BLD: use sqlalchemy 0.7.1 in 2.6 build (GH6340)

[jreback]

closes #6340

[jreback]

@jorisvandenbossche

ok.....good thing we tested!

this is 0.7.1 for sqlalchemy, and slightly older versions of pymsql and psycopg2

https://travis-ci.org/jreback/pandas/jobs/24212656

[jorisvandenbossche]

strange, the integer columns fail. We convert it to a string (https://github.com/pydata/pandas/blob/master/pandas/io/sql.py#L586), but for some reason sqlalchemy converts it back to an int. Have to investigate.

[jreback]

ok...my pr is up their...lmk when you want to merge (or you can simply take and rebase on fix ontop)
up 2 you

also ok with simply skipping that test on a too old version of sqlalchemy (seeing as its the only 1 that failed).

[jorisvandenbossche]

But possibly also a bug on our side (as here https://github.com/pydata/pandas/blob/master/pandas/io/sql.py#L520 we don't convert the column names to strings when constructing the dict which is passed to execute). Will look at it later.

In any case, say it is only this that is not working for sqlalchemy 0.7. Do we say they need 0.8, or just say 0.7 is possible but that does not work? Can we eg skip that one test when using sqlalchemy 0.7?

[jreback]

I would just say min in 0.7.1 and leave it at that

then skip this test and/or fix. good enough (in 0.14.1 can extend the testing if you'd like)

[jreback]

@jorisvandenbossche you are going to cover the build change in another PR, yes? or want me to keep this open

[jreback]

bumping this; for 0.14.1 can work this in (and test with MySqlDB as well), https://pypi.python.org/pypi/MySQL-python/1.2.5

[jorisvandenbossche]

@jreback Can you rebase this? Normally tests should now pass after #7456