[HipHop] WMF/HHVM status/check-in

[Tim Starling]

On 17/07/13 07:18, Sean Cannella wrote:
> Hi -
>
> I'm a relatively new engineer on the HipHop VM team and just joined
> our Open Source effort so I haven't met any of you yet. Hi!
>
> Since we haven't had any traffic on this list in the past month or
> so, I just wanted to check in; is WikiMedia waiting on something
> from Facebook / plugging away / other?
> We want to make sure silence doesn't mean that both sides of this
> effort think they are blocked on the other.
> Let us know if you need any support from us to proceed or, if not,
> generally how things are going with the port.

Hi Sean,

We (WMF Platform Engineering) are a pretty small team, with a lot of
other responsibilities. Nobody is assigned to work primarily on HipHop
at the moment, but we're planning a concerted project aimed at
production deployment starting in September or October.

In the meantime, I hope there will be some work from WMF volunteers,
WMF staff with spare time, and FB staff. The kinds of things I'd like
to see are:

* Updates to hphp/doc/*, for example, documentation of what hhvm is
and what its command-line parameters are.
* Successful execution of the MediaWiki unit test suite, if that's not
done already.
* Comparative benchmarks, such as parse times of large articles, and
complete page view times for parser cache hits.

I'd also be interested in having a discussion about strategies for
bytecode repo management. Should repos be pre-built before deployment,
or should the application servers just be allowed to populate the
repos while serving normal traffic? And what is the central/local
split meant to be for? Presumably SQLite's locking strategy wouldn't
work across a shared filesystem like NFS -- maybe you have a different
definition of "central"?

Should Repo.Authoritative be used?

-- Tim Starling