alint — apache/spark

4 languages, 49 Maven modules, 3 PySpark packaging variants, 72 GHA workflows, and 6 per-language lint scripts — alint sees the structural shape no per-language linter does.

Why this case study matters

apache/spark is the flagship “Maven-multi-module polyglot Apache TLP” story for the launch — the canonical big-data engine (~38k stars, used at every FAANG, every major bank, every cloud provider). Naming it as a target gives alint instant credibility with the JVM + data-engineering audience.

Spark is the canonical 4-language Apache TLP polyglot monorepo — Scala (the core engine) + Java (the ASM-level extension points) + Python (PySpark, packaged as three distinct PyPI distributions) + R (SparkR, packaged for CRAN). Glued together by a Maven multi-module build (49 pom.xml files, 48 declared <module> entries) and a 72-workflow GitHub Actions matrix.

Where apache/arrow has a parity-mandate shape (every type implemented in every language), apache/spark has a per-language-MODULE mandate shape: each language tier sits at a different layer of the Spark architecture, with its own Maven module set, its own packaging conventions, its own published-artifact namespace, and its own canonical lint script under dev/lint-<lang>.

Headline catch

arrow + spark + airflow give us 3 Apache TLPs with a converging structural-discipline pattern — promoting the proposed apache/governance@v1 bundled ruleset from “v0.10+ idea” to “v0.10 ship-target”. 9 of 12 governance artefacts converge across all 3 TLPs examined (with the 3 binary-distribution artefacts gated on whether the TLP ships binary).

That’s a strong enough signal to ship apache/governance@v1 in v0.10 alongside the existing compliance/apache-2@v1. Adopters can then drop down to a 3-line extends::

extends:
  - alint://bundled/oss-baseline@v1
  - alint://bundled/compliance/apache-2@v1
  - alint://bundled/apache/governance@v1   # ← v0.10 ship-target

…and pick up the entire Apache-discipline check set without restating the 6 governance rules in every TLP-specific config.

Where alint earns its keep here

No per-language linter sees the cross-language MODULE shape — scalastyle only sees Scala, checkstyle only sees Java, ruff/black/mypy only see Python, lintr only sees R. The invariants alint enforces (Maven multi-module integrity, per-language tool-config presence, the 3-PySpark-packaging-variants uniformity, R/pkg/DESCRIPTION + NAMESPACE + cran-comments.md set, two-tier license discipline, .asf.yaml ASF-infra config, RAT exclude registry presence) are exactly the layer alint owns.
apache/governance@v1 confirmed v0.10 ship-target — spark is the headline driver, with arrow + airflow as confirming sources.
The Maven multi-module shape (49 pom.xml files, 48 declared <module> entries) gives alint a structural primitive that pnpm + yarn + cargo workspaces also need but currently express via hard-coded for_each_dir lists. Adding xml_path_* (now v0.10 ship-target via dotnet/runtime’s scale, alongside spark’s 49 pom.xml) and registry_paths_resolve lands the workspace-registry consistency story in one shape across all build systems.
3 PySpark packaging variants × 2 required files each = 6 file-existence assertions wrapped in a single for_each_dir rule. No Python tool checks the layout because no Python tool sees the layout from above.

Future story angles

Launch tile — position alongside kubernetes, airflow, microsoft/typescript, next.js, arrow with the angle: “apache/spark has 4 languages, 49 Maven modules, 3 PySpark packaging variants, 72 GHA workflows, and 6 per-language lint scripts under dev/. alint is the layer that sees the structural shape that no per-language linter does — and the Apache-governance bundle ships in v0.10.”
Apache TLP narrative — pair with arrow + airflow for the “3 TLPs converge → apache/governance@v1” story. Position as “drop one line in your extends: and inherit the Apache TLP discipline that the ASF Incubator graduation process requires”.
JVM + data-engineering narrative — Spark adoption brings immediate credibility with JVM and data-engineering audiences who may not yet have seen alint.
Demand-saturation evidence for v0.10 candidates — spark is a source for registry_paths_resolve (8 sources total), xml_path_* (2 sources, promoted by dotnet/runtime scale), generated_file_fresh (6 sources), and cross_language_registry_consistency (refinement of the v0.11+ family).