[isabelle-dev] An experience report on the testboard

[Lars Hupel]

To add to my last mail:

> 3. Go to https://ci.isabelle.systems/jenkins/job/testboard-afp/ and
> await new build run to be scheduled.

There can currently only be one testboard build running at a time. That
means you should wait to push until the previous run is finished.

I understand that this is not ideal. It is fixable, but not very high
priority right now.

> A disadvantage is that you have to click a lot to find out the hash id
> in the distribution view. I guess that could be easily resolved by a
> suitable configuration, or hosting the distribution testboard repository
> in the same hosting environment as the AFP testboard repository.

An easier way is this: Go to <https://ci.isabelle.systems/status/> and
on the bottom of the page, click on the appropriate testboard button. It
will take you to an overview page of Jenkins where you can see the
changes which triggered the build (including links to hgweb) and also
links to the console outputs of the subjobs.

> 5. Then repeatedly check the URI of the build until results are available.

What kind of notification would you envision here?

> An inherent weakness of the testboard approach is that whenever you
> gradually improve your changesets you to rebuild ab initio. But the
> building times are compact enough that this shouldn't be too big a problem.

The new hardware brought the raw build time down to about an hour. An
obvious optimisation would be to not clean before building. But then
again, most often, changes being tested out affect Pure or HOL which
means everything needs to be rebuilt anyway.

Cheers
Lars