pg-gen: make sqlc-pg-gen the complete source of truth for pg_catalog.go #1809
Conversation
|
@kyleconroy I think given the fact that you have been pretty diligent about adding endtoend tests whenever you change something, I think as long as this passes the test, it should be safe. But if you can remember any other manual edits to pg_catalog as well, we'll definitely need to fix them - this could possibly be a breaking change if we don't catch one of those edits. |
This is needed for the end-to-end tests, at least until resolving a function call is smart enough to match functions parameter types
| @@ -79,7 +79,7 @@ ORDER BY bucket DESC | |||
| type ListMetricsRow struct { | |||
| Bucket interface{} | |||
| CityName sql.NullString | |||
| Avg string | |||
| Avg float64 | |||
There was a problem hiding this comment.
The query is:
SELECT time_bucket('15 days', time) AS bucket, city_name, AVG(temp_c)
FROM weather_metrics
WHERE DATE_SUB(NOW(), INTERVAL 6 MONTH)
GROUP BY bucket, city_name
ORDER BY bucket DESC;
So I think the float64 is actually correct here - the string was incorrect before. They are also "both wrong" since the order of the function definitions in pg_catalog.go is actually what is determining the types here.
| @@ -104,7 +104,7 @@ func Pgcrypto() *catalog.Schema { | |||
| Name: "digest", | |||
| Args: []*catalog.Argument{ | |||
| { | |||
| Type: &ast.TypeName{Name: "text"}, | |||
| Type: &ast.TypeName{Name: "bytea"}, | |||
There was a problem hiding this comment.
Changed the order of these, as they affected the end-to-end tests, and just wanted to preserve legacy behavior.
| }, | ||
| }, | ||
| ReturnType: &ast.TypeName{Name: "text"}, | ||
| }, |
There was a problem hiding this comment.
This concat is not in the pg_catalog on the main branch, so I removed it here too.
| }, | ||
| }, | ||
| ReturnType: &ast.TypeName{Name: "numeric"}, | ||
| }, |
There was a problem hiding this comment.
Simply changed the order of the functions here to preserve the behavior of the e2e tests.
This is sort of a yak-shaving task for #1748
Problem: there seem to have been some manual edits to pg_catalog.go over the course of time. This makes it hard (almost impossible actually) to do any change there with any confidence.
Solution: Do some detective work on the manual edits performed, and incorporate them into pg-gen. Once that is done the work on #1748 will be a lot easier.
Here are the 2 manual edits as far as I can tell:
9513155 (pg_catalog functions with variadic signatures - this is completely solved by this PR)
And
846c629 (manual edits to pg_catalog that I believe will be fixed "right way" once "direct args" vs "aggregated args" are separated in #1748 ).
Commits in order tell a story, probably best to review that way.