BLD: Build against numpy >= 2.0.0rc1 #1233
Conversation
mferrera
force-pushed
the
unpin-numpy-2
branch
from
c246439
to
25a09c4
Compare
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #1233 +/- ##
==========================================
+ Coverage 80.02% 80.95% +0.92%
==========================================
Files 98 92 -6
Lines 13680 12175 -1505
Branches 2203 1981 -222
==========================================
- Hits 10948 9856 -1092
+ Misses 1999 1664 -335
+ Partials 733 655 -78 ☔ View full report in Codecov by Sentry. |
This unpins numpy<2 and builds the SWIG C library against numpy >= 2.0.0rc1. Versions of numpy major 2 are backward compatible with numpy 1. Hence we should not see issues in environment where numpy<2 may be maintained for years to come, i.e. RMS environments. The numpy SWIG bindings committed in this repository are not out-of-date in a way that would break compatibility. The last update to them in numpy's repository was two years ago from the time of this commit.
mferrera
force-pushed
the
unpin-numpy-2
branch
3 times, most recently
from
291d1a9
to
164b00d
Compare
mferrera
force-pushed
the
unpin-numpy-2
branch
from
164b00d
to
251f83d
Compare
| if numpy_version >= (2, 0, 0): | ||
| assert x.dtype == np.int64 | ||
| else: | ||
| assert x.dtype == np.int32 |
There was a problem hiding this comment.
We should probably take another look into this and if there are implications for file sizes
There was a problem hiding this comment.
Does this indicate some problems with backward compatibility?
There was a problem hiding this comment.
Need to test this more
| "numpy==1.19.2; python_version == '3.8'", | ||
| "numpy==1.19.5; python_version == '3.9'", | ||
| "numpy==1.21.6; python_version == '3.10'", | ||
| "numpy==1.23.5; python_version == '3.11'", | ||
| "numpy==1.26.2; python_version == '3.12'", |
There was a problem hiding this comment.
I do not think this should be replaced without proper testing. The reason is that numpy version for RMS is pinned, and to avoid ABI version issues we should compile on the oldest version of numpy that is supported by a given python version.
There was a problem hiding this comment.
Agree that we will need to do some thorough testing. A few good indicators so far:
- numpy claims that ABI is backward compatible from 2 to 1.x.x
- All of the tests are passing in the RMS 14.2 workflow in this PR which runs against
numpy==1.24.3
Resolves #1209
This unpins numpy<2 and builds the SWIG C library against numpy >= 2.0.0rc1. Versions of numpy major 2 are backward compatible with numpy 1 with respect to the C api. Hence we should not see issues in environments where numpy<2 may be maintained for years to come, i.e. RMS environments.
The numpy SWIG bindings committed in this repository are not out-of-date in a way that would break compatibility. The last update to them in numpy's repository was two years ago from the time of this commit.
As a consequence of supporting numpy>=2 this PR deprecates support for Python 3.8, which hits EOL on October 2024.