Jack Pappas

Kevin, did you look at the code in the PR I submitted? The cache is not dynamically-sized, it's just a fixed-length char[] stored in a thread-static field. The cached array...

JB, if you're still planning to implement this, perhaps you could adapt the code from ResourceLib (http://resourcelib.codeplex.com/). It's an MIT-licensed project for manipulating Win32 resources in an assembly.

@aregm posted a nice writeup / summary on threading APIs in another thread, and it feels very relevant to this discussion of parallelism as well: [Omni Parallel Runtime_New.pdf](https://github.com/pydata-apis/workgroup/files/4784720/Omni.Parallel.Runtime_New.pdf)

@aregm Thank you for sharing that writeup, it contains some great insights into the various trade-offs in these threading layers; I thought the comparison of compute vs. I/O handling is...

The latter option (to do an in-place true division) doesn't seem feasible, at least when the dividend (the l.h.s. of the ``/=`` operator) has an integer dtype; to do that...

Thanks! Looks like this is to fix #105, is that right? Would you mind adding a quick unit test for this case (of the empty input list), both to demonstrate...

@3keepmovingforward3 Sorry about taking so long to respond to your question, I was caught up with some other projects this past week. The test at test_categorical.py line 1606-1611 is for...

You can create these, you just need to specify ``unicode=True`` when calling the FastArray constructor. By default, it checks whether all the strings can be safely converted to ASCII and...

Thank you for reporting this. Your suggestion for a fix seems reasonable, but I think with a small bit of extra work it could be made more general and powerful....

If you add this, please use the proper tz database name for the timezone: ``Australia/Sydney`` for easier compatibility with other libraries/tools. Ideally, riptable should get rid of the hard-coded timezone...