Count Patchula

Appearance

Phase 1 wave 2 coverage ​

All 13 Pkl examples validate (12 comparison repositories and benchmark-public). The offline engine coverage test applies every config to independently authored representative manifests and evaluates extracted source/versioning combinations. It asserts Cargo/Compose for fnox, Go for immich-go, Terraform for cafeteria-order-tracker, and Kubernetes/Kustomize/Flux/Helm values/requirements for maccrae-infra. Existing npm/Bun, lockfile, workflow and mise tests remain active. Manager, source and scheme dispatch cover every value in the bundled schema; lookup has no remaining "not implemented" branch for these enums.

This is a config/dispatch and fixture check, not a fresh inventory of the twelve repositories. Their live comparison remains the next integration check. No Renovate source, presets or fixtures were inspected. New fixture bodies came from public registry endpoints and public jdx repositories via the recording tasks; ordinary tests replay without constructing a network transport.

Release-mode offline replay on this workspace, with compilation outside measured intervals:

  • Wave 2 registry sources: 24.5–36.4 ms cold, 7.6–14.0 ms warm across two runs. Cold: 13 HTTP requests. Warm: 5 requests, including 3 conditional 304s. The second run used 370,616 bytes for SQLite, WAL and shared-memory files.
  • Existing public source fixture: 36.8 ms cold, 14.3 ms warm; 19 cold requests, 3 warm requests and 2 warm 304s.
  • Public two-repository engine fixture, now including Cargo: 337 ms cold, 82.9 ms warm; 58 cold requests, 6 warm requests, 1 warm 304, four unique catalog lookups and two warm extraction hits.

These are small fixture replays, not estimates of live network latency. The provided live renovate-runner baseline (4.0 s cold, 0.9–1.1 s warm, 0.85 MB cache) has not been remeasured here. New shared-index/registry reads coalesce overlapping requests; later sequential calls retain normal freshness/revalidation behavior. Token refresh is single-flight per repository, while manifests can fetch in parallel. Registry tags use complete lexical pagination because a known tag is not a safe stopping point. Catalogs and digest keys remain shared across a batch.

Remaining limits and recommendations ​

  • Registry authentication covers anonymous public Docker Hub/GHCR bearer flows. Private registry credentials, arbitrary cross-origin token services, registry mirrors, Terraform custom hosts and provider service discovery are not wired. Keep those explicit extensions; do not send GitHub credentials to registries.
  • Cargo git/custom registries, Terraform git/local modules, Kustomize remote git resources and pip editable/VCS/URL dependencies are not registry update targets. Cargo/Terraform locks enrich direct declarations; transitive lock entries do not become independent dependencies.
  • YAML image aliases/templates and rendered Helm templates need concrete scalar locations. Helm images support strings and repository/tag/digest maps, not all chart-specific naming conventions. Flux source references must be present in selected files; chartRef releases use the OCIRepository version separately. Digest-only OCI refs cannot infer a release tag without exhaustive manifests.
  • Requirements extraction handles versioned requirement lines with extras, markers and hashes. Unpinned requirements, arbitrary include/constraint file traversal and custom pip indexes need a separate policy/configuration design.
  • Go replace/exclude/retract semantics and Docker ARG default/conditional shell operators are not interpreted. Docker supports ordinary global ARG substitution and stage aliases. Go/toolchain directives have distinct kinds but share the public Go release backend.
  • Preserve/widen retains satisfied constraints; unsupported compound range edits fail explicitly. Retain the default exact policy for the comparison run.
  • Docker publication timestamps are unavailable in the standard tag list, so those releases have no age delay. Rate limits surface Retry-After; automatic retry storms are avoided. Decide on a larger retry budget only after observing the twelve-repo live run.
  • Mutation, dependency installation, lockfile rewriting and the CLI run command remain later-phase work. Lookup is read-only.

There are no blocking configuration questions for the requested lookup matrix. Recommended next check: run the twelve-repository live comparison, inspect warnings for these limits, and compare cold/warm request counts and cache size.