Re: [ANNOUNCE qgit-0.95]

Re: [ANNOUNCE qgit-0.95]

[Marco Costalba]

Petr Baudis wrote:

>
>Yes, it is several times faster than gitk, good work. Had to do this in
>order to be able to compile it with gcc-3.3.6, though (it complained
>about goto'ing after initialization of nc - that's bogus since there's
>the return, but...):


thanks, applied.

>
>Besides that, this is what I don't like about qgit:
>
>* It'd be nice to be able to choose to see all commits in the initial
>dialog by single click (I know about --all). Perhaps a radio buttons
>choosing between all and selected? Also, you might add the "diff against
>working copy" option to that dialog as well.

Yes.

>
>* The graph column is too narrow. It should be auto-sized so that the
>graph fits in, or at least there should be some clear indication that
>the graph does not fit to the column at the given point. This confused
>me a lot at first.

Auto fit the column width to the graph can be not always the best choice, see
0572e3da3ff5c3744b2f606ecf296d5f89a4bbdf (aka Linux v2.6.13-rc7) commit to see what I mean.

But an inidcation that the graph column is resizable is a good thing.

>
>* Could you make the grey background for odd commits span to the whole
>line, including the commit graph?
>

Yes, I am not sure of the results, there are already a lot of colors in the graphs, but I will
check.


>* The commit time is relative to now, which makes no sense to me. Also,
>it is in the second column instead of the last one like in gitk, which
>seems better to me. At least, the column is too narrow and then it
>blends together with the commit title.
>

I took the layout from gitweb where the order is -relative time - author - short log.
But this is the second time someone complains about time column so I will
send to last column: graph - shortlog - author - time


>* In the filter radio buttons groups, only first few letters of the
>labels are visible, and apparently no tooltips.

Yes. I will choose the tooltips way to save space, that line its already too long, this is
the reason I try to avoid adding buttons.

> Same problem in the
>settings dialog, most of the options have the labels cut around 2/3, and
>the window is non-resizable on top of that - why? I hate non-resizable
>windows, especially because they are usually too small. :-)
>

This is strange, I have resizable setting window and no cutted labels.......
This is for sure a bug but I can't reproduce on my box.

>* Single-clicking at a commit produces a significantly slower response
>than in gitk, where I see the difflist and stuff nearly instantly - it
>takes several hundreds of ms in qgit. That's quite annoying.
>

Please, is flag edit->settings->general->'load file names in background' set?


>* Getting to the diff view was non-obvious for me. It'd be nice to have
>some [diff] button as well somewhere. Or you could also show the diff in
>the bottom part of screen in the commit view, I think gitk solved this
>nicely.

I have tought a lot how to pass to the user the information that to see a diff you have to 
double click on the commit:

1) Adding a button with a 'double click' tooltip? --> use space forever for a one time
    information (once you know you can double click the button is useless, the space not).

2) Adding a menu entry with a tool tip? can be.

3) Add a tool tip directly on the commit line: can became annoying really soon.

4) Show always diff in commit info window? can be time consuming for merges (gitk doesn't shows
merges diff) and the space is limited or you need to resize forth and then back the commit info
window. Perpahs it's only me, but I need the whole screen to go deep in a diff when I really need
to understand the change.

In any case you are right, something has to be made.

>
>* Clicking on the file was supposed to bring some annotated view, but if
>it is so, it should write "Annotate" in the window title, and should
>indicate that it is computing it on the background (and is really slow
>in that, or I don't know, but always only single revision was shown
>there). It is unclear what the "pin file" checkbox is supposed to mean,
>and the whole dialog is just very confusing. :-)
>

Yes annotate can be slow with linux tree with some vip files ;-)

Please try with a small project or with a seldom
touched file: annotation should appear left aligned.

An example can be file daemon.c in git archive, you can reach 
from commit f8ff0c0641a14770a2214fffbd4271b1ea3a0d61

When you browse the file history list on top, the corresponding main view  
commit changes as does the file revision content: "pin file" checkbox simply avoids 
the file viewer content to change, only main view is updated.

This can be useful when you double click on the left align annotation number to jump to commit,
preserving the file view.


>* The lane information might be available as a tooltip in addition to
>right-clicking on it - that wouldn't occur to me.
>

Yes...now I see its also a little bit broken....bad luck ;-)


>* Why is the Edit menu aligned to the right?
>

Is it? really? I cannot reproduce, Its the only menu without shortcuts, but 
is correctly aligned here.

>* Well - I don't like the line graphics and prefer gitk's. This is
>perhaps a matter of taste, though. I find the graphics to be too tiny,
>thin and not so clear as gitk's, which makes it harder to see the commit
>flow. I also like diagonal lines more than angles, but that might be
>just because I'm used to them.
>

Diagonal line could be nicer but doesn't leave you play some tricks to greatly speed up
graph drawing. I really like those tricks ;-)

Then when the things get tough, with a lot of lines, the more 'regular' and geometrical
layout of qgit could help...but this is just a matter of taste.

>
>
>Nice work otherwise. :-)
>

Thanks for your very good feedback.



__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

Re: [ANNOUNCE qgit-0.95]

[Josef Weidendorfer]

Hi,

On Sunday 25 September 2005 07:52, Marco Costalba wrote:
> >* The commit time is relative to now, which makes no sense to me. Also,
> >it is in the second column instead of the last one like in gitk, which
> >seems better to me. At least, the column is too narrow and then it
> >blends together with the commit title.

I vote for absolute time, and the date column at the end, too.
Relative has no meaning when searching for an old commit.
Besides, QT allows to reorder the columns with the mouse (the order should be
saved in a config file at end to be persistant over program runs).

> >* Getting to the diff view was non-obvious for me. It'd be nice to have
> >some [diff] button as well somewhere. Or you could also show the diff in
> >the bottom part of screen in the commit view, I think gitk solved this
> >nicely.
>
> I have tought a lot how to pass to the user the information that to see a
> diff you have to double click on the commit:

I like the gitk solution better. 
What about making the commit list a QDockWindow, which can be docked to either 
side of the window (default: top as currently), but also made a floating 
window, so that the commit diff gets the whole main window?

> commit changes as does the file revision content: "pin file" checkbox
> simply avoids the file viewer content to change, only main view is updated.

The "pin" action is really not-obvious. Why not open new windows for different
file annotations? Or one window with tabs for each file?
Perhaps make the main window contain commit diffs and file annotations in 
tabs?

> Diagonal line could be nicer but doesn't leave you play some tricks to
> greatly speed up graph drawing. I really like those tricks ;-)

I think that diagonal lines as in gitk make it way easier to get an overview.
What makes drawing of diagonal lines slow?
It should be quite fast to subclass QListViewItem for commit entries and
overwrite QListViewItem::paintCell to use your own drawing; paintCell is 
called for visible entries only.

Another wish: The tag/head markers in gitk are really good. In qgit, I only 
get another background color, and miss the name.
You can do the same drawing as in gitk via paintCell, too.

> >Nice work otherwise. :-)

Yes, really nice.

Josef

Re: [ANNOUNCE qgit-0.95]

[Marco Costalba]

Robin Farine wrote:

>Josef Weidendorfer <Josef.Weidendorfer <at> gmx.de> writes:    
>  
>
>>On Sunday 25 September 2005 09:07, Marco Costalba wrote:    
>>
>>>>>src/rangeselectbase.h QSettings: error creating /.qt    
>>>>>QSettings: error creating /.qt    
>>>
>>>...    
>>>QSettings are there from day one :-<    
>>>I am not able to let them disappear.....very bad. In any case should be    
>>>harmless.    
>>
>
>    
>With SCons, key-value pairs that are meant to appear in process    
>environment need to be added to the env['ENV'] mapping. In this  
>case:  
>   
>    env = Environment(...)  
>    env['ENV']['HOME'] = os.environ['HOME']   
>   
>Environment() creates an SCons build environment which is unrelated  
>to subprocess environments (even though the same word is used :)).  
>As a special case, SCons handles the mapping assigned to env['ENV']  
>as process environment for subprocesses.  
>  
>Hope this helps,  
>  

This helps a lot!

I added the line

env['ENV']['HOME'] = os.environ['HOME'] 

to my SConstruct and the warning disappeard :-)  :-)  :-)

This fix is greatly appreciated, thanks a lot Robin.

Marco

P.S: I have just pushed the fix, togheter with some other updates, 
to http://digilander.libero.it/mcostalba/qgit.git



		
__________________________________ 
Yahoo! Mail - PC Magazine Editors' Choice 2005 
http://mail.yahoo.com

Re: [ANNOUNCE qgit-0.95]

[Marco Costalba]

Josef Weidendorfer wrote:

>Hi,
>
>On Sunday 25 September 2005 07:52, Marco Costalba wrote:
>
>>>* The commit time is relative to now, which makes no sense to me. Also,
>>>it is in the second column instead of the last one like in gitk, which
>>>seems better to me. At least, the column is too narrow and then it
>>>blends together with the commit title.
>
>
>I vote for absolute time, and the date column at the end, too.


...and 'date column at the end' wins for 3 votes against 0 ;-)


>Besides, QT allows to reorder the columns with the mouse (the order should be
>saved in a config file at end to be persistant over program runs).
>

Yes, good idea.

>>>* Getting to the diff view was non-obvious for me. It'd be nice to have
>>>some [diff] button as well somewhere. Or you could also show the diff in
>>>the bottom part of screen in the commit view, I think gitk solved this
>>>nicely.
>>
>>I have tought a lot how to pass to the user the information that to see a
>>diff you have to double click on the commit:
>
>
>I like the gitk solution better. 
>What about making the commit list a QDockWindow, which can be docked to either 
>side of the window (default: top as currently), but also made a floating 
>window, so that the commit diff gets the whole main window?
>

Should be nice, but it's a bit of work...maybe I don't know very well QDockWindow.

Peraphs is the commit info pane at the bottom left that could be implemented as a 
QDockWindow with the diff attached below the commit info gitk like.....


>
>I think that diagonal lines as in gitk make it way easier to get an overview.
>What makes drawing of diagonal lines slow?

with diagonal line you cannot draw graphs one rev at the time from left to right but you
need some information from previous one and eventually pass some information to next one.

The fact is diagonal lines are originated in a diffrent line from what you are drawing and
you need to bring with you that piece of information.

More, in qgit pixmaps are precalculated in main view ctor and filled in an array, so only 
array indexing is used to retrive and copy correct pixmap according to proper lane type.

Adding diagonals pixmaps is not obvious and not simple, at least for me. I tell you this 
because I have tried but the design of graph function became overly complex.


>It should be quite fast to subclass QListViewItem for commit entries and
>overwrite QListViewItem::paintCell to use your own drawing; paintCell is 
>called for visible entries only.
>

QListViewItem is already subclassed to paint odd/even lines background and tags/heads colors.
Pixmaps are not painted but, as told before, directly copied from a fixed pixmaps array this 
is also done in paintCell(). See ListViewLogItem::paintCell() in mainimpl.cpp if interested.

Because paintCell() is called for visible entries only, also commit line setup (columns text,
relative time calculation, tagging, etc) is done in paintCell() the first time item became
visible. So to push to the limit the 'lazy setup' policy and gain speed.


>Another wish: The tag/head markers in gitk are really good. In qgit, I only 
>get another background color, and miss the name.

You can see the name in the status bar when you select the item. Also the refs names are 
cumulative, i.e. if you select a tag that is also a branch head and, eventually also another kind
of ref, all this information is shown in the status bar. 

Status bar has more avaiable space then inline marker this is the reason I chose that way.

>
>Yes, really nice.
>

Thanks

Marco


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

Re: [ANNOUNCE qgit-0.95]

[Marco Costalba]

Alan Chandler wrote:

>On Sunday 25 Sep 2005 07:32, Alan Chandler wrote:
>...
>
>>However when I run make I get errors of the form shown below (QSettings
>>errors) and eventually the compile stage fails.
>>
>>/usr/share/qt3/bin/uic -o src/rangeselectbase.h src/rangeselectbase.ui
>>/usr/share/qt3/bin/uic -impl rangeselectbase.h -o
>>src/uic_rangeselectbase.cc src/rangeselectbase.ui
>>/usr/share/qt3/bin/moc -o src/moc_rangeselectbase.cc src/rangeselectbase.h
>>QSettings: error creating /.qt
>>QSettings: error creating /.qt
>>QSettings: error creating /.qt
>>QSettings: error creating /.qt
>>QSettings: error creating /.qt
>>QSettings::sync: filename is null/empty
>>QSettings: error creating /.qt
>>QSettings::sync: filename is null/empty
>>QSettings: error creating /.qt
>>QSettings::sync: filename is null/empty
>>g++ -O2 -DQT_THREAD_SUPPORT -D_REENTRANT -I/usr/include/qt3
>>-I/usr/share/qt3/include -c -o src/git_startup.o src/git_startup.cpp
>>src/git_startup.cpp: In member function `void Git::parseReadFromStdout()':
>>src/git_startup.cpp:473: error: jump to label `resume'
>>src/git_startup.cpp:439: error:   from here
>>src/git_startup.cpp:467: error:   crosses initialization of `newCommits*nc'
>>scons: *** [src/git_startup.o] Error 1
>>scons: building terminated because of errors.
>>make: *** [all] Error 2
>>alan@kanger qgit-0.95 $
>
>
>Although the QSettings are still there, fixing up git_startup.cpp as per 
>Pasky's patch fixed it for me
>

QSettings are there from day one :-<
I am not able to let them disappear.....very bad. In any case should be harmless.

This morining I have updated the downloads with Pasky's patch.

Marco


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

Re: [ANNOUNCE qgit-0.95]

[Josef Weidendorfer]

On Sunday 25 September 2005 09:07, Marco Costalba wrote:
> >> src/rangeselectbase.h QSettings: error creating /.qt
> >>QSettings: error creating /.qt
> ...
> QSettings are there from day one :-<
> I am not able to let them disappear.....very bad. In any case should be
> harmless.

These errors seam to appear because scons does NOT pass through the $HOME
environment variable to subprocesses. "moc", which is producing these
errors, obviously wants to access some config options in $HOME/.qt/.

Somebody knows how to change this?

Josef

Re: [ANNOUNCE qgit-0.95]

[Robin Farine]

Josef Weidendorfer <Josef.Weidendorfer <at> gmx.de> writes:    
  
> On Sunday 25 September 2005 09:07, Marco Costalba wrote:    
> > >> src/rangeselectbase.h QSettings: error creating /.qt    
> > >>QSettings: error creating /.qt    
> > ...    
> > QSettings are there from day one :-<    
> > I am not able to let them disappear.....very bad. In any case should be    
> > harmless.    
>     
> These errors seam to appear because scons does NOT pass through the $HOME    
> environment variable to subprocesses. "moc", which is producing these    
> errors, obviously wants to access some config options in $HOME/.qt/.    
>     
> Somebody knows how to change this?    
    
With SCons, key-value pairs that are meant to appear in process    
environment need to be added to the env['ENV'] mapping. In this  
case:  
   
    env = Environment(...)  
    env['ENV']['HOME'] = os.environ['HOME']   
   
Environment() creates an SCons build environment which is unrelated  
to subprocess environments (even though the same word is used :)).  
As a special case, SCons handles the mapping assigned to env['ENV']  
as process environment for subprocesses.  
  
Hope this helps,  
  
Robin    
 
    

[ANNOUNCE qgit-0.95]

[Marco Costalba]

This is a performance update release.

The detailed list of changes is below, but the bottom line is 
qgit-0.95 is _really_ fast.

It tooks about 4 seconds, on my box, to run qgit --all on today linux tree against 
about 65 seconds with gitk. Also the used memory it seems lower.

As a 'theoretical maximum speed' a bare:

git-rev-list --header --topo-order --parents HEAD > /dev/null

Runs in about 2 seconds.

So this is the main reason I post this release. I doubt a little bit about this results 
because the spread with gitk it seems too much, so I would like a double check.

NOTE: Before to test qgit you need to UNSET flag 

           edit->settings->cache->'diff against working dir'

to avoid a slow 'git-status' (with involved 'git-update-index --refresh') call.


CHANGELOG

- more scalable pixmap memory handling.
  Graph pixmaps are created only for visible items and freed
  when log lines became not visible so that the amount of memory
  used is constant and independent from archive size. This is also a
  performance improvement because of the 'lazy setup' policy.

- changed GUI update policy to group all the updates in one big chunk at
  fixed intervals. This is faster then an incremental update.

- more scalable file names handling.
  File paths are splitted in base and file name, both are then indipendently
  indexed. This gives huge memory savings and also better scalability because the
  'different file names' set grows much more slowly then new commits. Using two index 
  tables gives good memory saving when you have lot of files in the same (deep) directory.

- Rewritten git-rev-list parsing to avoid as much as possible to move data around.


INSTALLATION

You need scons and Qt developer libs version 3.3.4 or better already installed. You need 'mt'
version of Qt libraries.

QGit is NOT compatible with Qt4.

On some platforms (Debian) you should set QTDIR before to compile.


DOWNLOAD

Download link is:
http://prdownloads.sourceforge.net/qgit/qgit-0.95.tar.bz2?download

But now there is also a git archive: 
http://digilander.libero.it/mcostalba/qgit.git

Please use 'cg-clone http://digilander.libero.it/mcostalba/qgit.git' 
I still have to set-up a 'git clone' friendly archive.

If you have problems with the sources you can download a binary:
http://digilander.libero.it/mcostalba/qgit



Marco



		
__________________________________ 
Yahoo! Mail - PC Magazine Editors' Choice 2005 
http://mail.yahoo.com

Re: [ANNOUNCE qgit-0.95]

[Petr Baudis]

Dear diary, on Sat, Sep 24, 2005 at 06:06:40PM CEST, I got a letter
where Marco Costalba <mcostalba@yahoo.it> told me that...
> This is a performance update release.
> 
> The detailed list of changes is below, but the bottom line is 
> qgit-0.95 is _really_ fast.
> 
> It tooks about 4 seconds, on my box, to run qgit --all on today linux tree against 
> about 65 seconds with gitk. Also the used memory it seems lower.
> 
> As a 'theoretical maximum speed' a bare:
> 
> git-rev-list --header --topo-order --parents HEAD > /dev/null
> 
> Runs in about 2 seconds.
> 
> So this is the main reason I post this release. I doubt a little bit about this results 
> because the spread with gitk it seems too much, so I would like a double check.

Yes, it is several times faster than gitk, good work. Had to do this in
order to be able to compile it with gcc-3.3.6, though (it complained
about goto'ing after initialization of nc - that's bogus since there's
the return, but...):

diff --git a/src/git_startup.cpp b/src/git_startup.cpp
--- a/src/git_startup.cpp
+++ b/src/git_startup.cpp
@@ -469,7 +469,9 @@ again:
 			if (!resumeTimer.isActive())
 				resumeTimer.start(1, true); // with 0 ms there is an oops in libqt-mt
  			return; // we suspend here to let GUI updating
+		}
 
+		if (false) {
 resume:
 			suspended = false;
 			processTime.start();


Besides that, this is what I don't like about qgit:

* It'd be nice to be able to choose to see all commits in the initial
dialog by single click (I know about --all). Perhaps a radio buttons
choosing between all and selected? Also, you might add the "diff against
working copy" option to that dialog as well.

* The graph column is too narrow. It should be auto-sized so that the
graph fits in, or at least there should be some clear indication that
the graph does not fit to the column at the given point. This confused
me a lot at first.

* Could you make the grey background for odd commits span to the whole
line, including the commit graph?

* The commit time is relative to now, which makes no sense to me. Also,
it is in the second column instead of the last one like in gitk, which
seems better to me. At least, the column is too narrow and then it
blends together with the commit title.

* In the filter radio buttons groups, only first few letters of the
labels are visible, and apparently no tooltips. Same problem in the
settings dialog, most of the options have the labels cut around 2/3, and
the window is non-resizable on top of that - why? I hate non-resizable
windows, especially because they are usually too small. :-)

* Single-clicking at a commit produces a significantly slower response
than in gitk, where I see the difflist and stuff nearly instantly - it
takes several hundreds of ms in qgit. That's quite annoying.

* Getting to the diff view was non-obvious for me. It'd be nice to have
some [diff] button as well somewhere. Or you could also show the diff in
the bottom part of screen in the commit view, I think gitk solved this
nicely.

* Clicking on the file was supposed to bring some annotated view, but if
it is so, it should write "Annotate" in the window title, and should
indicate that it is computing it on the background (and is really slow
in that, or I don't know, but always only single revision was shown
there). It is unclear what the "pin file" checkbox is supposed to mean,
and the whole dialog is just very confusing. :-)

* The lane information might be available as a tooltip in addition to
right-clicking on it - that wouldn't occur to me.

* Why is the Edit menu aligned to the right?

* Well - I don't like the line graphics and prefer gitk's. This is
perhaps a matter of taste, though. I find the graphics to be too tiny,
thin and not so clear as gitk's, which makes it harder to see the commit
flow. I also like diagonal lines more than angles, but that might be
just because I'm used to them.

OTOH, qgit actually flows the commits around the 0.99.6 release and such
better than gitk, keeping them all in a single lane instead of flowing
off to the right. It is still a bit hard to see what actually went on,
but curing that would require sorting the commits by date.


Nice work otherwise. :-)

-- 
				Petr "Pasky" Baudis
Stuff: http://pasky.or.cz/
VI has two modes: the one in which it beeps and the one in which
it doesn't.

Re: [ANNOUNCE qgit-0.95]

[Alan Chandler]

On Saturday 24 Sep 2005 17:06, Marco Costalba wrote:
...
> INSTALLATION
>
> You need scons and Qt developer libs version 3.3.4 or better already
> installed. You need 'mt' version of Qt libraries.
>
> QGit is NOT compatible with Qt4.
>
> On some platforms (Debian) you should set QTDIR before to compile.


I am running Debian unstable with libqt3-mt-dev installed, and qt3-dev-tools.  
I set QTDIR to /usr/share/qt3. (and running with KDE incidentally)

However when I run make I get errors of the form shown below (QSettings 
errors) and eventually the compile stage fails.

/usr/share/qt3/bin/uic -o src/rangeselectbase.h src/rangeselectbase.ui
/usr/share/qt3/bin/uic -impl rangeselectbase.h -o src/uic_rangeselectbase.cc 
src/rangeselectbase.ui
/usr/share/qt3/bin/moc -o src/moc_rangeselectbase.cc src/rangeselectbase.h
QSettings: error creating /.qt
QSettings: error creating /.qt
QSettings: error creating /.qt
QSettings: error creating /.qt
QSettings: error creating /.qt
QSettings::sync: filename is null/empty
QSettings: error creating /.qt
QSettings::sync: filename is null/empty
QSettings: error creating /.qt
QSettings::sync: filename is null/empty
g++ -O2 -DQT_THREAD_SUPPORT -D_REENTRANT -I/usr/include/qt3 
-I/usr/share/qt3/include -c -o src/git_startup.o src/git_startup.cpp
src/git_startup.cpp: In member function `void Git::parseReadFromStdout()':
src/git_startup.cpp:473: error: jump to label `resume'
src/git_startup.cpp:439: error:   from here
src/git_startup.cpp:467: error:   crosses initialization of `newCommits*nc'
scons: *** [src/git_startup.o] Error 1
scons: building terminated because of errors.
make: *** [all] Error 2
alan@kanger qgit-0.95 $ 


Is there something else I should have installed.





-- 
Alan Chandler
http://www.chandlerfamily.org.uk

Re: [ANNOUNCE qgit-0.95]

[Alan Chandler]

On Sunday 25 Sep 2005 07:32, Alan Chandler wrote:
...
> However when I run make I get errors of the form shown below (QSettings
> errors) and eventually the compile stage fails.
>
> /usr/share/qt3/bin/uic -o src/rangeselectbase.h src/rangeselectbase.ui
> /usr/share/qt3/bin/uic -impl rangeselectbase.h -o
> src/uic_rangeselectbase.cc src/rangeselectbase.ui
> /usr/share/qt3/bin/moc -o src/moc_rangeselectbase.cc src/rangeselectbase.h
> QSettings: error creating /.qt
> QSettings: error creating /.qt
> QSettings: error creating /.qt
> QSettings: error creating /.qt
> QSettings: error creating /.qt
> QSettings::sync: filename is null/empty
> QSettings: error creating /.qt
> QSettings::sync: filename is null/empty
> QSettings: error creating /.qt
> QSettings::sync: filename is null/empty
> g++ -O2 -DQT_THREAD_SUPPORT -D_REENTRANT -I/usr/include/qt3
> -I/usr/share/qt3/include -c -o src/git_startup.o src/git_startup.cpp
> src/git_startup.cpp: In member function `void Git::parseReadFromStdout()':
> src/git_startup.cpp:473: error: jump to label `resume'
> src/git_startup.cpp:439: error:   from here
> src/git_startup.cpp:467: error:   crosses initialization of `newCommits*nc'
> scons: *** [src/git_startup.o] Error 1
> scons: building terminated because of errors.
> make: *** [all] Error 2
> alan@kanger qgit-0.95 $

Although the QSettings are still there, fixing up git_startup.cpp as per 
Pasky's patch fixed it for me

Sorry for the noise.


-- 
Alan Chandler
http://www.chandlerfamily.org.uk