[rfc] time to remove LANG=C run from make check

[Vincent Ladeuil]

>>>>> "martin" == Martin Pool <mbp at canonical.com> writes:

<snip/>

    martin> If you mean the load on your cpus, then use cloud
    martin> resources and (with common sense) expense them.

Hehe, I see a pattern emerging that I should really go the cloud
way.

Yet, in this case, it's not that easy, slaves connect to the
master which in turn send them builds.

Since common sense dictates to not let those slaves run 24/7
(like mine do[1]), this requires: starting them, wait for the master
to contact them and run the necessary builds, stop them. Tricky
at best (doable of course, but not available out of the box).

Hopefully Hudson may be used the other way around and the problem
will be moot when we switch.

Plus the load problem is on leopard, because it's my laptop and
is under-powered and I'd be surprised if we can run OSX images on
EC2 ;)

The linux slaves, as said elsewhere, even sharing the same
physical host, represent only ~30 minutes by run.

leopard, alone, requires 2 hours, but is always asleep (or
sometimes even not connected to the network) when the builds
starts, which means I wake it up in the morning and it run
against a different revision.

Nothing problematic, just a bunch of small details that could be
avoided by using another host for that (and I've exhausted my
macs[2] here :-)

I still want to have local slaves (configured even if not used in
the botnet) if only because it means I'm able to reproduce and
diagnose the failures locally. 

It has been proved repeatedly that the bugs we can't reproduce
ourselves are an order of magnitude harder to fix.

       Vincent

[1]: There was a reason I put 12GB of ram in that server...
[2]: The G3 and G4 ones don't count :)