Now that we have 1.1.0 out, I was thinking maybe now is the time to
merge the matplotlib-py3 branch into master. As a reminder, the main
downside is losing compatibility with Python 2.5 and earlier. We would
continue to have a 1.1.x maintenance branch for the foreseeable future
for small-yet-critical bugfixes, and can still make a Python
2.5-compatible bugfix release from that.
Any objections or concerns? Any reason to hold off?
I agree, we have to do this, and now is the time, before the work you
have already done gets too stale and hard to merge in.
My main worry is how to support the resulting py2.6-py3.2 master:
1) In the coding guide, it might be good to have notes (tips) about how
to maintain compatibility, or at least references to such notes. I have
read about py3 but have never worked with it.
2) This is going to make development significantly harder--having to
think about the compatibility requirements, and having to test
everything with 2.x and 3.x.
Yes. What is the status of the buildbots? (I've been really remiss at checking them for a very long time now). Having those would be a great help.
3) Most of the interactive backends will be unavailable, correct?
The working ones are Qt4Agg (presumably also Pyside, though untested), and Tk. WxPython seems like it's still a long way from supporting Python 3 (though I'm not terribly plugged into that community). Note, however, the wx backend will continue to work for users running Python 2.x, just as it always has. There is no need to chop this off completely. Gtk is probably ready to be used, I just haven't done the testing and twiddling for matplotlib yet.
4) I hope the 1.1.x branch doesn't have to be maintained too long; or if
it does, it would be good to have a single designated maintainer to take
care of it, backporting from master or applying custom fixes as needed.
I feel like merging in the Python 3 changes is only an improvement in this respect. Right now, I build/install/test three branches on a regular basis. After the merge, we go down to two.
On 10/28/2011 02:41 PM, Eric Firing wrote:
On 10/28/2011 07:50 AM, Michael Droettboom wrote: