Build better software to build software better

This post explains how the Slack/Quip build team reduced 60-minute build times by applying software engineering principles (caching and parallelization) to their Bazel-based build pipeline.

• •Build performance is modeled as a directed acyclic graph (DAG) where targets have explicit inputs, outputs, and commands
• •Caching works best when build units are hermetic (only use declared inputs) and idempotent (same inputs always produce same outputs)
• •Granularity of cache keys directly impacts cache hit rate — smaller, more focused build units yield better cache performance
• •Parallelization requires rigorously defined input/output boundaries so work can be distributed across CPU cores or remote build clusters
• •Bazel enforces hermeticity via sandboxing and automatically handles caching, parallelization, and minimal rebuild of only affected targets