Omnisci Core - complete the endpoint work to allow registration

dharhas commented 5 years ago

pearu commented 5 years ago

https://github.com/omnisci/mapd-core-internal/pull/3342

pearu commented 5 years ago

Current state:

[x] Feature complete.
[x] Ensure all CI tests succeed (1-2h)
[x] Needs a review from Omnisci [loadtime UDF triggers multiple symbols errors when GPU enabled]
[x] Debug loadtime UDF errors when GPU enabled (5h)

pearu commented 5 years ago

Alex left this comment to UDF PR: """ I think we want to keep this test behavior, as discussed in Slack: SELECT TRUNCATE(2016, 1) FROM test LIMIT 1; in current master fails because 2016 is a smallint (int16_t) constant but is interpreted as just an INTEGER in calcite so we have TRUNCATE() returning INTEGER in calcite, but smallint in the specialization this seems kinda like a nonsense test... on the other hand, we do have to worry about the following problem: if you define

EXTENSION_NOINLINE
int32_t udf_diff(const int32_t x, const int32_t y) {
   return x - y;
}

and then pass it 64-bit values, it will just truncate them to 32-bits silently this happens now, so I am not saying it is a blocker for pearu's work but we want to make sure that we are doing the appropriate casts, and not allowing automatic narrowing casts, in the future. """ I have to look at how to prevent automatic truncation.

pearu commented 5 years ago

The UDF PR has been approved and merged to omniscidb-internal repo.

Quansight / omnisci

Omnisci Core - complete the endpoint work to allow registration #27