walk_dict_to_hdf5: Explicitly use np.int64 #71
Conversation
`dtype=float` is equivalent to `dtype=np.float64`, but `dtype=int` is equivalent to `dtype=np.int_`, which is 32 bits wide on 32-bit systems and not necessarily wide enough for the data we're trying to convert. As shown (sort of) in https://bugs.debian.org/1103147, this caused test failures on 32-bit architectures.
|
Thanks, the fix makes sense to me. 👍 I'm surprised that a Debian package of karabo-bridge exists at all, though. This is an interface to European XFEL's Karabo control system, and EuXFEL's own clusters run Red Hat. There are some scenarios where it you could use karabo-bridge from another platform, but it's easy to install through pip, and many people will want it in a virtualenv or a conda environment anyway. Can I gently request that you remove the package from Debian? Obviously if someone wants it there then they have every right to package it, but it seems like a waste of effort, and we'd rather users install it with |
|
I'll leave that sort of question to Roland, who originally packaged it - I think they usually have their reasons and I don't really want to butt in. I'm just doing various cleanups in passing! |
|
Close/reopen to re-run tests |
dtype=floatis equivalent todtype=np.float64, butdtype=intis equivalent todtype=np.int_, which is 32 bits wide on 32-bit systems and not necessarily wide enough for the data we're trying to convert. As shown (sort of) in https://bugs.debian.org/1103147, this caused test failures on 32-bit architectures.