RLS: v0.19.2

[jorisvandenbossche]

Issue to follow-up 0.19.2 release.

0.19.2 milestoned issues/PRs: https://github.com/pandas-dev/pandas/milestone/45

0.19.1 was released Nov 3. In the meantime already quite some work for 0.19.2 is done (mainly thanks to Jeff!). Main things for 0.19.2 (besides general bug fixes):

• Python 3.6 compatibility (COMPAT: 3.6 #14679).
  • This is already done I think thanks to Jeff
• fix segfault with latest dateutil (done in BUG/WIP: segfault manifesting with dateutil=2.6 w.r.t. replace when timezones are present #14631)
• possible regressions (I don't think we have any urgent reported at the moment, only combine_first throws ValueError: Cannot convert NA to integer #14687)

The Python 3.6 release schedule is a rc1 at 2016-12-05, and final release at 2016-12-16.

Regarding timeline for pandas 0.19.2, the question is maybe whether we have to wait until after the final Python 3.6 release, or if we can do it earlier (eg after rc1, with the risk there are still some changes)

@pandas-dev/pandas-core

[stonebig]

hi, would it be possible to have a 0.19.2beta BEFORE rc1, to help check there no more bugs in python-3.6.0b4 itself ?

[jreback]

as soon as the latest 3.6 is on conda forge we can build with it - if u want to ping the maintainers

[jreback]

so 3.60b4 is now active (e.g. on this PR): https://travis-ci.org/pandas-dev/pandas/jobs/179477285

[stonebig]

no errors, wonderfull !

[jreback]

@stonebig yeah I don't think any actual changes in the latest betas matters much to us. (I fixed things on 3.60b3)

[stonebig]

no waiting that Christoph Gohlke does a binary ... you may perhaps tag a beta to trigger it.

[jreback]

@stonebig we don't usually do a pre for a minor release.

[jorisvandenbossche]

@stonebig Are you able to install pandas from master to test?

[stonebig]

the idea was to get a wheel. pythonnet projet generate an artefact wheel on every test on appveyor. would be nice of pandas could too

[jreback]

@stonebig we generate a tremendous number of PR's / commits. Building the package (via wheel) will just add a lot of time to the build, then have to manage the .whl and such. We are already generating conda packages (though not actually doing anything with them).

To be honest I don't really want to do it because it inevitable causes people to use them, and then complain if they are not super backwards compatible (IOW these are usually generated by a modern compiler / arch and not the compatible layers that say conda-forge / anaconda uses).

So I think people should just build from source if they really really need something in-between relatively frequent versions. Instructions are very clear, and even on windows this i quite easy now a days.

[stonebig]

ok, will do a try

[stonebig]

falling on second line of procedure:

python setup.py build --compiler=mingw32

that cannot possibly be ok on windows for Python-3.6, unless I miss an important detail.

also if I select branch-0.19.x in github, last commit is dated 26 days ago (0.19.1)... I miss a trick on this to get current 0.19.2

[jreback]

http://pandas-docs.github.io/pandas-docs-travis/contributing.html#creating-a-windows-development-environment

@stonebig not sure where you are looking, pls following the instructions above, you must use VS 14 to compiile >= 3.5.