Summary

pnpm lint has been failing repo-wide since the eslint 8→9 bump (eslint 9 only reads flat config; we only shipped .eslintrc.json). This consumes the shared flat config published from dev-utils (constructive-io/dev-utils#98) and gets pnpm lint to exit 0.

.eslintrc.json is deleted; eslint.config.mjs is the shared array plus repo-local adjustments. The bulk of the diff is one mechanical eslint . --fix commit (indentation + import sorting), kept separate from the config commit and from the ~20 files that needed a human.

Deviations from a literal port of .eslintrc.json

comma-dangle: ['error', 'never'] is dropped. It produced 57,145 errors — the codebase uses trailing commas essentially everywhere, and .prettierrc.json sets trailingComma: 'es5', so the rule and the formatter have been in direct contradiction (invisibly, since lint never ran). Prettier owns trailing commas now. If you'd rather enforce never, that's a separate mass-reformat, and prettier config has to change with it.

Generated and verbatim-copied files are ignored. Linting them is churn the next pnpm generate reverts — and in two cases it actually changes behavior:

• codegen output: sdk/{constructive-cli,constructive-react,constructive-sdk}/src, sdk/migrate-client/src/migrate, **/__generated__/**, **/*.generated.ts, postgres/pg-ast/src/{asts,wrapped}.ts (~50k of the original errors)
• graphql/codegen/src/core/codegen/templates/** — these are read verbatim at codegen time and written into every generated client, so import sorting there silently changes generated output (it broke the client-generator snapshots)
• **/__fixtures__/** — codegen output snapshots and standalone sqitch/pgpm fixture projects that tests compare against

The sdk ignore is per-package rather than sdk/*/src/**: the latter left sdk/migrate-client with zero lintable files, which eslint treats as an error.

Rules turned off for idioms this repo relies on: no-namespace (declare global { namespace Grafast { ... } }), no-unsafe-function-type (Function in plugin/callback signatures), no-empty-object-type (StrictSelect<S, Shape> = ... ? {} : never). Scoped: no-control-regex off for pgpm/logger, no-unused-expressions off for **/*.type-test.ts.

linterOptions.reportUnusedDisableDirectives: 'off'. Worth knowing about: eslint 9 defaults it to warn, and --fix deletes disable comments it considers unused — including /* eslint-disable @typescript-eslint/no-empty-object-type */ at the top of a generated file, leaving a stray blank line the generator would re-add. Turning it off leaves ~30 existing directives intact.

Hand fixes (all semantics-preserving)

-return stmt && !stmt.hasOwnProperty('TransactionStmt');
+return stmt && !Object.prototype.hasOwnProperty.call(stmt, 'TransactionStmt');

-const tokenRe = /(^|[\s\[])([^\s\[\]@]+)(?=$|[\s\]@])/g;
+const tokenRe = /(^|[\s[])([^\s[\]@]+)(?=$|[\s\]@])/g;

 const info = project.getModuleInfo();
-info.version;

plus comments in empty catch {}/finally {} blocks, and two targeted disables (no-this-alias in MigrateTestFixture, no-async-promise-executor in s3-utils download — the executor is genuinely async and rewriting it is out of scope for a lint PR).

Verified

pnpm lint → 0 errors / 191 warnings (all no-unused-vars, warn by design). pnpm build clean, full CI green.

Follow-up

CI doesn't run lint — which is exactly why this rotted for a whole major version. Adding a lint job to run-tests.yaml is the thing that keeps it fixed.

Link to Devin session: https://app.devin.ai/sessions/3b5f5e76ddc74eda83a3543d7da2c75f
Requested by: @pyramation