[QGIS-Developer] Realistically, we can't release in 2 days 😱

[Greg Troxel]

[dropping psc because it doesn't accept mail]

Nyall Dawson via QGIS-Developer <qgis-developer at lists.osgeo.org> writes:

> Everyone who has been actively testing 4.0 nightlies has probably seen this
> coming, but we're just NOT ready to publicly release 4.0 on the 20th. Aside
> from anything else, we still don't have a fix for the profile corruption
> bug* which should be a blocker alone. With this one a user's 3.x profile
> can get completely corrupted, and/or cause hangs on loading 3.x/4.0.

In general, I think firmly fixed release dates are not a reasonable
plan, except for micros along a stable branch which is kept stable at
all times.  What's reasonable is a firm feature freeze date.  Basically
one can stop destabilizing changes on command, but one can't make
stability on command.  (Yes, I have managed software development to
deadlines, and yes, I know it's really not fun at times.)

As a packager, the idea of a 4.0 (with massive change in dependencie)
coming out without a 4.0a1 or 4.0b1 or even an rc is uncomfortable.

I don't see how is served by tagging a 4.0 release before it's ready for
use by random users.  To me that's the key question.

So I'd say:

  It's past time to be in feature freeze on master.  Definitely bug
  fixes and doc fixes only.  (Probably that's true and has been for a
  while.)

  (The last few years we have been able to get away without a/b/rc
  because there has been little change in deps and it's mostly core code
  that changes in a way that doesn't tend to lead to trouble in
  environments other than the developer's.)

  If it mostly works, which it sounds like it does, there should be a
  4.0a1 tag laid down, and that can lead to 4.0a1 binary packages.
  (People who run an alpha should be making backups, even more than
  everybody else.)

  4.0.0 should be tagged when it's ready, not before.


I realize there is sort of ".0 releases are sort of rc", but I see that
as a higher-level risk mitigation rather than truly being an rc.