BUG: Bug in expressions evaluation with reversed ops, showing in series-dataframe ops (GH7198)

[jreback]

closes #7198
closes #7192

[jreback]

@cpcloud this is fine in 0.13.1 and broke in 0.14.0...any clue?

[jreback]

I think this is the same issues as in #7192 (as the subtraction is not commutative)

[cpcloud]

I think this is my fault. I added the op() calls for the rops. This particular issue was untested.

[jreback]

fessing up! :)

np....this tests this explicity

took me a bit to figure out that sub/div are not commutative while add/mul ARE!

[cpcloud]

Might be better to just remove the rops op($ calls than to add the reversed check

[cpcloud]

Unless that isn't the issue. But I think that's the only thing other than the raise on bool ops that has changed between 0.13.1 and 14

[jreback]

python somehow seems to handle them correctly (e.g. it in-line reverses)

__rtruediv__(x,y)
__truediv__(x,y)

do the right thing!

whereas in numexpr the 2nd is backwards

I don't have any idea WHEN python decides to call the reversed ops, but no biggie, just added a flag and it should work

[cpcloud]

It's done at the c level in pyobject_richcompare I followed it once for some Timestamp compare stuff. Good times

[jreback]

haha....ok...no worries