DOC: Fix reference warning for recarray.

[liang3zy22]

This is to fix reference warning for module numpy.rec. Put it in draft since I am not sure about current solution.
@rossbar , Please give comment. My current solution is add a new section for "record arrays" just after "masked arrays".
Do you think it is reasonable? If yes, I can add related content in "record arrays" section.

[melissawm]

Nitpick: if we're adding this to this page, should we organize these submodules alphabetically so that we have

np.char
np.ma
np.rec

?

[liang3zy22]

Nitpick: if we're adding this to this page, should we organize these submodules alphabetically so that we have

np.char
np.ma
np.rec

?

I put it after 'masked arrays' since they are both array objects and masked arrays are used more. I am not sure they should be alphabetically listed.

[rossbar]

Generally +1 for creating the recarray article for the refguide. I'm hesitant about exposing it in the array refguide article though, as I don't think recarrays should be used in new code (e.g. pandas is preferrable for tabular data).

[rossbar]

I'd vote to remove the mention from the top-level array refguide article, just because I don't think recarrays are recommended for new code.

[liang3zy22]

It can be removed. But a "document isn't included in any toctree" warning will be raised. I don't think adding it into routines.rst is a better choice.

[mattip]

Aren't they syntactical sugar over structured arrays to allow access by attribute rather than key? If so, I don't think they are so bad. Maybe the new reference could mention that pandas data frames are much more powerful.

[liang3zy22]

Aren't they syntactical sugar over structured arrays to allow access by attribute rather than key? If so, I don't think they are so bad. Maybe the new reference could mention that pandas data frames are much more powerful.

I added a note to use pandas DataFrame instead.

[mattip]

Sorry I didn't investigate this more closely before getting involved. What is the exact warning you see?

It seems rec array (and masked array) is already rendered under the arrays.classes heading two lines above. See the rendered page of arrays.rst. So maybe is would be better to remove both references, and improve the reference in the array subclasses page, arrays.classes.rst to avoid the warning, if possible, rather than adding a new page.

[liang3zy22]

Sorry I didn't investigate this more closely before getting involved. What is the exact warning you see?

It seems rec array (and masked array) is already rendered under the arrays.classes heading two lines above. See the rendered page of arrays.rst. So maybe is would be better to remove both references, and improve the reference in the array subclasses page, arrays.classes.rst to avoid the warning, if possible, rather than adding a new page.

You mean remove numpy.rec module completely? The warning I met is:
numpy/doc/source/reference/recordarray.rst: WARNING: document isn't included in any toctree

I think we can improve _arrays.classes.rec part, let routines.array-creation.rec part refer to _arrays.classes.rec part. And remove numpy.rec module reference from these two parts.

[mattip]

The warning I met is:

numpy/doc/source/reference/recordarray.rst: WARNING: document isn't included in any toctree

I mean the original warning, before this PR.

[liang3zy22]

The warning is:

arrays.classes.rst:514: WARNING: py:mod reference target not found: numpy.rec
routines.array-creation.rst:47: WARNING: py:mod reference target not found: numpy.rec
routines.array-creation.rst:50: WARNING: py:mod reference target not found: numpy.rec

[mattip]

Then yes, this

Let routines.array-creation.rec part refer to _arrays.classes.rec part. And remove numpy.rec module reference from these two parts.

I think would be a better solution.

[mattip]

here and here are the rendered changed pages. LGTM

[mattip]

Thanks @liang3zy22