Proposal to conditionally include numpy data types as known numerics.
We do want to add support for Numpy but it would have to be a bit more extensive than this. At the very least the serialisation code will probably also need looking at and tests will be required.
I agree that a pull request was probably not the best way to make this proposal. Could you point me to the place where serialization is done? I can try to add the required functionality there and write some tests. Does the change I made so far look sensible? As I said in my comment to Issue #49, it might be a good idea to have a public to add new types dynamically at some point, but NUMERIC_TYPES would probably have to be mutable for that.
safe_string does pretty much everything. It's a pain because the XML libraries won't serialise numbers automatically. And it's worth noting that precision is limited in Excel.
Support for date and time types should also be added and I suspect this will be harder to do: they have to be converted to Excel's serials.
I'd handle Numpy as a soft dependency just as we do with LXML so that we have a constant to check whether it's around. This would, for example, make it possible to convert to Numpy when reading files. This makes tests easier as well.
Currently, I don't see the need for a type registry as we're not really providing a serialisation library and read support is particularly convoluted, so that there is no easy API as there is say for JSON. Apart from that: constants should never be mutable.
Are you planning to add tests to this?
Not in this form. I've been to busy to revisit the problem until recently. Since then I've learned a couple of things about numpy that would make this approach unnecessarily complicated. I will close the PR and open a proper one in a bit.