Open Stack

On Tue, 27 Jan 2015, Robert Collins wrote:

> Yes, but the experience we have about the limitations of per-service
> venvs is still relevant, no? Are you really saying 'do per-service
> venvs because they work well', or are you agreeing with me that they
> don't solve the problems plaguing the gate?

To take this topic even further from its original subject I'd just
like to remind at least myself that the reason we have problems in the
gate is because there are real problems that actually exist or will
exist soon and that catching them before they leak out is is presumably
exactly why we have a gate.

Given a lot more resources we should always pip -U and never put upper
version caps in requirements.txt. Then when stuff breaks do the good
work to untangle the mess and fix the problem. Not just for us but for
the commonweal.

Breaking stuff tends to be how stuff gets fixed.

Yes, I'm talking idealistic impractical hooey here, but I think it is
important to remember these things. So whereas earlier I said I was
keen on virtualenvs, now that I actually think about it some I think
we're better off creating a chaotic crucible of everything in one fiery
cauldron of truth and using breakage in the gate as a way of marshalling
resources.

Do I expect to see that happen? Sadly, not really.

-- 
Chris Dent tw:@anticdent freenode:cdent
https://tank.peermore.com/tanks/cdent