search

tuwien-cms / xprec

Full quadruple precision (double-double) data type for numpy
MIT License
12 stars 4 forks source link

Fail to build with numpy 2.0 #17

Closed shinaoka closed 4 months ago

shinaoka commented 4 months ago

@mwallerb Can we fix it quickly?

image
mwallerb commented 4 months ago

Thanks for catching this, Hiroshi!

This is really bad ... they completely broke the way dtypes are created.

See numpy user dtypes for some examples, but there are no examples on how to do this in a 1.0 and 2.0 compatible way. The migration guide is also less than helpful. I am inclined to restrict numpy compatibility to < 2.0 for now. Maybe Tensorflow will figure this out, and we can crib their solution.