Repository navigation
Measure and bound coupling analysis on large monorepos #9
Description
Context
Coupling counts every pair of files in every commit with at most 50 universe files, keyed by string — up to 1,225 pairs per commit. Memory and time for very large monorepos (tens of thousands of commits and files) are unmeasured.
Goal
Know the cost of coupling analysis on a large repository and keep it bounded.
Acceptance criteria
- A measurement on a large public monorepo (for example 50k+ commits) records wall time and peak RSS in
docs/toolchain.mdor the README's known limits. - If peak memory exceeds a budget agreed in this issue, pair counting is bounded (for example integer file ids instead of string keys, or pruning pairs below
minSharedCommits), with the measurement repeated.
References
packages/engine/src/coupling/coupling.ts
Activity
typeweaver commentedon Sep 30, 2026
Budget used for this issue: on a public repository with 30k+ commits (
angular/angular, ~38.7k), the default 12-month window must finish in under 60 s with under 1 GB peak RSS;--since 10yis measured but not budgeted. Measured: ~10 s and ~140 MB for the Node process (git adds ~275 MB) — seedocs/toolchain.md→ Performance.