Skip to content


Python 3 compatilibility work #663

merged 44 commits into from

4 participants

IPython member

The eventual aim is to make one codebase that can be installed on Python 3 using 2to3. This doesn't complete that, but it does a substantial chunk of the work, and I think it's a good time to stop and go over it.

Most of the compatibility layer is in a new IPython.utils.py3compat module.

Stuff I know isn't yet done:

  • doctests (might not be needed - 2to3 installation with distribute should translate them automatically, but will it catch IPython doctests?)
  • Script names (e.g. ipython --> ipython3)
  • What do we do about external libraries? In particular, I've made changes to pexpect and simplegeneric for Python 3. I'm happy to push changes upstream, but that could take a while, and for pexpect would probably involve dropping some backwards comatibility (the pexpect website still mentions Python 1.5!)

Full view of the manual changes I've made for Python 3:

Also, I found we had some more-or-less duplicate code for breaking down user input, in inputsplitter and prefilter. I've made them use a common version.

One test is failing at present: IPython.core.tests.test_inputsplitter.test_split_user_input, which attempts to break down the line u"Pérez Fernando". My gut feeling is that it's OK to change the test itself, because there's nothing intrinsically 'correct' about the result it gave before; but perhaps I'm missing something.

IPython member

One other problem - the coloured display of the source you get with func?? doesn't appear properly (you just see the first character). It does work with %psource func, though. I'll look into this.

IPython member

OK, it now catches errors loading the default config file, and resets exec_lines to an empty list. But when I do ipython3 profile create (or ipython3 profile create python3), it creates the full default config file in .ipython/profile_python3. Doing ipython profile create python3 copies the file across as I'd expect, so I guess it's because python3 is the default profile for ipython3.

IPython member

Hey Thomas, sorry we let this one go stale, but right now it doesn't merge anymore. Could you rebase it against master and push again? I'll be happy to review it then so we can flush it quickly.

IPython member

Done. There were just a couple of small collisions. This can now be installed and run on Python 3, although I haven't tested it extensively. With Python 3, it now installs setuptools-style entry points as ipython3 etc. I've patched simplegeneric and sent the diff upstream, but not heard anything yet.

Before merging, we should consider the failing test for splitting user input containing unicode (I think it's alright to change the test, but I'm not sure why it was written).

Other questions I haven't yet addressed, but needn't block merging this: I need to work out a way to convert doctests, which currently cause a lot of failures. Pexpect will need some patches to work in Python 3. We should also give a bit of thought to how the python3 profile should fit in.

IPython member

no, that failing test is indeed a problem. I'm not sure what's going on, but it's really a breakage. In master, we get, correctly:

In [5]: from IPython.core import inputsplitter as isp

In [6]: isp.split_user_input(u'Pérez Fernando')
Out[6]: (u'', '', u'', u'P\xe9rez Fernando')

but with these changes, we get:

In [1]: from IPython.core import inputsplitter as isp

In [2]: isp.split_user_input(u'Pérez Fernando')
Out[2]: (u'', '', u'P', u'\xe9rez Fernando')

So the input is definitely being incorrectly split, because the third output value should be an empty string and we're getting a 'P' in there.

I'm too tired to dig into this one, but we do need to fix this one before merging, not changing the test. If you make any progress let me know...

IPython member

I know it changes what happens in that case, but I'm not sure why the behaviour in master is more correct. Non-ascii characters outside a string are invalid in Python 2 anyway, so does it matter how we split up the invalid input?

IPython member

More generally, split_user_input looked like it was designed to break out the first valid Python identifier (from a comment: "fpart has to be a valid python identifier"), optionally with IPython escape characters. But it didn't do this very well, so x =1 would be split, but x=1 wouldn't. I tweaked the regex a bit so that it behaved more consistently. So when presented with u'Pérez Fernando', it breaks out P as the first valid Python identifier. Is there some better definition of how it should behave?

IPython member

You're actually right, now that I looked at it more carefully. Go ahead and update the PR so that the test passes.

I've looked it over and I don't see any problems, but since it's pretty big, let me run using this as my default branch for a day and see if I notice any problems. If all looks OK we'll go ahead and merge it.

@mirnk, @ellisonbg: if you have any objection to this one, let us know. It's a fairly large set of changes, but it will make py3 support much easier in the future, and I think Thomas did a great job of isolating the compatibility hacks to very well-defined locations. So unless I find any problems in actual use (I've reviewed it and ran the test suite), I'll merge it in a day or so.

IPython member

Thanks, Fernando. I've updated it, and the there are no more test failures on Python 2.

(Tagging @minrk because a typo missed him above. Let us know of any objections to these changes.)

IPython member

Thanks for catching the typo. If @ellisonbg and @minrk have no objections, I'll merge this one tonight (unless I find a glitch in usage today). I've been using it as my normal ipython and haven't seen any problems so far...

IPython member

I just did a PR against @takluyver's branch to localise the Session as bytes/unicode, so we don't have to call encode('ascii') all over the place.

IPython member

Awesome, thanks! And @takluyver: given that this branch does have a few conflicts with the notebook stuff, which we really need for Euroscipy (Stefan and I are planning on using it), let's delay this merge until the nb branch lands. There will be a few small conflicts to fix (just stuff in setupbase, I think) and then we can merge it afterwards.

IPython member

OK, great. I'll get Min's bsession changes merged into this branch in the meantime, and possibly investigate converting the doctests.

IPython member

Fantastic. The nb branch is a lot of work and will still take us a day or two to mege, so we may just get to actually merging this when we're in Paris, which isn't the end of the world. Between now and Tuesday I've basically run out of available time for just about anything.

IPython member

Rebased after notebook branch merged.

IPython member

Rebased again.

takluyver and others added some commits
@takluyver takluyver Add module for Python 3 compatibility layer. eb1cd97
@takluyver takluyver Start using py3compat module. ee492f8
@takluyver takluyver More Python 3 compatibility fixes. 54a1718
@takluyver takluyver Spelling correction. 7cb775e
@takluyver takluyver Repair various failures in the test suite. e36691a
@takluyver takluyver More Python 3 compatibility fixes. 18b2497
@takluyver takluyver Incidental correction to config file. f707b28
@takluyver takluyver Add Python 3 profile. c165f2e
@takluyver takluyver Select default profile according to Python version. 4408c4c
@takluyver takluyver Fix in preparation for Python 3.3 compatibility. c7f7b48
@takluyver takluyver Python 3 compatibility for identifiers. 8a38592
@takluyver takluyver Fix for looking up print function in Python 3. a7e8c70
@takluyver takluyver Further Python 3 fixes in core. cf41e14
@takluyver takluyver Add simple implementation of Python 3 style open() d60f657
@takluyver takluyver Improvements in the code that breaks up user input. 17afbbe
@takluyver takluyver More Python 3 compatibility fixes in core. ef3358b
@takluyver takluyver Reuse common code for inputsplitter and prefilter. 76b8481
@takluyver takluyver Small correction to recent change in oinspect. 74ba4eb
@takluyver takluyver Convert source to native string before doing ANSI formatting.
Closes gh-665
@takluyver takluyver Internal machinery should not be printing out error messages. 64eab80
@takluyver takluyver Make installation with Python 3 possible. b9f6e70
@takluyver takluyver Backwards compatibility hack to use nested() in Python 3.2 328adae
@takluyver takluyver Python 3 compatibility in IPython.lib.pretty 5326df5
@takluyver takluyver Small changes towards supporting Python 3. e57ac64
@takluyver takluyver Make PickleShareDB inherit from collections.MutableMapping abc 1ec897c
@takluyver takluyver Fix for nested() c14b594
@takluyver takluyver Update bundled simplegeneric to version 0.7. bcd1047
@takluyver takluyver Patch bundled simplegeneric for Python 3 compatibility. d82fb26
@takluyver takluyver Fixes for UltraTB and PyColorize with Python 3 f866f01
@takluyver takluyver Fix various problems highlighted by the test suite. 6feb755
@takluyver takluyver test_oinspect works with Python 3. b3d8f63
@takluyver takluyver Don't monkeypatch nose in Python 3, as it must be a newer version. fd074fb
@takluyver takluyver Fix config loader test for Python 3. 1dcb5a6
@takluyver takluyver Fixes to 5299658
@takluyver takluyver Various Python 3 fixes in IPython.utils 2467d2e
@takluyver takluyver Allow 'python install' to work correctly for either Python 2…
… or 3.
@takluyver takluyver Fix for 'foo?' object inspection in Python 3. 326f739
@takluyver takluyver Safer loading of default config inside Python 3 config. 6c940b7
@takluyver takluyver Fix bugs in PyColorize 974068c
@takluyver takluyver Update split_user_input unicode test. 8c63d48
@minrk minrk add Session.bsession trait for session id as bytes c340839
@takluyver takluyver Add Trove classifiers for PyPI. a88bf20
@takluyver takluyver Override input(), not raw_input(), for Qt console in Python 3. 786b191
@grahame grahame Make PNG images in the Qt console work in Python 3. 307b083
IPython member

Rebased again.

IPython member

What still needs to be done here?

If all tests pass on 2.6, it would seem that this is ready for merge, unless there are specific tasks that need doing.

IPython member

I'd like to get it merged in. There's some more things to do (doctests, pexpect), but the changes that are already here keep ending up in conflict with changes in master. I should test on 2.6, though - so far I've only been testing on 2.7.

IPython member

OK, I've just run the tests on 2.6, and all looks good. Since we've had this in the wings for so long, I'm merging it now.

Thomas, thanks a lot for the great work! I hope having this now in trunk will make your life significantly easier re. py3 support. Please do yell at us if you see us doing things that make py3 support harder in master: since we don't use it regularly, we tend to slip up easily :)

@fperez fperez merged commit 37dd091 into ipython:master
IPython member

Thanks, Fernando, that will definitely make my life easier. I'll try to keep checking it.

When you've got a moment, can you also modify the description line at to say it's superseded by work in the main repo. Leave it around for the moment, though, because the diff might still be useful.

IPython member
IPython member

There's now a big problem. We can no longer run any files that are not utf8 encoded. Python itself respects the encoding header, but since we now hardcode the encoding in, all other codings are simply ignored. Fixed in PR #778. I'll merge right away if I get an okay, as it's a very serious issue.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Commits on Sep 7, 2011
  1. @takluyver
  2. @takluyver
  3. @takluyver
  4. @takluyver

    Spelling correction.

    takluyver committed
  5. @takluyver
  6. @takluyver
  7. @takluyver
  8. @takluyver

    Add Python 3 profile.

    takluyver committed
  9. @takluyver
  10. @takluyver
  11. @takluyver
  12. @takluyver
  13. @takluyver
  14. @takluyver
  15. @takluyver
  16. @takluyver
  17. @takluyver
  18. @takluyver
  19. @takluyver
  20. @takluyver
  21. @takluyver
  22. @takluyver
  23. @takluyver
  24. @takluyver
  25. @takluyver
  26. @takluyver

    Fix for nested()

    takluyver committed
  27. @takluyver
  28. @takluyver
  29. @takluyver
  30. @takluyver
  31. @takluyver
  32. @takluyver
  33. @takluyver
  34. @takluyver

    Fixes to

    takluyver committed
  35. @takluyver
  36. @takluyver
  37. @takluyver
  38. @takluyver
  39. @takluyver

    Fix bugs in PyColorize

    takluyver committed
  40. @takluyver
  41. @minrk @takluyver
  42. @takluyver
  43. @takluyver
  44. @grahame @takluyver
Something went wrong with that request. Please try again.