There are a few (hopefully very small) follow-ups that I forgot to point out in the original PR. But maybe they are good as follow-ups anyway:
- The docs don't mention the new dtypes yet!
- The
.imag/.real problem may need discussing? I suspect there isn't much to do except putting a big warning into the documentation, unfortunately.
- I put some random characters in and forgot to think about it again. We need something better!
- ideally the complex ones would be capitalized float ones (just for symmetry with NumPy). But, bfloat16 is
E which would be the code for complex32.
- I suppose we could actually use
c kind code for complex32, but maybe not for bcomplex32.
I'll open a PR soon, but if anyone has thoughts (especially on the character codes) please say so (no need to look what they are now, they are truly random).
There are a few (hopefully very small) follow-ups that I forgot to point out in the original PR. But maybe they are good as follow-ups anyway:
.imag/.realproblem may need discussing? I suspect there isn't much to do except putting a big warning into the documentation, unfortunately.Ewhich would be the code forcomplex32.ckind code forcomplex32, but maybe not forbcomplex32.I'll open a PR soon, but if anyone has thoughts (especially on the character codes) please say so (no need to look what they are now, they are truly random).