numpy pinned to an old version

[nickovs]

The setup.py file pins the version of numpy by having the requirement line:

"numpy<1.21"

The current version of numpy is 1.25.2. As a result attempting to install pygraphblas un-installs the current version of numpy.

Worse, the Numpy team no longer make binary wheels available for pre-1.21 numpy versions. As a result not only will the current version be removed but the old version will need to be built from source in many situation. Building numpy from source is both slow and prone to failures because of its extensive dependencies.

It would be very helpful if pygraphblas could officially support a current version of numpy so that it can be installed reliably without having to resort to manual dependency installation.

[bionicles]

@michelp @marci543 could you guys remove this numpy<1.21 pin and debug the tests? please don't give up on pygraphblas!

just needing to speed up nx.pagerank for a clinical medical project and thought I could try this. sadly, the install with pip did not work!

I cloned the repo and removed <1.21 and pip install -U -e . did work, however "pytest tests/did not work.py.test -n 2 tests/` failed the same way, implying these tests can be parallelized.

Looks like most of the issue is with suitesparse_graphblas having evolved a bit, the numpy upgrade seems OK or at least doesn't cause the issues..

Proposal:

• add Makefile with "debug" and "debug-command" and "test"
• add nodemon.json to rerun "debug-command" when we change code
• debug the code until it passes the old tests <-- need maintainer for this

#106 Created!