New build system problems
david.holmes at oracle.com
Tue Mar 5 23:30:34 UTC 2013
On 6/03/2013 9:17 AM, Martin Buchholz wrote:
> On Tue, Mar 5, 2013 at 2:36 PM, David Holmes <david.holmes at oracle.com
> <mailto:david.holmes at oracle.com>> wrote:
> Sorry but that is completely unacceptable. If you are providing
> changes that obviously impact multiple platforms (ie there are
> platform specific changes) then they _must_ be tested on all
> platforms. If the external author/committer can not do that then
> they must work with someone in Oracle who can assist with JPRT runs etc.
> There's always a tradeoff between agility and not breaking other folks.
And in this kind of case the right tradeoff is to not break things. You
simply can't push a <OS> change and not have tested a <OS> build!
> IMO the right approach is to improve processes so that bad commits don't
> cause other developers to lose time. Once upon a time, I was actually
> the tl gatekeeper and I implemented such a system. Today, I see there's
> a tl-gate, but there's close to zero testing between submission to
> tl-gate and "promotion" to tl-proper. In the system I implemented,
> there was a full build/test cycle in between.
The processes should be there to catch mistakes, not to encourage lack
of upfront testing.
If the "gate" provided such functionality it would be like submitting
each change via JPRT. While a nice idea it is completely impractical
given the resources it would require.
More information about the build-dev