Test identifier handling against real engines and hostile schemas

.github/workflows/ci.yml

@@ -63,3 +63,44 @@ jobs:
           chrome_path="$(npx --yes @puppeteer/browsers@latest install chrome@151.0.7922.76 | awk '{ $1=""; sub(/^ /,""); print }')"
           node test/e2e-smoke.mjs "$chrome_path"
         working-directory: packages/browser-extension
+
+  # Real engines, hostile schemas, no model. A mixed-case Postgres schema once failed every query
+  # and shipped that way, because every database test used tables we had written ourselves.
+  schema-regression:
+    runs-on: ubuntu-latest
+    services:
+      postgres:
+        image: postgres:16
+        env:
+          POSTGRES_PASSWORD: pg
+          POSTGRES_DB: schema_regression
+        ports:
+          - 5433:5432
+        options: >-
+          --health-cmd pg_isready --health-interval 5s --health-timeout 5s --health-retries 12
+      mysql:
+        image: mysql:8
+        env:
+          MYSQL_ALLOW_EMPTY_PASSWORD: 'yes'
+          MYSQL_DATABASE: schema_regression
+        ports:
+          - 3307:3306
+        options: >-
+          --health-cmd "mysqladmin ping -h127.0.0.1" --health-interval 5s --health-timeout 5s --health-retries 20
+    steps:
+      - uses: actions/checkout@v5
+      - uses: pnpm/action-setup@v5
+      - uses: actions/setup-node@v5
+        with:
+          node-version: 22
+          cache: pnpm
+      - run: pnpm install --frozen-lockfile
+      - run: pnpm build
+      # The harness exits non-zero if an engine is unreachable, so a silent skip cannot pass here.
+      - run: pnpm test:schemas -- --require=postgres,mysql,sqlite
+        env:
+          ASKSQL_PG_URL: postgres://postgres:pg@127.0.0.1:5433/schema_regression
+          ASKSQL_MYSQL_HOST: 127.0.0.1
+          ASKSQL_MYSQL_PORT: '3307'
+          ASKSQL_MYSQL_USER: root
+          ASKSQL_MYSQL_PASSWORD: ''

CONTRIBUTING.md

@@ -22,6 +22,7 @@ pnpm test          # the full suite (see gating below)
 pnpm format:check  # prettier, as CI runs it
 pnpm coverage      # what CI runs instead of `test`; the coverage floor applies only with --coverage
 pnpm test:packaged # installs the packed tarballs outside the workspace and exercises every export
+pnpm test:schemas  # identifier handling against real engines and hostile schemas; needs Postgres and MySQL
 pnpm verify        # all of the above in the order CI runs them; use this before pushing
 ```
 
@@ -71,6 +72,25 @@ To exercise them, provide:
   model. Per-provider model overrides use `ASKSQL_<PROVIDER>_MODEL`.
 - **Browser E2E** - a Chrome install; the tests drive it via `puppeteer-core`.
 
+### Schema regression
+
+`pnpm test:schemas` runs the engine's identifier handling against real databases using schemas
+built to break the rules: mixed case, reserved words, spaces, unicode, names the SQL parser treats
+as keywords. It needs no model, because this class of defect does not need one to surface and a
+model would only make the result non-deterministic.
+
+It exists because a mixed-case Postgres schema once failed **every** query and shipped that way for
+weeks. Every database test we owned used tables we had written ourselves, so they shared our blind
+spots by construction.
+
+Two rules keep it honest, and both matter more than the pass count:
+
+- On an engine that folds unquoted names, each fixture first asserts the **bare form genuinely
+  fails**. A fixture that passes before the fix is testing nothing.
+- An unreachable engine is reported as **skipped, never as a pass**. CI passes
+  `--require=postgres,mysql,sqlite` so the job cannot go green on the embedded engine alone when the
+  service containers fail to start.
+
 The security boundary is developed **test-first**: add or extend a case in the
 `guard-security` / `guard-fuzz` suites before changing the guard.
 

package.json

@@ -17,6 +17,7 @@
     "test:watch": "vitest",
     "coverage": "vitest run --coverage",
     "test:packaged": "node tools/packaged-consumer-test.mjs",
+    "test:schemas": "node tools/schema-regression.mjs",
     "test:real-db:load": "node tools/real-db-load.mjs",
     "test:real-db": "node tools/real-db-e2e.mjs",
     "test:schema-sweep": "node tools/schema-sweep.mjs",

packages/jetbrains/src/main/kotlin/com/rahulmahadik/asksql/ide/engine/IdentifierCase.kt

@@ -24,10 +24,10 @@ object IdentifierCase {
         return "$quoteChar$name$close"
     }
 
-    /** Where a literal or comment ends, or -1 when the position starts neither. */
     /** A dollar-quoted body is a literal in Postgres and DuckDB, and may contain anything. */
     private val DOLLAR_OPEN = Regex("""\$[A-Za-z_]\w*\$|\$\$""")
 
+    /** Where a literal or comment ends, or -1 when the position starts neither. */
     private fun skipTo(sql: String, i: Int, doubleQuoteIsLiteral: Boolean, backslashEscapes: Boolean = false): Int {
         val ch = sql[i]
         val next = if (i + 1 < sql.length) sql[i + 1] else ' '