Python 3 support

[QuLogic]

Most of the simple compatibility stuff is finished (i.e., #1657, etc.). So the question is how to do stuff that is incompatible between 2 and 3?

• six
• future

There are a boatload of iter* calls, either due to performance or just plain copy & paste. In that regard, it would be much nicer to just use Python 3 syntax via future. Any thoughts on the two?

Remaining items:

• Py3k installable #1701 - Python 3 installable
• Context-ified file IO #1702 - File I/O in contexts
• Portable comparisons (for sorting) #1709 - Portable comparisons
• Python3 builtin changes #1770 - Handle changes to builtins
• Python 3 exceptions/warnings #1771 - Handle changes to exceptions and warnings
• Python3 binary I/O #1773 - Handle binary I/O consistently
• Python 3 hashing #1772 - Add hashes to some types where necessary
• Python 3 __dict__ inheritance #1777 - Fix some methods due to inheritance changes
• Py3k: Use consistent sorting #1778 - Ensure tests work on sorted data
• Python3 grib fixes #1774 - Fix some GRIB issues
• Py3k consistent unicode handling #1782 - Fix Unicode handling
• Py3k compatibility fixes for pp #1781 - Fix some new stuff in PP loader
• Python 3 compatibility for C extension #1787 - Python 3 compatibility for C extension
• Enable Python 3 #1842 - Final tweaks and Travis enablement

[QuLogic]

Ping?

[rhattersley]

We've used six in Cartopy, so sticking with that would avoid the need for yet another dependency.

[QuLogic]

Ah, it seems it's a good thing I started trying to do it that way already, then.

[QuLogic]

Currently reaching about 90% test passes now, but most of the failures are now due to GRIB's issues, or with something I don't yet understand.

[pelson]

Currently reaching about 90% test passes now, but most of the failures are now due to Grib's issues, or with something I don't yet understand.

It isn't unreasonable to state that GRIB is not supported on Python3. We do precisely that for Windows. GRIB is an optional dependency for Iris after all.

[ajdawson]

It isn't unreasonable to state that GRIB is not supported on Python3. We do precisely that for Windows. GRIB is an optional dependency for Iris after all.

Agreed, with the following caveat: When we announce Python 3 support we must make sure to declare very openly which components are not supported.

[QuLogic]

Hrm, you're right that's how it's documented, but the tests seem to disagree.

[QuLogic]

What is the encoding for the text fields in the NAME, NIMROD formats?

[QuLogic]

----------------------------------------------------------------------
Ran 3432 tests in 86.103s

FAILED (SKIP=16, errors=4, failures=6)

99% there...

[rhattersley]

99% there...

👏 Impressive. 😄

[QuLogic]

I've got about 8 branches split out into common changes and ready to make PRs. But I'm not sure how co-dependent they are, so I'm not going to open any PRs until the other 4 are merged, or I'll have to rebase them repeatedly.

[rhattersley]

If there are still multiple steps to achieve this then perhaps a checklist in the description, or split it out into multiple issues.

[rhattersley]

Thanks for adding the checklist @QuLogic 😄

[QuLogic]

Hmm, looks like there are no 3.3 dependencies in conda. And a few missing things in 3.4, too.

[pelson]

And a few missing things in 3.4, too.

https://github.com/SciTools/conda-recipes-scitools/blob/master/pyugrid/meta.yaml#L14

Hmm, looks like there are no 3.3 dependencies in conda.

I decided to keep the matrix small. It is easy enough to fix, but has the cost of an extra load of built distributions. Given we want to keep the channel as small as possible, do you feel that py33 is critical @QuLogic?

[QuLogic]

Well, that's up to you. Cartopy, biggus, etc. are tested on 3.3.

[QuLogic]

It turned out to be pretty easy to split apart the smaller branches, so there are 5 new PRs to review.

[QuLogic]

The first build to pass on Travis without trouble is finally up (except 3.3 has no conda deps.)

[pelson]

The first build to pass on Travis without trouble is finally up (except 3.3 has no conda deps.)

Is there a good reason to support py33 (other than the fact that we can)? Otherwise, I'm inclined to keep our lives as simple as possible and state py27 / py34 / (py35) support only.

[QuLogic]

It used to be the latest supported on RHEL, but now software collections has a 3.4 backport, so perhaps it is not so necessary.

[pp-mo]

As @pelson already suggested, we probably only need support Python 2.7 and 3.4 : comment
Against 3.3 : few people would specifically need it
Against 3.5 : available conda builds are too restrictive, especially regarding our current need for matplotlib=1.3.1. It will make much more sense to progress this when we have fixed our testing dependency on an older matplotlib version (or at least updated it), but that has to be future work, not a pre-requisite for 1.9.

So just 2.7 + 3.4 sounds like a a nice simple goal
(= stated support; tests work; testing in Travis)

@QuLogic @pelson @rhattersley are we all agreed that will be good enough ?

[QuLogic]

Sounds good to me. I think most of the hard work for updating matplotlib has to do with cartopy, but that's already in progress.

[rhattersley]

I'd be very happy with 2.7 + 3.4. 👍

[rhattersley]

So... this is done?!? 😁

[pp-mo]

So... this is done?!? 😁

I believe it is : everything here has now been included somewhere else.
Agreed @QuLogic ?

[QuLogic]

Pretty much; I was holding off unless any last minute issues came up.

[pelson]

🎉 🎉 - woohoo! Welcome to the 21st century 😜 ! 👍

[rhattersley]

Pretty much; I was holding off unless any last minute issues came up.

Superb! Well done everyone... and especially @QuLogic! 👏

I'd love to close this PR now, but I'd rather @QuLogic did it. (If anything else comes up I think we should just raise a specific issue.)

[QuLogic]

🎂 🎉

[rhattersley]

🤘

[ocefpaf]

Thanks all! And special thanks to Elliott!

I am looking forward for the next release!

Iris was the last major dependency I needed to move to py3.

On Nov 20, 2015 7:46 AM, "Richard Hattersley" notifications@github.com
wrote:

—
Reply to this email directly or view it on GitHub.