[Sugar-devel] CIA bot disabled

[Bernie Innocenti]

El Sat, 17-07-2010 a las 14:17 +0000, Sascha Silbe escribió:
> > To avoid polluting the history with spurious merges, people should
> > rebase their patches on top of the remote branch before pushing them. 
> That only works if the changes are local - i.e. not if you've published them on gitorious clone.

Yes. (one could maintain a separate clean tree for pushing, but things
start to get complicated)


> > I suspect that many irc floods are caused by activity on clones. Perhaps
> > we could exclude those.
> Perhaps we should for the time being. While clones contain interesting
> activity as well (IMO even more interesting than mainline),

I agree.

>  I have no idea how we could filter out the merge spam. I already
> filter out the merge commits themselves for my clone, but filtering
> the "merged" commits is beyond my git foo.

Actually, it strikes me as odd that we see them at all. Consider a
mainline tree with 3 commits:

 gitorious' mainline: a--b--c

You clone, and add your commit 'D' locally

 sascha's mainline: a--b--c--D

Meanwhhile, someone commits a few more patches to gitorious:

 gitorious' mainline: a--b--c--e--f

Then you pull:

 sascha's mainline: a--b--c--e--f--*
                           \---D--/

Now you have a merge on your tree (marked with '*').

If you push this to gitorious, what does the post-update script see?
Head changed from the old location (f) to the new one (*).

So, you should get only D... and the merge itself.

A quick test on a merge point of the olpc-os-builder repository reveals
that my theory is indeed correct:

 git rev-list 50219e37 ^f9a4be2e --pretty=oneline
 50219e37770d55b7835633917a08d3afaacd16cd Merge branch 'master' of git+ssh://dev.laptop.org/git/projects/olpc-os-builder
 41fae8d940a15d2bfc531773d5886fce77cc731a add usb_modeswitch, fixes #7172
 fb74bbaf81fe2d095fbadefaea218fdb740e93f2 sugar_activity_group: configure activity group as systemwide by default

So, maybe something is wrong with the oldrev passwd by the post-receive
hook? To make sense of it, I added $oldrepo to the debug output of
ciabot-post-receive.sh. Next time let's keep our eyes open.

I also noticed an oddity: the scrpt was invoking "git-rev-list" instead
of "git rev-list" with no hyphen. I'm puzzled on how it could have
possibly worked before.

-- 
   // Bernie Innocenti - http://codewiz.org/
 \X/  Sugar Labs       - http://sugarlabs.org/