You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
#2307
For 2299, we were only resolving the column type, and not the types within default values, so we ended up with an unresolved type that was persisted in the schema. Since we assume that every deserialized type is resolved, this would cause a panic when fields were missing.
For 2307, we return the schema from the unanalyzed node obtained right after parsing the node. Our analysis passes handle type conversions between GMS types (with EXISTS being a node defined in GMS), and therefore we were returning a GMS type (which was being automatically converted to the wrong type). We now attempt an analysis pass before determining the schema, which should work in most cases, but may fail for some bindings. In those cases, we ignore the analysis error and return the original schema. This isn't a complete solution, but it at least gets us further in all other cases.
This most likely fixes:
#2301 2301 seems related to 2299. We don't appear to have any intrinsic problem with many CREATE TABLE statements within Rust's sqlx library (as used in the issue filed, 30 statements work fine), so it appears tied to the other issue. It's possible that there are further issues masked by the aforementioned one, but that cannot be determined without the user's migration data.