Seth Morton

Thanks - I am working on refactoring in such a way that it should eliminate the possibility of this occurring.

Bummer. This is related to the block of code starting on https://github.com/SethMMorton/fastnumbers/blob/c283f27a1f6c38102e52e0ebace68cc74053e9a8/include/fastnumbers/parsing.h#L78 There is logic to handle that different compilers give different levels of precision. Right now, it assumes GCC/CLANG...

If someone is willing to update my `.travis-ci.yml` to also support ARM, I can try to fix this.

Currently, no - the `allow_nan` option is only for strings. The way I have found to check for nan quickly is with `x != x` - which is only true...

I would bet that `x != x` would actually be faster than calling a function because of the overhead of calling functions in python.

I'm on the fence about this, for the reason that technically `float("nan")` *is* a float. But, if I were to implement this I would do it like so: I would...

So, when implementing a function in C, it's possible to actually have an execution path for when an argument is not given (as opposed to set to `None`), by defining...

Copied from (a now-deleted) comment by @argenisleon in #39 that was intended from this issue. --- I think fastnumbers could handle 'nan' as True. From the readme ```python >>> #...

I have cleaned/moved up some conversation that belonged as part of #39.

@argenisleon The issue with the proposal to have `isfloat` always handle `"nan"` is that it removes flexibility from users. Not everybody has the same goal in mind when using these...