Public SourceMage Discussion List

[seth AT swoolley.homeip.net]

On Sat, May 06, 2006 at 12:42:03PM +0200, Flavien Bridault wrote:
> Just a small question (maybe silly :-/ ), I agree that we need
> gatekeepers to grant the integration in stable-rc; but do you really
> need them to perform the integration ? Can't we trust the maintainer to
> do that ? 

No, we cannot trust them, for a few reasons.

I've seen too many bad integrations to not trust people to do p4 diff 
(or similar) to double-check the integration.

This is more true for stable than for stable-rc, but I would still like 
to see control on stable-rc.  'advisory closed' just isn't hard enough 
for me, especially when I want to know of _every_ integrate to 
stable-rc and know that the people doing it know what they are doing.  I 
explain this more in response to your next question.

> Or is this technically to much different from an integration
> from devel to test ? 

It's not much technically different, even though there are technical 
differences in layout (since we include the version).  The most 
problematic part is that integrates to stable-rc and stable often 
involve cherry-picking.  I don't really want to leave that to just any 
guru.  Our grimoire structure is complex enough now with FUNCTIONS, 
etc., that it's a non-trivial operation.  I want people who know to 
integrate the smallest or least intrusive change and can understand the 
ramifications of their action on the whole.  Early gurus don't 
necessarily get this.  I want to open up guruing in the trunk, but keep 
stable release candidates and release branches to be tightly qa'd.  If 
we let any guru access stable-rc, we have to be more careful who we let 
have access to test (the trunk).

Moreover, while stable-rc might not seem so important to have tight 
control over, the fact is that once stable-rc is branched, we start 
testing immediately on the entire thing.  That means that at that point 
anything integrated after testing has to be re-tested or limited such 
that testing can be encapsulated to the specific change (unit tested).  
Some changes cannot be well unit tested and that's why we need tight 
control over this process.  Unless we want to extend our stable-rc 
testing to be atomic against any change (making global retesting
requisite), we cannot give just anybody access.

Remember, once code enters the QA process, the less developers can 
mingle with it, the cleaner the final output will be.  This of course is 
weighed against flexibility, but with each ounce of flexibility we add, 
we need to add a pound of process to account for that.

> I could even imagine (but maybe we don't have the technology yet ;-) )
> to allow a temporary access to the spell in stable-rc once the
> gatekeeper granted the integration.

Even if we had this, for the above reason, particularly with regard to 
testing processes, I would be uncomfortable with it.

Seth