On Tue, 2010-06-29 at 13:43 +0200, Zygmunt Bazyli Krynicki wrote:

Hello everyone.

Our current developer guidelines for python development state that we should support both python2.5 and python2.6.

I suggested that because it should be fairly easy to achieve this as long as we don't use any 2.6-only things that depend on new language features (e.g. the abc module or class decorators). I no longer think supporting 2.5 is a good idea, though, given that it adds some more constraints and python2.6 has been released nearly 2 years ago.

Also, yesterday I suggested we might want to not worry about 2.5 for now and, if necessary, backport our code to 2.5 if it's ever needed. I don't think that's a good idea either -- my gut feeling is that backporting code that uses class decorators or the abc module will be quite painful.

I'd like to ask if python2.5 can be dropped from the list. I just took the steps required to test python2.5 (and removed 2.6-only features in the process). I had to use Hardy VM to do so because python 2.5 is no longer available in the Lucid archive. Doing additional testing in a VM is cumbersome and error-prone as we can accidentally slip in a 2.6-only code or trigger 2.5 bug we didn't know about and it might go untested.

I think it's fine to drop 2.5 support.

Can we rationalise the requirements for supporting python2.5 and if so, devise a sensible testing plan for our python development?

BTW: It's worth publishing our python development practices on the Linaro wiki.

Definitely