angular/angular

Why this case study matters

Angular is one of the most-watched JavaScript framework repos on GitHub (~95k stars, ~25k forks) — a TypeScript mega-repo (~3M LoC across 16 published @angular/* packages) wired through a Bazel workspace and orchestrated by a bespoke in-house CLI (@angular/ng-dev). It’s a distinct shape from microsoft/typescript (which IS the TS compiler) and from vercel/next.js (a hybrid pnpm+Cargo dual-workspace). angular sits in the active TS-monorepo + Bazel segment that no per-language linter catches end-to-end.

For adopters in enterprise TS-monorepos using Bazel (Google’s internal Angular build, Salesforce’s Lightning Web Runtime, etc.) the same shape is replicated — and the same gaps. A repo this disciplined, with this much custom in-house tooling, is exactly the kind of place where alint is the layer that sees the cross-language API parity contract that no per-language linter does.

Headline catch

6 source files with non-canonical @license header — including packages/core/src/defer/interfaces.ts with a leading UTF-8 BOM byte that prettier doesn’t catch and tslint doesn’t see. Plus 5 packages without an index.api.md golden that pnpm public-api:check silently no-ops on (the existing tool only diffs against goldens that already exist; a missing golden is a free pass). Plus 2 packages with non-canonical version field (zone.js at 0.16.1, benchpress at 0.4.0-PLACEHOLDER — the latter won’t substitute correctly during release).

These are drifts no existing tool in angular’s pipeline catches: tslint sees source semantics, prettier sees formatting, pnpm public-api:check only diffs goldens that exist. alint sits exactly in the gap.

Where alint earns its keep here

• goldens/public-api/<pkg>/ discipline is THE canonical example of the v0.11+ cross_language_implementation_complete shape applied within a single language — TypeScript source ↔ TypeScript API surface golden, 50 goldens locking the public surface of 13 of 16 packages. angular is one of 5 saturated demand sources for this primitive (alongside arrow’s per-language schema parity, TF’s 1,185 textproto goldens, protobuf’s 10 in-tree language bindings, and flutter’s 6 native-OS embedders).
• The .ng-dev/ in-house CLI is a fascinating contrast to the more-typical husky+lint-staged+changesets stack. angular built its own toolchain and the structural invariants around its config surface are exactly what alint expresses well.
• Per-package manifest discipline — 16 package.json files enforced in single-digit milliseconds, vs. the ~2-3 s of warm-cache startup that sequential node -e "require()" calls would burn.
• The same repository.directory regression class that bit react-refresh in the facebook/react case study is caught here at config-parse time.

Future story angles

• Launch tile candidate — position as the canonical “TS-monorepo + Bazel-built mega-framework with strict per-package discipline” data point alongside kubernetes, airflow, microsoft/typescript, vercel/next.js, facebook/react.
• Cross-language parity narrative — pair angular’s goldens/public-api discipline with arrow’s per-language schema test fixtures, TF’s textproto goldens, protobuf’s bindings, and flutter’s embedders for a “5 saturated sources” launch-post on cross_language_implementation_complete shipping in v0.11+.
• Enterprise TS-monorepo angle — same shape replicates in Google internal Angular, Salesforce LWR, and many other Bazel+TS monorepos. A “what alint does for your enterprise TS-monorepo” case study can generalise from this.