Support extract(... from ...)
#450
Conversation
pdobacz
force-pushed
the
piotr/extract-datetime
branch
from
0b83f64
to
474eb7b
Compare
pdobacz
force-pushed
the
piotr/extract-datetime
branch
4 times, most recently
from
ae71b67
to
f747e84
Compare
| if ((aggoid == g_oid_cache.sum_noise || | ||
| aggoid == g_oid_cache.avg_noise) && | ||
| TypeCategory(linitial_oid(aggref->aggargtypes)) == TYPCATEGORY_DATETIME) | ||
| FAILWITH_LOCATION(aggref->location, "Unsupported aggregate in query."); |
There was a problem hiding this comment.
Why are these aggregators allowed by type system while sum(timestamp) isn't? Can't we make the argument type identical with the original?
There was a problem hiding this comment.
Hm, I'm not sure I can explain this well, but I think this comes of the fact, that sum comes in many flavors, and sum(timestamp) isn't one. We track and rewrite all of these flavors to "our" anon_sum(any).
With _noise it is not rewritten but it is our own user-defined aggregate and it comes in one flavor diffix.sum_noise(any) from the "beginning". We could in theory define multiple flavors of dummy sum_noise and rewrite them to sum_noise_anon, to have the same semantics as sum. I don't remember anymore, why this path wasn't chosen (simplicity probably). I'm not sure if we have enough justification to do it now or if it's possible at all.
pdobacz
force-pushed
the
piotr/extract-datetime
branch
from
f747e84
to
8e82bad
Compare
... with a bonus bugfix