eich at suse.de
Wed Apr 19 00:17:03 PDT 2006
Adam Jackson writes:
> On Tuesday 18 April 2006 13:12, Egbert Eich wrote:
> > Adam Jackson writes:
> > > If this is ever the case (and it is), you're not releasing often enough.
> > > Wasn't "release early and often" the _mantra_ for modularization?
> > Well, before you do a release you still need some people to test your stuff
> > - don't you? The more there are the more secure you are that what you
> > release is really OK.
> > Looks like you are advocating snapshot releases. Release and forget as soon
> > as the next one comes out.
> I want to defuse this particular point. Remember that our version numbers are
> multivariant. Major, minor, bugfix, buildfix, if you want to split it that
> far down. It may well be that the first usable 1.5.x version of the i810
> driver is 1.5.4. That's just the way it goes. It doesn't mean you should
> hold off from releasing 1.5.0 until you're sure it's perfect.
I don't think we are in contradition here.
> You _will_ ship with bugs. We always have and we always will. Accept that
> reality and design your processes to reflect it. For most modules, this
> means release for almost any user-visible change and okay yeah your version
> number will get large in the last digit, so what.
People wdo appreciate if versions they pull have received sufficient
soak time before they are declared to be a release. Nobody will expect
anything to be entirely free of bugs.
Furthermore if you realease a module you must expect that people pull
it and report bugs on it. This may happen half a year or longer down
The more often you release the harder it will be to keep track which
problem has been fixed when - or tell them not to bother to report bugs
unless they have pulled the latest release.
I know that the argument held against this is that people should get
bits from their 'Linux vendor' - which is expected to have given them
soak time and picks up the QA work.
There are two points to this:
1. It supports the argument that vendors have a strong interest
to cherry pick fixes from SCM.
2. The world lis not commercial + Linux vendors only.
Leaving the QA process almost entirely to downstream may cause
problems to other communities. Maybe I'm wrong here. Somebody
from the BSDs should probably comment here.
> I am absolutely not advocating snapshot-and-forget in the general case.
> Now for the server, which has a pretty high rate of churn, I think regular
> snapshots are valuable, in that they give a quick and relatively fine-grained
> way of stating when bugs or features were introduced. The level of
> integration testing required for a releasable server is generally higher than
> that for the leaves of the module tree, so longer release cycles with
> mostly-usable snapshots is probably a better choice than me stamping out
> 1.1.0 today and waiting for the bug reports to come flying back in.
> Point is: pick a process and go with it, as long as it gets code released in a
> fashion that's timely, or predictable, or both.
Our QA 'guidelines' seem to be very much in the flux as pretty much
everything else. At least I don't seem to completely understand them.
Depending on who you ask you get different opinions (which may also
change over time).
More information about the xorg