Mammoth Cache for Gradle® and Apache Maven™

Transparent, incremental dependency caching for Gradle and Maven builds on GitHub Actions — from simple single-job pipelines to large parallel fan-out workflows.

One action wraps your build step. Before the build, it restores the newest compatible immutable cache generation. After a material change, a standalone job publishes a new complete generation; distributed workers instead exchange precise deltas through an aggregator. No manual cache-key juggling, and unchanged standalone runs publish no duplicate generation.

Pre-release status: Mammoth Cache does not have a published release yet. The links below lead to unreleased development documentation. The current source tree does not contain the generated action bundles and is not directly installable through a uses: reference.

---

Preview the unreleased development guide Development docs (unreleased)

Docs cover workflow usage, configuration, cache-partition behaviour, security notes, and maintenance guidance.

Current scope

Mammoth Cache currently targets dependency and build-tool cache orchestration on GitHub Actions; it is not a general build-output cache, build sandbox, or dependency-provenance verifier. Distributed use requires a dedicated aggregator job. Other CI providers remain future work.

Highlights

🔁 Distributed deltas — worker artifacts package only files changed relative to their restored base; the aggregator merges them into a complete generation for later runs.

🔒 Secure Gradle wrapper provisioning — wrapper JARs are checksum-validated and GPG-verified against the official Gradle release signing key before any code runs.

🗂 Content-fingerprinted cache keys — changing the cache partition layout automatically produces a new key lineage; no manual version bumps required.

🧹 Age-based cleanup — best-effort timestamp garbage collection removes old managed Gradle and Maven cache entries before a changed generation is published.

🚫 Read-only on pull requests — PRs restore the shared cache but never write back, keeping the main cache clean by default.

Single-job and distributed workflows

Mammoth Cache works transparently for both simple and complex pipeline shapes.

Single-job (standalone)

One action call is all it takes. The action runs transparently before and after your build step.

flowchart LR
    RC([Restore\ngeneration]) --> B([Build]) --> SD([Publish changed\ngeneration])

Distributed multi-job

Worker jobs run in parallel and each publish a delta artifact. The aggregator job downloads, merges, and applies them all into a single coherent cache that every future run benefits from.

flowchart LR
    WA([Worker A]) --> AG
    WB([Worker B]) --> AG
    WC([Worker C]) --> AG
    AG([Aggregator\nmerge + apply]) --> C[(Shared\ncache)]