[darcs-users] buildbot changes

[Petr Rockai]

Hi,

you might have noticed that I have done a couple of changes in the
buildbot. First thing, the look has changed: we have upgraded to
buildbot 0.8.1, which has a new default style. Second, since that
version of buildbot supports templating the HTML pages, we have gained a
"quick links" bar at the top of each page for better navigation.

However, more importantly (and interestingly): I have split the
waterfall into three pieces:

- http://buildbot.darcs.net/waterfall?category=head
- http://buildbot.darcs.net/waterfall?category=release
- http://buildbot.darcs.net/waterfall?category=try

(links to all these are available near the top of each page, and on the
buildbot frontpage -- http://buildbot.darcs.net/)

Another new feature is the "console" view
(http://buildbot.darcs.net/console) -- unfortunately, this is still
somewhat experimental and the two branches we have: head and release,
get somewhat interspersed, so patches pushed to one will show up on the
console and they will get coloured from the following build of that
branch (which is obviously a buildbot bug). This will be less of a
problem when the release branch is inactive, which should be most of the
time. Nevertheless, it does a better job than the waterfall in
visualising which patches are in what state (on which builder).

Yet another new feature we now have is that pushes to the release branch
now automatically trigger a build on the RELEASE builders, so no manual
action is needed anymore. The Release Manager will hopefully find this
useful.

The TRY Service
---------------

Last but not least, a new feature is available to the members of the
review team (join today!) -- we can submit our work-in-progress (or
review-in-progress, more likely) code to buildbot and get it
crunched. This is what the TRY builders and waterfall are
for. Unfortunately, this comes with a caveat: to use this feature, you
will need to install buildbot on your machine, and likely that needs to
be at least version 0.8.1 (although you may try with older: if it fails,
you probably need to upgrade).

To actually submit a build, you should just run
./contrib/buildbot-try.sh: this script will create a repository that is
an intersection of your current one and darcs.net, apply all your local
patches and working copy changes on top and submit the overall diff
against darcs.net to the buildbot, which will then check out the
intersected repository, apply the generated diff and do a normal build
(which includes benchmarks on one of the slaves). It took some
persuasion, but the windows slaves are now participating in this TRY
service as well. You can pass --wait to the script to have it wait until
the build(s) finish, and --builder="..." (e.g. "6.10.3 Vista TRY") to
limit the build to just some of the builders. See "buildbot try --help"
for details.

Since the service can be used to execute arbitrary user-submitted code,
the access is restricted to the review team: you need to be able to log
into darcs-users at darcs.net (the buildbot-try.sh script will use ssh to
fetch a password that is then sent to the buildmaster).

When you submit a build, the update box in the TRY waterfall will have a
"patch" link in it, which you can inspect to see what changes are being
built/tested.

I think that is it. I will also update the wiki with the above info
about the TRY service, when I get around to that...

Yours,
   Petr.

PS: We are now also collecting fairly robust data from benchmarking, so
in near future, we should be able to visualise the evolution of HEAD
darcs, with a dataset for each push. Also, I have darcs-benchmark almost
integrated into the windows buildslave, so that should provide us with
further data when done.