[tr1v3r]

Fixes #350.

Problem

Since v1.6.1 (PR #340), Migrator().DropIndex() and Migrator().RenameIndex() fail in two ways:

1. No explicit schema (the common default): CurrentSchema returns clause.Expr{SQL: "CURRENT_SCHEMA()"}, and the new DROP INDEX ?.? / ALTER INDEX ?.? RENAME TO ? templates splice it into an identifier position:

   DROP INDEX CURRENT_SCHEMA()."idx_users_name"
   ERROR: syntax error at or near "CURRENT_SCHEMA" (SQLSTATE 42601)

   CURRENT_SCHEMA() is a function call; PostgreSQL cannot use it to qualify an identifier.

2. Explicit schema (TableName() like public.users): the schema is returned as a plain string, which ? binding sends as a parameter:

   DROP INDEX $1."idx_users_name"

   also invalid — identifiers must be quoted, not bound.

This breaks AutoMigrate flows that drop/recreate indexes (e.g. gormigrate) and keeps the gorm core repo's postgres CI matrix red.

Fix

Qualify the index only when the table has an explicit schema, using a quoted identifier:

• CurrentSchema returns a string (explicit schema from schema.table or TableExpr) → DROP INDEX "schema"."idx" / ALTER INDEX "schema"."idx" RENAME TO "idx_new"
• otherwise → DROP INDEX "idx" / ALTER INDEX "idx" RENAME TO "idx_new" (pre-v1.6.1 behavior; the index resolves via the search_path)

Tests

New migrator_drop_index_test.go records the SQL through an injected gorm.ConnPool (no live PostgreSQL needed) and asserts both scenarios for DropIndex and RenameIndex. Verified red → green: pre-fix the tests fail with exactly the two invalid statements above; post-fix all pass, full package go test ./... green.

[tr1v3r]

The red run-tests check here is a repo-wide workflow issue, not this change: the workflow pins go: 1.23.0 while master's go.mod requires go >= 1.25.0 (fails identically on #352). Proposed workflow fix in #354; this PR goes green once it lands (rebase or re-run).

For reviewers comparing with #351: this PR deliberately leaves CurrentSchema untouched, so all existing call sites keep the inline CURRENT_SCHEMA() comparison (evaluated by PostgreSQL itself) and no extra SELECT current_schema() round-trip is added per migrator operation. Only DropIndex/RenameIndex change their behavior — the two methods that are broken today. Both failure modes from #350 are covered by the new gorm.ConnPool-based regression tests (red → green verified).