[Pandas-dev] On bug-fix releases and maintenance branches

[P C]

Apologies if the formatting for this is messed up, I have just joined the
list but would like to reply to a thread that started before I joined, so I
have to copy and paste relevant items in.

Specifically, I have comments on this:


   - What are other people's experiences with upgrading pandas? And would
   more bug-fix releases actually ease the upgrading?


I work for a large company with an established Python code base (3+ years
of maintenance, tens of users). Pandas and numpy are heavily used
throughout the code base. In our experience, upgrading pandas is usually
quite an involved exercise.

When we upgraded from 0.15.2 to 0.17, it took several passes, as follows:

* attempted to upgrade from 0.15.2 to 0.16.1: Found the following bugs,
raised the following PRs

https://github.com/pydata/pandas/issues/10181 /
https://github.com/pydata/pandas/pull/10188
https://github.com/pydata/pandas/issues/10195 /
https://github.com/pydata/pandas/pull/10196
https://github.com/pydata/pandas/issues/10193 /
https://github.com/pydata/pandas/pull/10379

* The above fixes went into 0.16.2 and 0.17.0 so we decided to wait for
0.17.0.
* Tried 0.17.0: Found https://github.com/pydata/pandas/issues/11372 and
https://github.com/pydata/pandas/issues/11370
* https://github.com/pydata/pandas/issues/11372 was fixed in 0.17.1 but not
https://github.com/pydata/pandas/issues/11370 (targeted 0.18.0)
* In the end we went with an intermediate build: 0.17.1.post78+g158e85a in
the normal git versioning parlance. (
https://github.com/pydata/pandas/pull/11427 was the specific item that we
needed- to fix issue 11372 above). We chose not to wait for 0.18.0 because
that contains possibly breaking changes to resample. We judged that it was
better to go with a known non-official intermediate build of master, that
passed all our tests, than to risk another cycle of chasing bugs. We found
that from 0.15.2 to 0.17.1, each release fixed one or more issues that
affected us but also introduced other issues.

I describe this just to add a data point and relate our experience with
upgrading, rather than to shout for massive changes in the release
process.The main things that hit us were edge cases breaking when core
classes were refactored.

Thanks,

PC
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/pandas-dev/attachments/20160223/97288c0f/attachment.html>