Add support for REAL data type.

[eliran-lightricks]

Trying to solve #37

[andrewgross]

Just a heads up, Redshift considers REAL to be FLOAT4 (or float32() in SQLAlchemy). I think if you use float64() you will have issues reading the data with Spectrum.

[c-nichols]

@eliran-lightricks from reading the documentation I think @andrewgross is correct, REAL columns should use type pa.float32().

Relevant documentation below -- note that REALs consume 4 bytes of space, indicating pa.float32() is sufficient.

https://docs.aws.amazon.com/redshift/latest/dg/r_Numeric_types201.html

Would you mind testing with pa.float32() and see how it goes?

[eliran-lightricks]

I did some tests and it seems to work with uploading to S3 and copying back. But for some reason the tests are failing, trying to figure out exactly where.

[c-nichols]

@eliran-lightricks try using self.assertAlmostEqual(self.data, written_data, places=5) I think it's just the floating point precision causing issues with strict comparison

[eliran-lightricks]

@c-nichols Had to do some alterations since self.assertAlmostEqual can't work directly on lists. Let me know if that's okay with you.

[anomitra]

@c-nichols Did you release this on PyPI? I pip install'd this today, and this error cropped up -- got fixed when I applied the diff from the PR.

[c-nichols]

Releasing is still on my todo list :see_no_evil: I will do so within the next couple days

[anomitra]

Please do! Don't make me monkeypatch this on production 😂

[c-nichols]

@anomitra new release is up on pypi. "a few" == 18 apparently