Fix: make some SQL builders pure

[georgesittas]

No description provided.

[tobymao]

can you double check and_, not_, or_ and, and paren?

[georgesittas]

Yeah will take a look, thanks for reviewing

[georgesittas]

can you double check and_, not_, or_ and, and paren?

>>> import sqlglot
>>> c = sqlglot.exp.Condition(this="x")
>>> a = sqlglot.exp.and_(c, "true")
>>> id(a.this)
4368906528
>>> id(c)
4368906528

Should we add a copy=True parameter and pass it everywhere to ensure copying by default?

[tobymao]

@georgesittas yes, and i'd just double check where it's used and see if we need to put false

[georgesittas]

Got it, will look into this soon

[georgesittas]

@tobymao can you take another look? Fixed a couple of TPC-DS tests that broke due to these changes.

For example, check this part in TPC-DS 53. From what I can understand, the optimizer converts this disjunctive normal form to conjunctive, using the distributive property of the OR / AND operators. The current diff can be seen here.

We previously had 7 clauses combined through conjunction in the optimized query, but I think we need 9 of them (the other 2 shown in the diff). A small proof of this claim:

let x = i_category IN ( 'Books', 'Children', 'Electronics' )
let y = i_class IN ( 'personal', 'portable', 'reference', 'self-help' )
let b = i_brand IN ( 'scholaramalgamalg #14', 'scholaramalgamalg #7', 'exportiunivamalg #9', 'scholaramalgamalg #9' )

let z = i_category IN ( 'Women', 'Music', 'Men' )
let w = i_class IN ( 'accessories', 'classical', 'fragrances', 'pants' )
let d = i_brand IN ( 'amalgimporto #1', 'edu packscholar #1', 'exportiimporto #1', 'importoamalg #1' )

let a = x AND y
let c = z AND w

(a AND b) OR (c AND d) ->
(a OR c) AND (a OR d) AND (b OR c) AND (b OR d) ->
((x AND y) OR (w AND z)) AND ((x AND y) OR d) AND (b OR (w AND z)) AND (b OR d) ->
(x OR w) AND (x OR z) AND (y OR w) AND (y OR z) AND (x OR d) AND (y OR d) AND (b OR w) AND (b OR z) AND (b OR d) 

If we substitute x, y, b, z, w and d here we'll get the current optimized version of TPC-DS 53 (the terms are in different order but they're all there).

The other TPC-DS test had a similar issue.