test numpy==2.0.0rc2

[gojomo]

Test what happens to our builds if numpy==2.0.0rc2 is specified, as Numpy 2.0 release is expected on June 16 and there may be breaking changes that torpedo our current numpy >= 1.18.5 dependency.

[mpenkov]

Looks like some parts of the numpy API we were using aren't around anymore:

gensim/models/word2vec_inner.c: In function ‘__pyx_f_5numpy_PyDataType_SHAPE’:
gensim/models/word2vec_inner.c:9378:39: error: ‘PyArray_Descr’ {aka ‘struct _PyArray_Descr’} has no member named ‘subarray’

I don't see "subarray" as member in the docs, either.

https://docs.rs/numpy/0.18.0/numpy/struct.PyArrayDescr.html

[mpenkov]

@gojomo Please continue to look into this. I'm removing it from the upcoming milestone, just to get things out the door quicker.

[gojomo]

@mpenkov I'll be tinkering, but it'll be a few minutes here and there, so unsure I'll even have a good sense of the magnitude of required changes anytime soon. I suspect the numpy release will cause hiccups & scrambles across the Python data/ML ecosystem.

So, my hunch remains: next Gensim fix release should just require numpy<2.0.0, to minimize immediate problems when it arrives, pending further work.

[jc-louis]

next Gensim fix release should just require numpy<2.0.0, to minimize immediate problems when it arrives

@gojomo numpy 2.0 stable was released yesterday

[gojomo]

@gojomo numpy 2.0 stable was released yesterday

Yes, as expected & mentioned above.

[mballintijn]

What is the expected timeline for numpy 2.x support?

[gojomo]

What is the expected timeline for numpy 2.x support?

So far I've not seen much discussion or estimation of the effort required, or anyone stepping up to take the lead – so AFAICT, timeline uncertain.