ENH: Add utf-8 support for savetxt

[DavidToca]

Add support for utf-8 strings for savetxt

related issues

• savetxt cannot handle unicode objects #4939
• in python 3 numpy.savetxt can not save matrices with unicode data #4543

[pv]

These should not be touched. Any changes should be restricted to npyio.py.

[pv]

I don't think hardcoding an encoding like this is the correct way forward.

There are a number of things to address in savetxt/loadtxt, and the strategy should be explicitly discussed to be sure to get it right:

• The savetxt handling of byte string formatting is buggy on Py3 --- this traces back to % formatting being not supported for bytes on Python < 3.5.
• Writing to text streams vs. byte streams.
• Selecting the encoding to use for cases where conversions are necessary.

You also need to add corresponding tests to the test suite for a PR to be accepted.

[DavidToca]

@pv thanks for your review, so, should I add this as an issue, or post it in on the dev mailing list in order to be discussed?

[pv]

What I mean by "explicitly discussed" is that I think it would be useful to first write down an overview/spec of how you think it should work, for example in the pull request description above. For example, what should happen with dtype='S and type='U' strings etc., how to select an encoding, whether there should be a possibility to write/read text streams instead of just byte streams --- once there's such an overview/spec of the intended behavior, it's easier to check whether the code does what was intended, and whether what is intended is consistent.

[juliantaylor]

savetxt has an encoding option in 1.14 so this PR can be closed