3.6 Release Experience, Parrot Developer Summit, Roadmap Goals

[James E Keenan]

Fellow parroteers,

3.6 was my first experience serving as Parrot Release Manager.  It was 
an arduous experience.  Unfortunately, it did not qualify as boring. 
Kudos to cotto++, NotFound++ and Coke++ for interventions at key points. 
  The release is not complete yet, as I don't yet have the permissions 
necessary to upload the HTML versions of our documentation.

Two issues had an adverse impact on 3.6's degree of difficulty.  Both 
are issues of long-standing.  Both reflect cultural factors within our 
community.  Both are likely to re-emerge in the future.  Hence, not 
surprisingly, I would like to see both discussed at the Parrot Developer 
Summit to take place on July 30 or 31, and both should be considered for 
elevation to Roadmap Goal status.

I. Smoke testing

smolder.parrot.org was down from, IIRC, the first week of May until late 
June.  We survived the 3.4 and 3.5 releases without Smolder mostly by 
(a) taptinder++; (b) ignoring the possibility that we were getting 
failure reports on OSes other than Linux (see below).

When Smolder came back on line, there were several things that limited 
its usefulness.

* It took time for the human smoke testers to learn that it was back on 
line and resume sending smoke reports.

* One of the automated smokers, a NetBSD box at the compile farm, was 
stuck on a Parrot release from February and gave the illusion that we 
were still passing on that OS.  (We were, but only under certain build 
conditions.  And we don't have anyone doing actual development on 
NetBSD, so diagnosing those failures is difficult.)

* We weren't/aren't getting any smoke reports on OpenBSD.

* Most critically, we were/are getting a different set of test failure 
reports from each Windows box submitting a smoke.

This meant that we were helpless when one of our veteran developers 
presented us with a build failure on OpenBSD just before the release. 
And while we did manage to make code revisions in order to get PASSes on 
some of our Windows boxes submitting smoke reports, other Windows boxes 
are still reporting failures.  Hence, the lack of consistent smoke 
testing throughout the quarterly release period limited the degree of 
quality assurance we were able to provide in the week leading up to the 
3.6 release.

II. Operating systems other than Linux

Let's face it.  Our project's members like to develop *on* and *for* 
Linux.  We react to working on or (especially) for OSes other than Linux 
as either being told to floss (Darwin, the BSDs) or having teeth pulled 
(Windows).

* People *state* that it would be good to have more Parrot developers on 
Windows, but they really would like those developers to be *somebody else*.

* Given the number of us who use Macs, you would expect many more 
Darwin/i386 smoke reports than we actually get.  But long-standing 
corner-case bugs on Darwin go undiagnosed.

Our mantra: "Parrot is a virtual machine aimed at all dynamic 
languages."  The reality: "Parrot is a virtual machine aimed at all 
dynamic languages, provided you're on Linux."

So there's dissonance here between what we know, at an intellectual 
level, is good for the Parrot project and what we as individual 
participants in the project find fun to do.

I don't have any solutions to these problems at this point -- which is 
all the more reason I think they need to be on the agenda for the Summit 
in 11 or 12 days.

Thank you very much.
kid51