fix: #1574 nullable return types for groupby functions #1631
Conversation
|
I haven't found any docs that mention what postgres instance I should use to generate pg_catalog. This is needed to complete the pull request and fix tests. Could anyone direct me on where to look? |
| @@ -23,7 +23,8 @@ SELECT p.proname as name, | |||
| format_type(p.prorettype, NULL), | |||
| array(select format_type(unnest(p.proargtypes), NULL)), | |||
| p.proargnames, | |||
| p.proargnames[p.pronargs-p.pronargdefaults+1:p.pronargs] | |||
| p.proargnames[p.pronargs-p.pronargdefaults+1:p.pronargs], | |||
| CASE WHEN p.prokind = 'a' THEN TRUE ELSE FALSE END | |||
There was a problem hiding this comment.
I mentioned this in the discussion we were having, but I thought it is worth mentioning here.
We probably should do a CASE WHEN p.prokind = 'a' AND p.name != 'COUNT ....' since COUNT is an exception to the nullability rules.
Also, I don't think there needs to be a CASE since you should just be able to use p.prokind = 'a' AND p.name != 'COUNT' as a boolean expression
| } | ||
| cols = append(cols, &Column{ | ||
| Name: name, | ||
| DataType: columns[0].DataType, |
There was a problem hiding this comment.
I think that unfortunately the fix is a lot more complex than this, this is saying that any function with more than 0 args the return type should the type of its first arg?
This is true for most aggregates, (min, max, avg, every) but is not generally true of any function call - in particular count, array_agg and probably any user-defined aggregate.
https://www.postgresql.org/docs/15/functions-aggregate.html
I think special cases for these common aggregates is probably a useful change.
This is a proposed fix for #1574. I am unsure if this change breaks any tests because I don't have a correctly setup postgres instance to generate
pg_catalogcorrectly with all extensions. However, I have checked that this change fixes the bug.