.e. it is not illegal to
have refs/heads/frotz and refs/tags/frotz at the same time).
When talking about a specific commit, the interpretation of
'frotz' has always been "use tag and then check branch",
although we warn when ambiguities exist.
However "git checkout $name" is defined to (1) first see if it
matches the branch name, and if so switch to that branch; (2)
otherwise it is an instruction to detach HEAD to point at the
commit named by $name. We did not follow this definition when
$name appeared under both refs/heads/ and refs/tags/ -- we
switched to the branch but read the tree from the tagged commit,
which was utterly bogus.
will consider references in "refs/heads" of the remote repository
"remote-name", instead of the ones in "refs/remotes/other", which
the given refspec clearly means.
Signed-off-by: Alex Riesen <raa.lkml@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-gui: Changed blame header bar background to match main window
The main window's diff header bar background switched from orange
to gold recently, and I liked the effect it had on readability of
the text. Since I wanted the blame viewer to match, here it is.
Though this probably should be a user defined color, or at least
a constant somewhere that everyone can reference.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Only get the expensive branch mapping from the p4 server when not
syncing with the help of an origin remote (which we instead then use
to get new branches from).
Don't attempt to set the initialParent on multi-branch imports (useless).
At some point the code paths should be unified, but for now I need a working
git-p4 :)
Signed-off-by: Simon Hausmann <shausman@trolltech.com>
Fix common path "calculation" from logs of multiple branches.
Need to use min instead of max for prev/cur to avoid out-of-bounds
string access. Also treat "i" as index of the last match instead of
a length because in case of a complete match of the two strings
i was off by one.
Signed-off-by: Simon Hausmann <shausman@trolltech.com>
This uses "git-apply --whitespace=strip" to fix whitespace errors that have
crept in to our source files over time. There are a few files that need
to have trailing whitespaces (most notably, test vectors). The results
still passes the test, and build result in Documentation/ area is unchanged.
Fix clone to setup the origin if its name ends with .git
The problem is visible when cloning a local repo. The cloned
repository will have the origin url setup incorrectly: the origin name
will be copied verbatim in origin url of the cloned repository.
Normally, the name is to be expanded into absolute path.
Signed-off-by: Alex Riesen <raa.lkml@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Currently git-cvsimport tries to create tag objects directly via git-mktag
in a very broken way, e.g the stuff it writes into the tagger field of
the tag object doesn't really resemble the GIT_COMMITTER_IDENT. This makes
gitweb and possibly other tools that try to interpret tag objects to be
confused about tag date and authorship.
Fix this by calling git-tag instead. This also has a nice side effect of
not creating the tag object but only the lightweight tag as that's the only
thing CVS has anyways.
Signed-off-by: Elvis Pranskevichus <el@prans.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
pack-check: Sort entries by pack offset before unpacking them.
Because of the way objects are sorted in a pack, unpacking them in
disk order is much more efficient than random access. Tests on the
Wine repository show a gain in pack validation time of about 35%.
Signed-off-by: Alexandre Julliard <julliard@winehq.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* sv/objfixes:
Don't assume tree entries that are not dirs are blobs
git-cvsimport: Make sure to use $git_dir always instead of .git sometimes
fix documentation of unpack-objects -n
Accept dates before 2000/01/01 when specified as seconds since the epoch
git-cvsimport: Make sure to use $git_dir always instead of .git sometimes
CVS import was failing on a couple repos I was trying to import.
I was setting GIT_DIR=newproj.git and using the -i flag, but this bug
was thwarting the effort... evil CVS.
Accept dates before 2000/01/01 when specified as seconds since the epoch
Tests with git-filter-branch on a repository that was converted from
CVS and that has commits reaching back to 1999 revealed that it is
necessary to parse dates before 2000/01/01 when they are specified
as seconds since 1970/01/01. There is now still a limit, 100000000,
which is 1973/03/03 09:46:40 UTC, in order to allow that dates are
represented as 8 digits.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
filter-branch: also don't fail in map() if a commit cannot be mapped
The map() function can be used by filters to map a commit id to its
rewritten id. Such a mapping may not exist, in which case the identity
mapping is used (the commit is returned unchanged).
In the rewrite loop, this mapping is also needed, but was done
explicitly in the same way. Use the map() function instead.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Acked-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
filter-branch: Use rev-list arguments to specify revision ranges.
A subset of commits in a branch used to be specified by options (-k, -r)
as well as the branch tip itself (-s). It is more natural (for git users)
to specify revision ranges like 'master..next' instead. This makes it so.
If no range is specified it defaults to 'HEAD'.
As a consequence, the new name of the filtered branch must be the first
non-option argument. All remaining arguments are passed to 'git rev-list'
unmodified.
The tip of the branch that gets filtered is implied: It is the first
commit that git rev-list would print for the specified range.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Acked-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-submodule: clone during update, not during init
This teaches 'git-submodule init' to register submodule paths and urls in
.git/config instead of actually cloning them. The cloning is now handled
as part of 'git-submodule update'.
With this change it is possible to specify preferred/alternate urls for
the submodules in .git/config before the submodules are cloned.
Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
cvsimport: add <remote>/HEAD reference in separate remotes more
When in separate remote mode (via -r <remote>) we can now use
the name HEAD for the CVS HEAD. In keeping with git-clone
remotes/<remote>/HEAD is creates as a symbolic ref to the user
specified name for the HEAD which defaults to master.
Signed-off-by: Andy Whitcroft <apw@shadowen.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
cvsimport: add support for new style remote layout
cvsimport creates any branches found in the remote CVS repository
in the refs/heads namespace. This makes sense for a repository
conversion. When using git as a sane interface to a remote CVS
repository, that repository may well remain as the 'master'
respository. In this model it makes sense to import the CVS
repository into the refs/remotes namespace.
Add a new option '-r <remote>' to set the remote name for
this import. When this option is specified branches are named
refs/remotes/<remote>/branch, with HEAD named as master matching
git-clone separate remotes layout. Without branches are placed
ion refs/heads, with HEAD named origin as before.
Signed-off-by: Andy Whitcroft <apw@shadowen.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
$EMAIL is a last resort fallback, as it's system-wide.
$EMAIL is a system-wide setup that is used for many many many
applications. If the git user chose a specific user.email setup,
then _this_ should be honoured rather than $EMAIL.
Signed-off-by: Pierre Habouzit <madcoder@debian.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
filter-branch: prevent filters from reading from stdin
stdin is the list of commits when the env, tree and index
filter are executed. The filters are not supposed to read
anything from stdin so the best is to give them /dev/null
for reading.
Signed-off-by: Matthias Lederhofer <matled@gmx.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-gui: Favor the original annotations over the recent ones
Usually when you are looking at blame annotations for a region of
a file you are more interested in why something was originally
done then why it is here now. This is because most of the time
when we get original annotation data we are looking at a simple
refactoring performed to better organize code, not to change its
semantic meaning or function. Reorganizations are sometimes of
interest, but not usually.
We now show the original commit data first in the tooltip. This
actually looks quite nice as the original commit will usually have an
author date prior to the current (aka move/copy) annotation's commit,
so the two commits will now tend to appear in chronological order.
I also found myself to always be clicking on the line of interest
in the file column but I always wanted the original tracking data
and not the move/copy data. So I changed our default commit from
$asim_data (the simple move/copy annotation) to the more complex
$amov_data (the -M -C -C original annotation).
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Improve our labeling of blame annotation types
It feels wrong to call the -M -C -C annotations "move/copy tracking"
as they are actually the original locations. So I'm relabeling
the status bar to show "copy/move tracking annotations" for the
current file (no -M -C -C) as that set of annotations tells us who
put the hunk here (who moved/copied it). I'm now calling the -M
-C -C pass "original location annotations" as that's what we're
really digging for.
I also tried to clarify some of the text in the hover tooltip.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Use three colors for the blame viewer background
To prevent neighboring lines that are different commits from using
the same background color we now use 3 colors and assign them
by selecting the color that is not used before or after the line
in question. We still color "on the fly" as we receive hunks from
git-blame, but we delay our color decisions until we are getting
the original location data (the slower -M -C -C pass) as that is
usually more fine-grained than the current location data.
Credit goes to Martin Waitz for the tri-coloring concept.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
When the user clicks on a commit link within one of the columns
in the blame viewer we now jump them not just to that commit/file
pair but also to the line of the original file. This saves the
user a lot of time, as they don't need to search through the new
file data for the chunk they were previously looking at.
We also restore the prior view when the user clicks the back button
to return to a pior commit/file pair that they were looking at.
Turned out this was quite tricky to get working in Tk. Every time
I tried to jump the text widgets to the correct locations by way
of the "yview moveto" or "see" subcommands Tk performed the change
until the current event finished dispatching, and then reset the
views back to 0, making the change never take place. Forcing Tk
to run the pending events before we jump the UI resolves the issue.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we have commit data from both the simple blame and the
rename/move tracking blame and they differ than there is a
bigger story to tell. We now include data from both commits
so that the user can see that this link as moved, who moved
it, and where it originated from.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Run blame twice on the same file and display both outputs
We now perform two passes over any input file given to the blame
viewer. Our first pass is a quick "git-blame" with no options,
getting the details of how each line arrived into this file. We
are specifically ignoring/omitting the rename detection logic as
this first pass is to determine why things got into the state they
are in.
Once the first pass is complete and is displayed in the UI we run
a second pass, using the much more CPU intensive "-M -C -C" options
to perform extensive rename/movement detection. The output of this
second pass is shown in a different column, allowing the user to see
for any given line how it got to be, and if it came from somewhere
else, where that is.
This is actually very instructive when run on our own lib/branch.tcl
script. That file grew recently out of a very large block of code
in git-gui.sh. The first pass shows when I created that file, while
the second pass shows the original commit information.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Display the "Loading annotation..." message in italic
If the user clicks on a line region that we haven't yet received
an annotation for from git-blame we show them "Loading annotation".
But I don't want the user to confuse this loading message with a
commit whose first line is "Loading annotation" and think we messed
up our display somehow. Since we never use italics for anything
else, I'm going with the idea that italic slant can be used to show
data is missing/elided out at the time being.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Rename fields in blame viewer to better descriptions
Calling the commit message pane $w_cmit is a tad confusing when
we also have the $w_cgrp column that shows the abbreviated SHA-1s.
So w_cmit -> w_cviewer, as it is the "commit viewer"; and
w_cgrp -> w_amov as it is the "annotated commit + move tracking"
column. Also changed line_data -> amov_data, as that list is
exactly the results shown in w_amov.
Why call the column "move tracking"? Because this column holds
data from "git blame -M -C". I'm considering adding an additional
column that holds the data from "git blame" without -M/-C, showing
who did the copy/move, and when they did it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Label the uncommitted blame history entry
If the user runs the blame viewer on a working directory file
instead of a specific commit-ish then we have no value for the
commit SHA1 or the summary line; this causes the history menu
to get an empty entry at the very bottom. We now look for this
odd case and call the meny entry "Working Directory".
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Switch internal blame structure to Tcl lists
The Tcl list datatype is significantly faster to work with than
the array type, especially if our indexes are a consecutive set
of numbers, like say line numbers in a file.
This rather large change reorganizes the internal data structure
of the blame viewer to use a proper Tcl list for the annotation
information about a line. Each line is given its own list within
the larger line_data list, where the indexes correspond to various
facts about that particular line.
The interface does seem to be more responsive this way, with less
time required by Tcl to process blame, and to switch to another
version of the same file. It could just be a placebo effect, but
either way most Tcl experts perfer lists for this type of work over
arrays.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Cleanup redundant column management in blame viewer
The code to handle our three different text widgets is a bit
on the messy side as we issue the same command on all three
widgets one at a time. Adding (or removing) columns from the
viewer is messy, as a lot of locations need to have the new
column added into the sequence, or removed from it.
We also now delete the tags we create for each commit when
we switch to display another "commit:path" pair. This way the
text viewer doesn't get bogged down with a massive number of tags
as we traverse through history.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The array variable "order" used to be used to tell us in what
order each commit was received in. Recent changes have removed
that need for an ordering and the "order" array is now just a
boolean 'do we have that commit yet' flag.
The colors were moved to fields, so they appear inside of the
blame viewer instance. This keeps two different concurrently
running blame viewers from stepping on each other's ordering
of the colors in group_colors.
Most of the other fields were moved around a little bit so
that they are organized by major category and value lifespan.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Remove unused commit_list from blame viewer
This list used to store the commits in the order we received
them in. I originally was using it to update the colors of
the commit before and the commit after the current commit,
but since that interface concept turned out to be horribly
ugly and has been removed we no longer need this list.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Automatically expand the line number column as needed
After we finish reading a chunk of data from the file stream
we know how many digits we need in the line number column to
show the current maximum line number. If our line number column
isn't wide enough, we should expand it out to the correct width.
Any file over our default allowance of 5 digits (99,999 lines)
is so large that the slight UI "glitch" when we widen the column
out is trivial compared to the time it will take Git to fully do
the annotations.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Make the line number column slightly wider in blame
Most source code files are under 9,999 lines of text, so using a
field width of 5 characters meant that we should have had one char
padding on the left edge (because we right-justify the line number).
Unfortunately when I added the right margin earlier (when I removed
the padding) I ate into the extra character's space, losing the left
margin. This put the line numbers too close to the commit column in
any file with more than 999 lines in it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The colors I originally picked out on a Mac OS X system look a
tad too dark on a Windows 2000 system; the greys are dark enough
to make it difficult to read some lines of text and the green used
to highlight the current commit was also difficult to read text on.
I also added a third grey to the mix, to try and help some files
that wind up with a number of neighboring chunks getting the same
colors.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Remove unnecessary space between columns in blame viewer
On Mac OS X the OS has "features" that like to draw thick black
borders around the text field that has focus. This is nice if
you want to know where your text is going and are blind as a bat,
but it isn't the best thing to have in a table that is being
faked through the abuse of Tk text widgets.
By setting our takefocus, highlightthickness and padx/y we can
get rid of this border and get our text widgets packed right next
to each other, with no padding between them. This makes the blame
background color smoothly run across the entire line of commit data,
line number and file content.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>