1
0
mirror of https://github.com/git/git.git synced 2024-11-18 06:03:53 +01:00
Commit Graph

329 Commits

Author SHA1 Message Date
Paul Mackerras
f806f0fba8 gitk: Handle updating with path limiting better
When updating the graph, gitk uses a git log command with commit
limiting in order to get just the new commits.  When path limiting
is also in effect, git log rewrites the parents of the commits it
outputs in order to represent just the subgraph that modifies the
listed paths, but it doesn't rewrite the parents on the boundary
of the graph.  The result is that when updating, git log does not
give gitk the information about where the new commits join in to
the existing graph.

This solves the problem by explicitly rewriting boundary parents
when updating.  If we are updating and are doing path limiting,
then when gitk finds an unlisted commit (one where git log puts a
"-" in front of the commit ID to indicate that it isn't actually
part of the graph), then gitk will execute:

    git rev-list --first-parent --max-count=1 $id -- paths...

which returns the first ancestor that affects the listed paths.
(Currently gitk executes this synchronously; it could do it
asynchronously, which would be more complex but would avoid the
possibility of the UI freezing up if git rev-list takes a long time.)

Then, if the result is a commit that we know about, we rewrite the
parents of the children of the original commit to point to the new
commit.  That is mostly a matter of adjusting the parents and children
arrays and calling fix_reversal to fix up the graph.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-02-24 12:16:56 +11:00
Paul Mackerras
841ea82449 gitk: Fix bug where arcs could get lost
Because we weren't fixing up vlastins when moving an arc from one
place to another, it was possible for us later to decide to move
an arc to the wrong place, and end up with an arc disconnected from
the rest of the graph.  This fixes it by updating vlastins when
necessary.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-02-18 10:44:33 +11:00
Paul Mackerras
585fb5985d Merge branch 'master' into dev 2008-02-16 22:24:52 +11:00
Paul Mackerras
f1bf4ee6d7 gitk: Cope better with getting commits that we have already seen
This fixes a bug in updating the graph after we have cherry-picked
a commit in gitk and then added some new stuff externally.  First,
we weren't updating viewincl with the new head added by the cherry-
pick.  Secondly, getcommitlines was doing bad things if it saw a
commit that was already in the graph (was already in an arc).  This
fixes both things.  If getcommitlines sees a commit that is already
in the graph, it ignores it unless it was not listed before and is
listed now.  In that case it doesn't assign it a new arc now, and
doesn't re-add the commit to its arc.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-02-16 22:22:22 +11:00
Johannes Sixt
6675ea4240 [PATCH] gitk: Heed the lines of context in merge commits
There is an edit box where the number of context lines can be chosen.
But it was only used when regular diffs were displayed, not for
merge commits.   This fixes it.

Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-02-14 10:49:44 +11:00
Linus Torvalds
1407ade93c [PATCH] gitk: learn --show-all output
It's really not very easy to visualize the commit walker,
because - on purpose - it obvously doesn't show the
uninteresting commits!

We will soon add a "--show-all" flag to the revision walker,
which will make it show uninteresting commits too, and they'll
have a '^' in front of them.

This is to update 'gitk' to show those negative commits in gray
to futureproof it.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-02-13 23:08:38 +11:00
Paul Mackerras
b8a938cf78 gitk: Fix cherry-picking to insert a real row not a fake row
The insertrow/removerow functions were really only suitable for
inserting/removing a fake row such as the ones used for showing
the local changes.  When used to insert a real new row from a
cherry-pick, they left things in an inconsistent state which then
caused various strange layout errors.

This renames insertrow/removerow to insertfakerow/removefakerow
and adds a new insertrow that does actually go to all the trouble
of creating a new arc and setting it up.  This is more work but
keeps things consistent.

This also fixes a bug where cherrypick was not setting mainheadid,
and one where selectline wasn't always resulting in targetrow/id
being set to the selected row/id.  Also insert/removefakerow now
adjust numcommits and call setcanvscroll.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-02-13 22:12:31 +11:00
Gerrit Pape
48750d6a84 [PATCH] gitk: properly deal with tag names containing / (slash)
When creating a tag through gitk, and the tag name includes a slash (or
slashes), gitk errors out in a popup window.  This patch makes gitk use
'git tag' to create the tag instead of modifying files in refs/tags/,
which fixes the issue; if 'git tag' throws an error, gitk pops up with
the error message.

The problem was reported by Frédéric Brière through
 http://bugs.debian.org/464104

Signed-off-by: Gerrit Pape <pape@smarden.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-02-12 10:05:59 +11:00
Steffen Prohaska
b9b86007e2 [PATCH] gitk: Add checkbutton to ignore space changes
Ignoring space changes can be helpful.  For example, a commit
claims to only reformat source code and you quickly want to
verify if this claim is true.  Or a commit accidentally changes
code formatting and you want to focus on the real changes.

In such cases a button to toggle of whitespace changes would be
quite handy.  You could quickly toggle between seeing and
ignoring whitespace changes.

This commit adds such a checkbutton right above the diff view.

However, in general it is a good thing to see whitespace changes
and therefore the state of the checkbutton is not saved. For
example, space changes might happen unintentionally.  But they are
real changes yielding different sha1s for the blobs involved.

Signed-off-by: Steffen Prohaska <prohaska@zib.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-02-12 10:05:58 +11:00
Michele Ballabio
3d2c998e30 [PATCH] gitk: Fix "Key bindings" message
The "Key bindings" message under the "Help" menu was too long
and could not be parsed by the translation engine.

Fix both issues by translating one line at a time.

Signed-off-by: Michele Ballabio <barra_cuda@katamail.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-02-12 10:05:57 +11:00
Paul Mackerras
5a7f577dce gitk: Fix bug causing Tcl error when no commits are selected
Some of the stuff that commit 31c0eaa8cc10944ebca8aa31fb59f0b77b1b6e77
added to drawvisible isn't appropriate to do when we have no commits,
and this was causing a Tcl error if gitk was invoked in such a fashion
that no commits were selected.  This fixes it by bailing out of
drawvisible early if there are no commits displayed.

Bug reported by Johannes Sixt.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-15 22:45:43 +11:00
Paul Mackerras
e7297a1c5f gitk: Fix bug where editing an existing view would cause an infinite loop
This fixes a bug where changing the commit range or file list for an
existing view and then clicking OK would cause gitk to go into an
infinite loop.  The problem was that newviewok was invoking reloadcommits
via "run reloadcommits", but reloadcommits wasn't explicitly returning
0, and whatever it was returning was causing dorunq to run it over
and over again.  This fixes it by making reloadcommits return 0.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-15 22:30:40 +11:00
Paul Mackerras
46308ea1da gitk: Select something appropriate on cherry-pick, branch reset and checkout
This makes gitk select the new commit when cherry-picking, and select
the new checked-out head when resetting or checking out a branch.
This feels more natural because the user is usually more interested
in that commit now than whatever was selected before.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-15 22:16:32 +11:00
Johannes Schindelin
646f3a149d [PATCH] gitk: make Ctrl "+" really increase the font size
Only Ctrl "=" was bound to increase the font size, probably because
English keyboards have the plus on the same key as the equal sign.
However, not the whole world is English, and at least with some
other keyboard layouts, Ctrl "+" did not work as documented.

Noticed by Stephan Hennig.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-14 08:55:04 +11:00
Paul Mackerras
3e76608d39 gitk: Select head of current branch by default
Instead of selecting the first commit that appears, this makes gitk
select the currently checked out head, if the user hasn't explicitly
selected some other commit by the time it appears.  If the head hasn't
appeared by the time the graph is complete, then we select the first
real commit.

This applies both for graph updates and when the graph is being read
in initially.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-13 17:26:30 +11:00
Paul Mackerras
17529cf9bc gitk: Fix a bug in make_disporder
The make_disporder function has an optimization where it assumed that
if displayorder was already long enough and the first entry in it for
a particular arc was non-null, then the whole arc was present.  This
turns out not to be true in some circumstances, since we can add a
commit to an arc (which truncates displayorder to the previous end of
that arc), then call make_disporder for later arcs (which will pad
displayorder with null elements), then call make_disporder for the
first arc - which won't update the null elements.

This fixes it by changing the optimization to check the last element
for the arc instead of the first.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-12 21:46:31 +11:00
Christian Stimming
7d8856d091 gitk: Update German translation.
Now 100% complete (163 strings).

Signed-off-by: Christian Stimming <stimming@tuhh.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-11 00:51:13 -08:00
Christian Stimming
55e34436e1 gitk: Fix typo in user message.
Signed-off-by: Christian Stimming <stimming@tuhh.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-11 00:51:13 -08:00
Charles Bailey
66e15a6b83 gitk: Fix the Makefile to cope with systems lacking msgfmt
The po2msg.sh script and the .gitignore in the po directory have been
shamelessly copied from the current git-gui.  This enables the top
level "make NO_MSGFMT" to work consistently for git across the git-gui
and gitk sub-projects.

This is the same effective patch that has previously been posted as a
git.git patch which more succinctly described the copying of
po/.gitignore and po/po2msg.sh from git-gui.

Signed-off-by: Charles Bailey <charles@hashpling.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-11 00:51:08 -08:00
Paul Mackerras
5be25a8f85 gitk: Fix handling of flag arguments
Despite the name, the --revs-only flag to git rev-parse doesn't make
it output only revision IDs.  It makes it output only arguments that
are suitable for giving to git rev-list.  So make start_rev_list and
updatecommits cope with arguments output by git rev-parse that aren't
revision IDs.  This way we won't get an error when an argument such as
"-300" has been given to gitk and the view is updated.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-10 21:44:39 +11:00
Paul Mackerras
6636b88ea1 Merge branch 'master' into dev 2008-01-09 14:23:30 +11:00
Gerrit Pape
f0c8426819 [PATCH] gitk: use user-configured background in view definition dialog
Have the text fields in the view definition dialog (View->New view...)
use the background color as configured through the preferences, instead
of hard-coded 'white'.

This was suggested by Paul Wise through http://bugs.debian.org/457124

Signed-off-by: Gerrit Pape <pape@smarden.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-09 14:19:47 +11:00
Christian Stimming
9e8ad094bd [PATCH] gitk: Update German translation
Signed-off-by: Christian Stimming <stimming@tuhh.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-09 14:13:22 +11:00
Christian Stimming
73c4f22850 [PATCH] gitk: Update and fix Makefile
This Makefile uses the template provided at git.git/gitk-git/Makefile
by Junio and adds the rules for the i18n files.

Signed-off-by: Christian Stimming <stimming@tuhh.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-09 14:12:18 +11:00
Paul Mackerras
476ca63dbc gitk: Index [fnvr]highlights by id rather than row
This means that we don't have to keep clearing them out whenever we
change the row numbers for some commits.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-07 22:16:31 +11:00
Paul Mackerras
7e92e257af Merge branch 'master' into dev 2008-01-06 22:19:03 +11:00
Paul Mackerras
b039f0a62d gitk: Restore some widget options whose defaults changed in Tk 8.5
The default options for panedwindows in Tk 8.5 make the sash
virtually invisible -- the handle is not shown and the relief is
flat.  This puts the defaults back to showing the handle and a
raised relief on the sash, as in Tk 8.4.

This uses the option command to do this, and also uses the option
command to set the default font for various UI elements to the
UI font ("uifont").

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-06 15:55:39 +11:00
Paul Mackerras
c8c9f3d9cc gitk: Fix potential bug with fake commit IDs in renumbervarc
When a fake row is added, we add its (fake) ID to the children list
for its (fake) parent.  If renumbervarc were to then renumber the
parent it would incorrectly use the fake child.  This avoids the
problem by adding a last_real_child procedure which won't return
a fake ID, and using it in renumbervarc.  For symmetry this also adds
a first_real_child procedure and uses it in ordertoken.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-06 13:54:58 +11:00
Paul Mackerras
cd2bcae798 gitk: Fix a couple of bugs in the find function
First, findmore would sometimes get a Tcl error due to relying on
varcorder and vrownum having valid values for the rows being searched,
but they may not be valid unless update_arcrows is called, so this
makes findmore call update_arcrows if necessary.

Secondly, in the "touching paths" and "adding/removing string" modes,
findmore was treating fhighlights($row) == -1 as meaning the row
matches, whereas it only means that we haven't received an answer from
the external git diff-tree process about it yet.  This fixes it.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-02 21:44:06 +11:00
Paul Mackerras
42a671fc00 gitk: Fix some corner cases in the targetid/targetrow stuff
* Make sure targetrow is never >= numcommits
* Don't try to do anything about the target row if the targetid is
  no longer in the view; it'll just cause Tcl errors
* In insertrow, increment targetrow if we are inserting the fake
  commit at or before the target row
* In removerow, if we are removing the target row, make it the next
  one instead.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-01-02 09:59:39 +11:00
Paul Mackerras
33de5c0da8 Merge branch 'master' into dev 2007-12-31 12:49:22 +11:00
Paul Mackerras
a495e5a908 gitk: Recode de.po to UTF-8
Somehow de.po got recoded to latin-1 in the process of committing it.
This recodes it back to UTF-8.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-31 12:49:13 +11:00
Paul Mackerras
31c0eaa8cc gitk: Keep the same commits visible as other commits come in
Since commits come in out of order and get sorted as we see them,
we can have commits coming in and being placed before the commits
that are visible in the graph display pane.  Previously we just
displayed a certain range of row numbers, meaning that when
incoming commits were placed before the displayed range, the
displayed commits were displaced downwards.  This makes it so
that we keep the same set of commits displayed, unless the user
explicitly scrolls the pane, in which case it scrolls as expected.

We do this by having a "target" commit which we try to keep in the
same visible position.  If commits have come in before it we scroll
the canvases by the number of rows that it has moved in the display
order.

This also fixes a bug in rowofcommit where it would test
cached_commitrow before possibly calling update_arcrows, which is
where cached_commitrow gets invalidated if things have changed.
Now we call update_arcrows if necessary first.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-30 22:41:14 +11:00
Paul Mackerras
eb5f8c9c00 gitk: Don't try to show local changes from a head that isn't shown
When updating the display, if the checked-out head has moved on and
isn't currently shown, and there are local changes, we could try to
insert a fake row with a parent that isn't displayed, leading to a
Tcl error.  This is because we check whether the checked-out head
is displayed before rereading the references (which is when we discover
that the head has moved).  This fixes it.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-29 21:13:34 +11:00
Paul Mackerras
fc2a256f4a gitk: Fix another collection of bugs
* Fixed a bug that occasionally resulted in Tcl "can't use empty string
  as argument to incr" errors - rowofcommit was sometimes not calling
  update_arcrows when it needed to.

* Fixed a "no such element in array" error when removing a fake row,
  by unsetting currentid and selectedline in removerow if the row we
  are removing is the currently selected row.

* Made the "update commits" function always do "reread references".

* Made dodiffindex et al. remove the fake row(s) if necessary.

* Fixed a bug where clicking on a row in the graph display pane didn't
  account for horizontal scrolling of the pane.

* Started changing things that cached information based on row numbers
  to use commit IDs instead -- this converts the "select line" items
  that are put into the history list to use "select by ID" instead.

* Simplified redrawtags a bit, and fixed a bug where it would use the
  mainfont for working out how far it extends to the right in the graph
  display pane rather than the actual font (which might be bold).

* Fixed a bug where "reread references" wouldn't notice if the currently
  checked-out head had changed.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-26 23:03:43 +11:00
Paul Mackerras
00abadb9dd Merge branch 'master' into dev 2007-12-20 10:25:50 +11:00
Christian Stimming
eadcac9218 [PATCH] gitk i18n: Recode gitk from latin1 to utf8 so that the (c) copyright character is valid utf8.
When using translations, the target language must be encoded in utf-8
because almost all target languages will contain non-ascii characters.
For that reason, the non-translated strings should be in utf-8 as well
so that there isn't any encoding mixup inside the program.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-20 10:19:07 +11:00
Christian Stimming
b007ee20dc [PATCH] gitk i18n: More markup -- various options menus
Signed-off-by: Christian Stimming <stimming@tuhh.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-20 10:19:07 +11:00
Christian Stimming
15bc7bae91 [PATCH] gitk i18n: Initial German translation
Signed-off-by: Christian Stimming <stimming@tuhh.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-20 10:19:07 +11:00
Christian Stimming
d990cedf9c [PATCH] gitk i18n: Markup several strings for translation
This just marks up plain strings, that aren't used in any unusual way.

Signed-off-by: Christian Stimming <stimming@tuhh.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-20 10:19:07 +11:00
Christian Stimming
663c3aa9c8 [PATCH] gitk i18n: Import msgcat for message string translation; load translation catalogs
By setting the environment variable GITK_MSGSDIR, one can manually set
the directory where the .msg files are located.  This is quite handy
during development with GITK_MSGSDIR=po.

Signed-off-by: Christian Stimming <stimming@tuhh.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-20 10:19:07 +11:00
Christian Stimming
25ec938484 [PATCH] gitk i18n: Add Makefile with rules for po file creation and installation
The compiled .msg files will be installed into $(sharedir)/gitk/lib/msgs
according to Junio's mailing list proposal on 2007-07-28.

Signed-off-by: Christian Stimming <stimming@tuhh.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-20 10:19:07 +11:00
Paul Mackerras
f3ea5ede23 gitk: Implement date mode in the new framework
This restores date mode, which lists commits by date, as far as possible
given the constraint that parents come after all their children.  To
implement this in the new framework, we (1) only join a new commit onto
an existing arc if the arc is the last arc created, (2) treat arcs as
seeds unless they have a child arc that comes later, and (3) never
decrease the token value for an arc.

This means we get lots of "seeds", which exposed some quadratic behaviour
in adding and removing seeds.  To fix this, we add a vbackptr array, which
points to the arc whose vleftptr entry points to us, and a vlastins array,
which shows where in an arc's vdownptr/vleftptr list we last inserted a
parent, which acts as a hint of a good place to start looking for where to
insert a new child.

This also ensures the children array elements stay in sorted order at all
times.  We weren't resorting the children lists when reassigning tokens
in renumbervarc.  Since the children lists are now always sorted, we don't
have to search through all elements to find the one with the highest token;
we can just use the last element.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-20 10:03:35 +11:00
Paul Mackerras
24f7a667e6 gitk: More bug fixes and cleanups
* Add/remove fake commits (for local changes) when updating the view
  even if nothing else has changed.
* Get rid of unused getdbg variable.
* Get rid of vseeds and uat.
* Fix bug where removerow would throw a "no such element in array" error.
* Clear out cached highlights when line numbers change.
* Make dodiffindex remove the fake commit rows if they currently exist
  but there are now no local changes.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-19 09:35:33 +11:00
Paul Mackerras
e5b37ac1ec gitk: Fix more bugs resulting in Tcl "no such element in array" errors
First, update_arcrows was being overly aggressive in trimming
displayorder, resulting in calls to rowofcommit sometimes trimming off
commits that layoutrows had asked for in make_disporder and was relying
on having present.  This adds a vrowmod($view) variable that lets
update_arcrows be more precise in trimming off the invalid bits of
displayorder (and it also simplifies the check in make_disporder).
This modifies modify_arc and its callers so that vrowmod($view) is
updated appropriately.

Secondly, we were sometimes calling idcol with $i==-1, which resulted
in a call to ordertoken with the null string.  This fixes it by
forcing $i to 0 if it is less than zero.

This also fixes a possible infinite recursion with rowofcommit and
update_arcrows calling each other ad infinitum.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-12 18:13:51 +11:00
Paul Mackerras
0c27886e46 gitk: Fix a couple of bugs
First, if we invalidate the layout for all rows (i.e. from row 0 on),
we were calling undolayout with an empty string as the argument.
Second, the comparison in make_disporder that tests if we need to
call update_arcrows was the wrong way around.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-11 20:09:53 +11:00
Paul Mackerras
9257d8f761 gitk: Compute row numbers and order tokens lazily
Instead of computing ordertok values and arc row numbers in
getcommitlines, this defers computing them until they are needed.
So getcommitlines no longer calls update_arcrows; instead it gets
called from rowofcommit and make_disporder.  Things that modify arcs
now call modify_arc instead of setting vtokmod/varcmod directly,
and modify_arc does the undolayout that used to be in update_arcrows.

Also, idcol and make_idlist now use a new ordertoken function instead
of the ordertok variable.  ordertoken uses ordertok as a cache, but
can itself compute the ordering tokens from scratch.  This means that
the ordering tokens (and hence the layout of the graph) is once again
determined by the topological ordering we put on the graph, not on the
order in which we see the commits from git log, which improves the
appearance of the graph.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-11 10:45:38 +11:00
Paul Mackerras
f78e7ab736 gitk: Fix bug in parsing multiple revision arguments
If the user specified multiple revisions arguments on the command
line or for a view, we were passing the whole list of arguments to
git rev-parse as a single argument, and thus git rev-parse didn't
interpret it as revisions.  This fixes it by adding an eval so the
arguments get passed to git rev-parse as separate arguments.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-07 20:50:09 +11:00
Paul Mackerras
38dfe93916 gitk: Fix bug in assigning row numbers to arcs
We weren't setting vtokmod and varcmod in renumbervarc, so after a
call to renumbervarc we sometimes weren't reassigning row numbers to
all the arcs whose row numbers had changed.  This fixes it.

This also collapses layoutmore and showstuff into one procedure and
gets rid of the phase variable.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-06 20:50:31 +11:00
Paul Mackerras
7fcc92bff4 gitk: Use git log without --topo-order and reorganize the commits ourselves
This very large patch implements code to organize the commits from
git log into "arcs" (sequences of commits where each pair of adjacent
commits are the only parent and child of each other), and orders the
arcs so as to get a topological ordering of the commits.  This means
we can use git log without --topo-order and display the commits as we
get them, incrementally, which makes the cold-cache start up time much
faster, particularly on unpacked repos.

One beneficial effect of this is that the File->Update menu item now
just adds any new commits to the existing graph instead of rereading
the whole thing from scratch, which is much faster.  (If you do want
to reread the whole graph from scratch you can use File->Reload.)

At an implementation level, this means that the displayorder and
parentlist lists are no longer fully valid at all times, and the
commitrow array has gone.  New procedures commitinview and commitonrow
replace the commitrow array, and make_disporder ensures that
displayorder and parentlist are valid for a range of rows.

The overall time to load the kernel repository has gone up a bit, from
~9 seconds to ~11 seconds on my G5, but I think that is worth it given
that the time to get a window up with commits displayed in it has gone
from ~3 seconds to under 1 second.

Signed-off-by: Paul Mackerras <paulus@samba.org>
2007-12-03 10:33:01 +11:00