Raise exception on non-unique column index in to_hdf for fixed format.

[filmor]

Fixes #7761.

[jreback]

hmm, maybe make this a NotImplementedError (as its not invalid, just not supported). Also maybe a note in the io.rst (in the Fixed section)

[filmor]

Well, with the current fixed format it is in fact invalid. There is no way to get this right without changing block_items to point on columns using integers, i.e. their position in the zeroth axis.

[jreback]

it could be implemented (it just isn't). You are preventing the writing (which invalidates the reading) either way. The exception will be more informative (ValueError is saying something about the input is not valid. It IS valid, just not in writing :).

Actually you might want to have a nice exception message saying that you CAN store this in 'table' format.

[jreback]

say, you can store this in table format, however.

[filmor]

It could not be implemented because block_items contains only strings. You'd have to change the fixed format to implement this, for all other cases (i.e. append and compression) a ValueError is raised.

If this is the only thing keeping you from merging I'll change it to NotImplementedError ;)

[jreback]

@filmor thanks for this!

[jreback]

this is technically for a non-unique info_axis (e.g. columns for df, but 0 axis for Series/Panel). Can you add some tests for this? And update the note in v0.15.0 (I moved it a bit)