Add dimension_names array metadata field

[jbms]

No description provided.

[jstriebel]

Fixes #73

[jstriebel]

Crosslinking #149 (comment)

[jstriebel]

LGTM 👍

[jstriebel]

@rabernat I just resolved a merge conflict which dismissed our old reviews, and mine doesn't count anymore. Could you re-approve this one please? Thanks :-)

[shoyer]

I don't mean to hold things up with bike-shedding, but did we reach concensus on using "" vs null for missing dimension names?

[jbms]

I don't mean to hold things up with bike-shedding, but did we reach concensus on using "" vs null for missing dimension names?

Is your intention to allow the empty string as a valid dimension name distinct from null, and not meaning an unlabeled dimension?

[shoyer]

I don't mean to hold things up with bike-shedding, but did we reach concensus on using "" vs null for missing dimension names?

Is your intention to allow the empty string as a valid dimension name distinct from null, and not meaning an unlabeled dimension?

My thinking was that null instead of a string would indicate missing dimension names, with no special meaning for empty strings.

I prefer null because it's less ambiguous from a typing perspective -- if somebody writes code that doesn't anticipate partially missing dimension names, it will fail noisily rather than silently.

[jstriebel]

@jbms As discussed last week, I changed this here to use null objects for unnamed dimensions, as requested by @shoyer. I believe this is ready to be merged now. My approval does not count anymore since I did some changes, @joshmoore could you approve this one?

[rabernat]

I got you @jstriebel! 🚀