Repository navigation
MNT: Drop Numpy 1.16 - #10664
MNT: Drop Numpy 1.16#10664
Conversation
|
Oooo... Failures with Numpy 1.19! UPDATE: Looks like just |
|
@astrofrog or @Cadair , how do I upgrade Scipy version in the failed |
|
Possibly of interest: SciPy is currently having a discussion about dropping Python 3.6 and NumPy 1.14 / 1.15 support – so slightly offset to this plan, but some arguments e.g. about available versions in the various Linux distros might apply. |
|
I think a good point being made in that thread is that when a user updates astropy, updating numpy at the same time is essentially trivial (if not already automatic), but for python itself one is much more dependent on OS, etc. |
|
Dropping Python is more controversial indeed. Hoping to discuss at astropy/astropy-APEs#62 |
mhvk
left a comment
mhvk
left a comment
There was a problem hiding this comment.
Mostly looks good - nice to see it does really simplify things! But some comments...
|
|
||
| out[out_name] = col_cls.info.new_like(cols, n_out, metadata_conflicts, out_name) | ||
|
|
||
| if not NUMPY_LT_1_17 or issubclass(col_cls, (Column, Time)): |
There was a problem hiding this comment.
not NUMPY_LT_1_17 will now always be true, so I think we can de-indent what follows and remove the else clause! (I always wondered why we didn't use NUMPY_GE_<something>!)
| raise NotImplementedError('__array_wrap__ should not be used ' | ||
| 'with a context any more, since we require ' | ||
| 'numpy >=1.16. Please raise an issue on ' | ||
| 'numpy >=1.17. Please raise an issue on ' |
There was a problem hiding this comment.
Let's remove the version here - maybe "since all use should go through array_function. Please ..."
| table = cls._substitute_iers_b(table) | ||
|
|
||
| q_where = cls._quantity_where if NUMPY_LT_1_17 else np.where | ||
| q_where = np.where |
There was a problem hiding this comment.
Ideally, we'd just replace q_where with np.where... (sorry to have set up this extra work for you!)
This comment has been minimized.
This comment has been minimized.
|
I hacked around the CircleCI failures. The one remote data failure is unrelated and hopefully will go away when numpy nightly is refreshed. |
mhvk
left a comment
mhvk
left a comment
There was a problem hiding this comment.
Looks all good to me! Thanks!
|
Before I merge, I think there was consensus that dropping 1.16 was OK, right? |
|
I haven't heard much feedback. No feedback is good feedback? 🤷 |
|
I'm 👍 on this! |
|
OK, let's get it in! Thanks, @pllim! |
The oldest version of NumPy that Astropy supports is 1.17 (see pull request astropy#10664), but the documentation still contains several remarks about older NumPy versions. This commit removes such obsolete references.
4 participants
Description
This pull request is to bump minimum supported version of Numpy from 1.16 to 1.17. This also adds CI for Numpy 1.19.
This is a prerequisite for #10644 .
cc @astropy/coordinators