DEPR: Deprecate as_recarray in read_csv

[gfyoung]

1. Documented and deprecate as_recarray
2. Added as_recarray functionality to Python engine
3. Fixed bug in C engine in which usecols was not being respected in combination with as_recarray

[jreback]

check this on Windows

I had disabled a number of these tests because of weird crashing

[gfyoung]

I don't have Windows 64 though, which is what I think was crashing based on the SkipTest message. Don't we have Appveyor to check this for us (I notice that it isn't a requirement for merging PR's however)?

[jreback]

ok, sign up for appveyor, we auto run on this, its a free account, and I think that you can even remote login to debug.

In any event I remove the skip for the tests I marked on win64.

Also remove the test marked for 2.6 (its a slow one).

It passed for me on 3.5 on win-64, but let's see what appveyor says.

[gfyoung]

Okay, so if I understand this correctly, remove the SkipTest on test_compact_ints_as_recarray as well as the one on test_file in common.py?

[jreback]

yep

[codecov-io]

Current coverage is 84.24%

Merging #13373 into master will increase coverage by <.01%

@@             master     #13373   diff @@
==========================================
  Files           138        138          
  Lines         50739      50749    +10   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits          42740      42750    +10   
  Misses         7999       7999          
  Partials          0          0          

Powered by Codecov. Last updated by 27448d9...e71aa1e

[gfyoung]

@jreback : Removed the SkipTest as requested, and Travis is happy. Ready to merge if there is nothing else.

[jreback]

what about appveyor ?

[jreback]

@gfyoung can you make the update that @jorisvandenbossche requested here: #13360 (IOW, just edit buffer_lines)

[jorisvandenbossche]

Broader question: do we actually want to support this feature? (if we want to prune in read_csv keywords, this is maybe a good candidate). Not sure how much used this is.

Has this a significant performance/memory benefit over just using read_csv and to_records afterwards? (I didn't look at the implementation, but I can imagine this could be a reason to keep the keyword)

[jreback]

@jorisvandenbossche good point. yes I think we should deprecate this as well.

@gfyoung xref to #10631 (check box 1) for fixing those tests. The test suite works for me on windows 64 (2.7 & 3.5)

[gfyoung]

@jreback : Appveyor passed for me as well, so I think we can consider that Windows issue resolved?

@jorisvandenbossche : That's a good point. I'll add the deprecation too. That also would help with decoupling pandas from numpy as well.

[jreback]

do we test these guarantees?

[gfyoung]

Of course. The test I added here speaks for itself.

[gfyoung]

@jreback @jorisvandenbossche : deprecated as_recarray, made the documentation fix from my previous PR, and Travis is happy. Ready to merge if there are no other concerns.

[jreback]

ty