Ensure that jit, vectorize and guvectorize functions can mutually call each other

[seibert]

We still have some situations where calls to one kind of function cannot be embedded in the other.

[mhlr]

In particular I am finding that @jit functions can not call @guvectorize functions

[joanneong]

Has this issue been resolved yet? I am facing the same problem as well...in particular, Numba ufuncs does not seem to be able to infer the types of other ufuncs or dufuncs...

[briantilley]

I'm having the same problem on version 0.38. @seibert from the tags, it looks like we'll have to wait quite some time for the fix in 1.0?

[stuartarchibald]

This turned out to be quite complicated. I added the stubs to do this here: #2469, direct calls with known bounds will work, but this is no different to just having a njit function that is inlined/dispatched. To get full guvectorize support loop nests will need generating or it will need to go via NumPys gufunc mechanics.

[rivasd]

Any ETA on this? I'm using guvectorize to compute a result between matrices of mismatched shapes, so I need to resort to to looping overt one of them. Being able to use a @jit function for this outer looping would be awesome but currently fails.

[seibert]

No firm ETA yet. Fixing this is a blocker for Numba 1.0, and we'd like to get that out this year.

[sklam]

#5938 is helping in this direction

[jasonml]

Are there any updates on this? I understand that it's intricate to implement, but if there's anything an amateur could do to help, let me know. Either way, are there some spots in the repo to start looking?

Thanks for all the hard work!

[duburcqa]

I'm also waiting for this feature.