I totally missed this obvious optimization in the checkout code path.
If our current repository HEAD is actually at the commit we are moving
to, and we agreed to perform this switch earlier, then we have no files
to update in the working directory and any stale mtimes are simply not
of consequence right now. We can pretend like we ran a read-tree and
skip right into the post-read-tree work, such as updating the branch
and setting the symbolic-ref.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we are given a merge type we don't understand in checkout_op there
is probably a bug in git-gui somewhere that allowed this unknown merge
strategy to come into this part of the code path. We currently only
recognize three merge types ('none', 'ff' and 'reset') but are going
to be supporting more in the future. Rather than keep editing this
message I'm going with a very generic "Uh, we don't do that!" type of
error.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
In both the ff and reset merge_types supported by checkout_op the
result is the same if the merge base of our target commit and the
existing commit is the existing commit: its a fast-forward as the
existing commit is fully contained in the target commit.
This minor cleanup in logic will make it easier to implement a
new kind of merge_type that actually merges the two trees with a
real merge strategy, such as git-merge-recursive.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
I've decided to teach checkout_op how to perform more than just a
fast-forward and reset type of merge. This way we can also do a full
recursive merge even when we are recreating an existing branch from
a remote. To help with that process I'm saving the merge-base we
computed during the ff/reset/fail decision process, in case we need
it later on when we actually start a true merge operation.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
A few users have been seeing crashes in Tk when using the undo key
binding to undo the last few keystroke events in the commit buffer.
Unfortunately that means the user loses their commit message and
must start over from scratch when the user restarts the process.
git-gui now saves the user's commit message buffer every couple of
seconds to a temporary file under .git (specifically .git/GITGUI_BCK).
At exit time we rename this file to .git/GITGUI_MSG if there is a
message, the file exists, and it is currently synchronized with the
Tk buffer. Otherwise we do our usual routine of saving the Tk buffer
to .git/GITGUI_MSG and delete .git/GITGUI_BCK, if it exists.
During startup we favor .git/GITGUI_BCK over .git/GITGUI_MSG. This
way a crash doesn't take out the user's message buffer but instead
will cause the user to lose only a few keystrokes. Most people do
not type more than 200 WPM, and with 30 possible saves per minute
we are unlikely to lose more than 7 words.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
I'm working on refactoring the UI of the merge dialog, because as it
currently stands the dialog is absolutely horrible, especially when
you have 200+ branches available from a single remote system.
In that refactoring I plan on using the choose_rev widget to allow
the user to select exactly which branch/commit they want to merge.
However since that only selects a single commit I'm first removing
the code that supports octopus merges.
A brief consultation on #git tonight seemed to indicate that the
octopus merge strategy is not as useful as originally thought when
it was invented, and that most people don't commonly use them. So
making users fall back to the command line to create an octopus is
actually maybe a good idea here, as they might think twice before
they use it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we haven't yet loaded any commit information for a given line but
our tooltip timer fired and tried to draw the tooltip we shouldn't;
there is nothing to show.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This is a essentially a copy of Paul Mackerras encoding support from
gitk. I stole the code from gitk commit fd8ccbec4f, as Paul has
already done all of the hard work setting up this translation table.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Running global takes slightly longer than just accessing the variable
via its package name, especially if the variable is just only once in
the procedure, or isn't even used at all in the procedure. So this is
a minor cleanup for some of our commonly invoked procedures.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Users who merge often may want to access the merge action quickly,
so we now bind M to the merge action.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Users shouldn't see this menu option if they startup a browser or
blame from the command line, especially if they are doing so on a
bare repository.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
I've found that the class code makes it a whole lot easier to create
more complex GUI code, especially the dialogs. So before I make any
major improvements to the merge dialog's interface I'm going to first
switch it to use the class system, so the code is slightly cleaner.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Showing only five lines of heads/tags is not very useful to a user
when they have about 10 branches that match the filter expression.
The list is just too short to really be able to read easily, at
least not without scrolling up and down. Expanding the list out
to 10 really makes the revision picker easier to read and access,
as you can read the matching branches much more quickly.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Users who are new to Git may not realize that visualizing things in
a repository involves looking at history. Adding in a small amount
of text to the menu items really helps to understand what the action
might do, before you invoke it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
We now allow users to pick which commit they want to browse through
our revision picking mega-widget. This opens up in a dialog first,
and then opens a tree browser for that selected commit. It is a very
simple approach and requires minimal code changes.
I also clarified the language a bit in the Repository menu, to show
that these actions will access files. Just in case a user is not
quite sure what specific action they are looking for, but they know
they want some sort of file thing.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Like our blame subcommand the browser subcommand now accepts both
a revision and a path, just a revision or just a path. This way
the user can start the subcommand on any branch, or on any subtree.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
A long time ago Linus Torvalds tried to run git-gui on a bare
repository to look at the blame viewer, but it failed to start
because we required that the user run us only from within a
working directory that had a normal git repository associated
with it.
This change relaxes that requirement so that you can start the
tree browser or the blame viewer against a bare repository. In
the latter case we do require that you provide a revision and a
pathname if we cannot find the pathname in the current working
directory.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
By moving our feature option determination up before we look for GIT_DIR
we can make a decision about whether or not we need a working tree up
front, before we look for GIT_DIR. A future change could then allow
us to start in a bare Git repository if we only need access to the ODB.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
I'm moving the code related to looking to see if we should GC now
into a procedure closer to where it belongs, the database module.
This reduces our script by a few lines for the single commit case
(aka citool). But really it just is to help organize the code.
We now perform the check after we have been running for at least
1 second. This way the main window has time to open up and our
dialog (if we open it) will attach to the main window, instead of
floating out in no-mans-land like it did before on Mac OS X.
I had to use a wait of a full second here as a wait of 1 millisecond
made our console install itself into the main window. Apparently we
had a race condition with the console code where both the console and
the main window thought they were the main window.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Some users may do odd things, like tag their own private version of
Git with an annotated tag such as 'testver', then compile that git
and try to use it with git-gui. In such a case `git --version` will
give us 'git version testver', which is not a numeric argument that
we can pass off to our version comparsion routine.
We now check that the cleaned up git version is a going to pass the
version comparsion routine without failure. If it has a non-numeric
component, or lacks at least a minor revision then we ask the user to
confirm they really want to use this version of git within git-gui.
If they do we shall assume it is git 1.5.0 and run with only the code
that will support.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Instead of running a full git-count-objects to count all of the loose
objects we can get a reasonably close approximation by counting the
number of files in the .git/objects/42 subdirectory. This works out
reasonably well because the SHA-1 hash has a fairly even distribution,
so every .git/objects/?? subdirectory should get a relatively equal
number of files. If we have at least 8 files in .git/objects/42 than it
is very likely there is about 8 files in every other directory, leaving
us with around 2048 loose objects.
This check is much faster, as we need to only perform a readdir of
a single directory, and we can do it directly from Tcl and avoid the
costly fork+exec.
All of the credit on how clever this is goes to Linus Torvalds; he
suggested using this trick in a post commit hook to repack every so
often.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we have more than our desired number of objects and we try to
open the "Do you want to repack now?" dialog we cannot include a
-parent . argument if the main window has not been mapped yet.
On Mac OS X it appears this window isn't mapped right away, so we
had better hang avoid including it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Not every caller of 'git' or 'git_pipe' wants to use nice to lower the
priority of the process its executing. In many cases we may never use
the nice process to launch git. So we can avoid searching our $PATH
to locate a suitable nice if we'll never actually use it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The git-gui version check doesn't handle versions of the form
n.n.n.GIT which you can get by installing from an tarball produced by
git-archive.
Without this change you get an error of the form:
'Error in startup script: expected version number but got "1.5.3.GIT"'
Signed-off-by: Julian Phillips <julian@quantumfyre.co.uk>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
On Windows (which includes Cygwin) Tcl defaults to leaving the EOF
character of input file streams set to the ASCII EOF character, but
if that character were to appear in the data stream then Tcl will
close the channel early. So we have to disable eofchar on Windows.
Since the default is disabled on all platforms except Windows, we
can just disable it everywhere to prevent any sort of read problem.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
My prior change to allow git-gui to run with a version of Git
that was built from a working directory that had uncommitted
changes didn't account for the pattern starting with -, and
that confused Tcl.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If the user is running a 'dirty' version of git (one compiled in a
working directory with modified files) we want to just assume it
was a committed version, as we really only look at the part that
came from a real annotated tag anyway.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This cat-file was done on maint, where we did not have git_read
available to us. But here on master we do, so we should make
use of it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
From Johannes Sixt <J.Sixt@eudaptics.com>:
> It seems that MSYS's wish does some quoting for Bourne shells,
> in particular, escape the first '{' of the "^{tree}" suffix, but
> then it uses cmd.exe to run "git rev-parse". However, cmd.exe does
> not remove the backslash, so that the resulting rev expression
> ends up in git's guts as unrecognizable garbage: rev-parse fails,
> and git-gui hickups in a way that it must be restarted.
Johannes originally submitted a patch to this section of commit.tcl
to use `git rev-parse $PARENT:`, but not all versions of Git will
accept that format. So I'm just taking the really simple approach
here of scanning the first line of the commit to grab its tree.
About the same cost, but works everywhere.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Because we are trying to execute /bin/sh we know it must be a real
Windows executable and thus ends with the standard .exe suffix.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
We cannot execute the git directory, it is not a valid Tcl command
name. Instead we just want to pass it as an argument to our sq
proc.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we get more than 80 characters of text in a single line odds
are it is output from git-fetch or git-push and its showing a
lot of detail off to the right edge that is not so important to
the average user. We still want to make sure we show everything
we need, but we can get away with that information being off to
the side with a horizontal scrollbar.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Our file browser was showing bad output as it did not properly buffer
a partial record when read from `ls-tree -z`. This did not show up on
my Mac OS X system as most trees are small, the pipe buffers generally
big and `ls-tree -z` was generally fast enough that all data was ready
before Tcl started to read. However on my Cygwin system one of my
production repositories had a large enough tree and packfile that it
took a couple of pipe buffers for `ls-tree -z` to complete its dump.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
We now embed any GIT_* and SSH_* environment variables as well as
the path to the git wrapper executable into the Mac OS X .app file.
This should allow us to restore the environment properly when
we restart.
We also try to use proper Bourne shell single quoting when we can,
as this avoids any sort of problems that might occur due to a path
containing shell metacharacters.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Now that we are pretty strict about setting up own absolute paths to
any git helper (saving a marginal runtime cost to resolve the tool)
we can do the same in our console widget by making sure all console
execs go through git_read if they are a git subcommand, and if not
make sure they at least try to use the Tcl 2>@1 IO redirection if
possible, as it should be faster than |& cat.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we cannot locate a .exe for a git tool that we want to run than
it may just be a Bourne shell script as these are popular in Git.
In such a case the first line of the file will say "#!/bin/sh" so
a UNIX kernel knows what program to start to parse and run that.
But Windows doesn't support shbang lines, and neither does the Tcl
that comes with Cygwin.
We can pass control off to the git wrapper as that is a real Cygwin
program and can therefore start the Bourne shell script, but that is
at least two fork+exec calls to get the program running. One to do
the fork+exec of the git wrapper and another to start the Bourne shell
script. If the program is run multiple times it is rather expensive
as the magic shbang detection won't be cached across executions.
On MinGW/MSYS we don't have the luxury of such magic detection. The
MSYS team has taught some of this magic to the git wrapper, but again
its slower than it needs to be as the git wrapper must still go and
run the Bourne shell after it is called.
We now attempt to guess the shbang line on Windows by reading the
first line of the file and building our own command line path from
it. Currently we support Bourne shell (sh), Perl and Python. That
is the entire set of shbang lines that appear in git.git today.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
We know that the version subcommand of git is special. It does not
currently have an executable link installed into $gitexecdir and we
therefore would never match it with one of our file exists tests.
So we forward any invocations to it directly to the git wrapper, as
it is a builtin within that executable.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we cannot locate a command in $gitexecdir on our own then it may
just be because we are supposed to run it by `git $name` rather than
by `git-$name`. Many commands are now builtins, more are likely to
go in that direction, and we may see the hardlinks in $gitexecdir go
away in future versions of git.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The master Makefile in git.git installs gitk into bindir, not
gitexecdir, which means gitk is located as a sibling of the git
wrapper and not as though it were a git helper tool.
We can also avoid some Tcl concat operations by letting eval do
all of the heavy lifting; we have two proper Tcl lists ($cmd and
$revs) that we are joining together and $revs is currently never
an empty list.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Rather than making the C library search for git every time we want
to execute it we now search for the main git wrapper at startup, do
symlink resolution, and then always use the absolute path that we
found to execute the binary later on. This should save us some
cycles, especially on stat challenged systems like Cygwin/Win32.
While I was working on this change I also converted all of our
existing pipes ([open "| git ..."]) to use two new pipe wrapper
functions. These functions take additional options like --nice
and --stderr which instructs Tcl to take special action, like
running the underlying git program through `nice` (if available)
or redirect stderr to stdout for capture in Tcl.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Sometimes switching between branches can take more than a second or
two, in which case `git checkout` would normally have shown a small
progress meter to the user on the terminal to let them know that we
are in fact working, and give them a reasonable idea of when we may
finish.
We now do obtain that progress meter from read-tree -v and include
it in our main window's status bar. This allows users to see how
many files we have checked out, how many remain, and what percentage
of the operation is completed. It should help to keep users from
getting bored during a large checkout operation.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Now that we have a fancy status bar mega-widget we can reuse that
within our main window. This opens the door for implementating
future improvements like a progress bar.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Our blame viewer has had a very fancy progress bar at the bottom of
the window that shows the current status of the blame engine, which
includes the number of lines completed as both a text and a graphical
meter. I want to reuse this meter system in other places, such as
during a branch switch where read-tree -v can give us a progress
meter for any long-running operation.
This change extracts the code and refactors it as a widget that we
can take advantage of in locations other than in the blame viewer.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If the user double clicks a branch in the checkout dialog then they
probably want to start the checkout process on that branch. I found
myself doing this without realizing it, and of course it did nothing
as there was no action bound to the listbox's Double-Button-1 event
handler. Since I did it without thinking, others will probably also
try, and expect the same behavior.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we have specifications listed in our revision picker mega-widget
then we should default the selection within that widget to the first
ref available. This way the user does not need to use the spacebar
to activate the selection of a ref within the box; instead they can
navigate up/down with the arrow keys and be done with it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If the end-user feeds us an abbreviated SHA-1 on the command line for
`git gui browser` or `git gui blame` we now unabbreviate the value
through `git rev-parse` so that the title section of the blame or
browser window shows the user the complete SHA-1 as Git determined
it to be.
If the abbreviated value was ambiguous we now complain with the
standard error message(s) as reported by git-rev-parse --verify,
so that the user can understand what might be wrong and correct
their command line.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This is a major rewrite of the way we perform switching between
branches and the subsequent update of the working directory. Like
core Git we now use a single code path to perform all changes: our
new checkout_op class. We also use it for branch creation/update
as it integrates the tracking branch fetch process along with a
very basic merge (fast-forward and reset only currently).
Because some users have literally hundreds of local branches we
use the standard revision picker (with its branch filtering tool)
to select the local branch, rather than keeping all of the local
branches in the Branch menu. The branch menu listing out all of
the available branches is simply not sane for those types of huge
repositories.
Users can now checkout a detached head by ticking off the option
in the checkout dialog. This option is off by default for the
obvious reason, but it can be easily enabled for any local branch
by simply checking it. We also detach the head if any non local
branch was selected, or if a revision expression was entered.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
