Repository navigation
perf(api): cap or paginate the /analysis/{id}/architecture payload #508
Description
Problem
GET /analysis/{repository_id}/architecture returned about 9.6 MB (343 nodes, 13,834 edges) for PARTHA's own repository. The frontend now avoids rendering all of it (#502), but the full payload is still downloaded and parsed on every visit.
Idea
Keep the sealed-snapshot contract intact while letting clients ask for less: for example a default summary (nodes plus an edge count and per-type totals) and a separate endpoint or query parameters to fetch the edges for one node or one page.
Acceptance criteria
- Default response size on the repository above drops substantially.
- Full data is still reachable and consistent with the snapshot (same hash, same revision identity).
- The OpenAPI contract is regenerated and drift-checked (
npm --prefix apps/frontend run generate:api-contract -- --check). - Backend tests added; existing owner-scoping behaviour unchanged (another owner still gets 404).
Pointers
apps/backend/app/api/routes/analysis.py, docs/CAPABILITIES.md.
Activity
SHAURYAKSHARMA24 commentedon Oct 8, 2026
Audit F11 remediation and remaining compatibility decision
PR #550 now bounds endpoint SQL parameters and omits nested symbol keys unused by module inventory. All 28 architecture/authentication-explanation tests pass. The public response is unchanged: an 88-node/640-edge disposable fixture remains 399,311 bytes and its JSON is identical before/after. PR #554 adds measured browser coverage, #549 finer extraction progress, and #558 an explicit edge-type filter. No evidence is silently capped.
The default response-size acceptance criterion is still open. Existing consumers expect complete
ArchitectureResponse.nodesand.edges; silently replacing either with a partial set would be a breaking contract change, reserved to Parth by the remediation request.Options:
Recommendation: the additive compatibility path, followed by representative large-repository benchmarks. The present PR is intentionally partial and does not close #508. No snapshot schema/hash change is proposed. Approval is needed for the default-contract transition, not for the internal fixes already implemented.