Properly cast reductions when necessary

[cpcloud]

closes #1050

also restricts the scope of most reductions (notable exceptions are count and nelements)
to only work on sqlalchemy.elements.ColumnElements, rather than the generic ClauseElement.

[cpcloud]

cc @llllllllll can you try this out on a few things on your side? this touches some often hit code paths, i want to make sure nothing major is borked.

[cpcloud]

ugh it looks like spark 1.2 doesn't allow casting to anything but DECIMAL

[cpcloud]

@llllllllll i'd like to get this in, any chance you could review this tomorrow and/or try it out to make sure there are no major breakages? i've run it on some simple redshift queries (this was partly the impetus for this change) and things are working

[llllllllll]

Sure thing, I will test this tomorrow morning at work

[cpcloud]

awesome thx!

Best,
Phillip Cloud

On Sun, Jun 14, 2015 at 11:04 PM, Joe Jevnik notifications@github.com
wrote:

Sure thing, I will test this tomorrow morning at work

—
Reply to this email directly or view it on GitHub
#1130 (comment).

[llllllllll]

To test, I created a table with a single bool column in postgres; however, I could not cast bools to floats

ProgrammingError: (psycopg2.ProgrammingError) function sum(boolean) does not exist
LINE 1: SELECT sum(test.c) AS c_sum 
               ^
HINT:  No function matches the given name and argument types. You might need to add explicit type casts.
 [SQL: 'SELECT sum(test.c) AS c_sum \nFROM test']

[llllllllll]

bz=# select cast(False as int);
 int4 
------
    0
(1 row)

bz=# select cast(False as float);
ERROR:  cannot cast type boolean to double precision
LINE 1: select cast(False as float);
               ^
bz=# select cast(False as int);
 int4 
------
    0
(1 row)

bz=# select cast(cast(False as int) as decimal);
 numeric 
---------
       0
(1 row)

It looks like we can actually cast to int an then to numeric though

[cpcloud]

i think casting to int is probably okay for now

[cpcloud]

closing in favor of #1137