Python 2.7 for Maverick

[Scott Kitterman]

"Barry Warsaw" <barry at canonical.com> wrote:

>On Jun 15, 2010, at 11:46 PM, Scott Kitterman wrote:
>
>>I went through making a new Python version the default in the same release
>>cycle in which it was introduced (with 2.6).  It might be doable for main,
>>but not for Universe.
>
>Since you've been through it, and I haven't, I'd like to learn from your
>experience.  Why is it not doable for universe?  Is it simply the sheer number
>of uploads that have to happen?  The vast majority of those will be mindless
>version bumping right?  Can't that be automated?  Or is there more to it that
>I'm missing?

The needed rebuilds can be scripted.  That's not the primary concern.  It will be first trouble shooting the build failures (some of which will be directly Python related and some won't). Then it will be trying to find and fix issues with packages not working correctly with 2.7.

>We're only talking about builds, right?  We don't actually know how many of
>those packages are not compatible with Python 2.7.  This is I think the bigger
>risk, however, in my experience as Python's release manager, we will never
>really know until we make Python 2.7 the default.  People just don't test
>their stuff until forced to.

To an extent that's true,  but I'd much rather make it the default during tool chain setup for Maverick +1 so we get the full cycle to work at it and more of the needed rebuilding will get done as part of the natural churn of the archive. 

I'd also like time to get this into Debian Experimental and see what benefit accrues from it.

I'll point to the 200+ bugs just now filed in Debian about string exceptions being removed as an example of how latent issues can be left for a long time when Ubuntu gets too far ahead of Debian. 


>I can understand the argument for letting upstream feel the pain first.  OTOH,
>there are only a couple of changes that I can see that might cause some
>sporadic incompatibilities (not global ones).  I think we'll get a good sense
>of those as we try to run our desktop on 2.7.

People can do this for testing without imposing it on end users who just want their computers to work. 

>>If this were the cycle before the LTS, I'd be all for pushing hard to reduce
>>LTS risk, but it's not.  There's no reason to kill ourselves now to make it
>>default in 10.10.
>>
>>Python 2.7 can be available and supported for Python people that want the new
>>features. For 10.10, I think that's enough.
>
>It's certainly not an unreasonable argument.  OTOH, we're calling this release
>Maverick for a reason. :) But I really want to better understand why you think
>it can't be done.  If it can't, it can't, but I do think we should continue to
>push for it.  Before beta 1, it'll be easy to fall back to 2.6 if necessary,
>right?
>

Falling back is doable (I think), but not easy. Moving the default Python to a lower version is not one a use case considered in the current design. 

Scott K