Bugfix releases (RE: profiler results for __getattr__, am I reading this correctly ? )

[Andrew Dalke]

Richard Jones:
>Actually, my biggest reason for not going to 2.2 is that it introduces a
>bunch of new stepwise changes to the code. New code == new bugs as far as
I'm
>concerned. This is not the way to go if your prime intention is to aim for
>less bugs.

It's a worthy concern.

On the other hand, we've been upgrading from Python 1.5.2 to
Python 2.1.  We only came across a handful of problems, and our
regression tests caught them all.

They were things like

   data.append(x, y)   instead of    data.append( (x, y) )

Even ugly code like

   def f(x):
       from somewhere import *

will work in 2.2.  Why?  Because when the static scoping changes
went in, enough people (like me :) complained about backwards
compatibility issues so there's support for this.

Hmm, I should try PyChecker to see what we missed.

Actually, our biggest problem hasn't been Python.  It's been
third party libraries, which had slow API changes over the
time we weren't in sync with the main Python world.  The next
biggest problem has been convincing people to upgrade to the
newer system, even though there have been zilch problems so
far on the test systems.

                    Andrew
                    dalke at dalkescientific.com