yer 0.1.0

Parse Yocto error-report.txt files into ranked build findings.

yer — Yocto error-report analyzer

A command-line tool that reads Yocto/OpenEmbedded error-report.txt files, highlights the errors, failures, and anomalies in the embedded build logs, and turns them into:

• Rich terminal output for developers debugging a failed build.
• Self-contained static HTML + JSON + SARIF artifacts for CI to publish.
• Compact, token-bounded summaries designed to be fed to Claude for automated root-cause analysis and fix suggestions.
• Cross-run trends — new / recurring / regressed / fixed failures, with a --fail-on-new regression gate.

error-report.txt files are the JSON payloads produced by OpenEmbedded's report-error bbclass / send-error-report. See docs/data-format.md.

Status

✅ Implemented (v0.1.0). The full pipeline (ingest → parse → analyze → summarize → render) plus cross-run trends is built, tested against the real report corpus, and green in CI. Milestones M0–M6 are complete — see docs/roadmap.md and CHANGELOG.md.

Why

A failed Yocto build drops a large JSON report whose real content is one or more raw task logs — up to ~2 MB each. Finding the actual root cause means scrolling past thousands of lines of NOTE:/DEBUG: noise. yer extracts the handful of lines that matter, classifies the failure (compile / configure / patch / QA / fetch / dependency), deduplicates repeats, and presents a ranked list of findings — the same findings drive the terminal view, the HTML report, and the LLM summary.

Quick example

$ yer analyze error-reports/error_report_20260701180755.txt
[ERROR] gz-gui9 [do_configure] - configure (0.85)
    package "Qt5" considered NOT FOUND
    at CMakeLists.txt:86
      | -- Could NOT find Qt5QuickControls2 (missing: Qt5QuickControls2_DIR)
      | but it set Qt5_FOUND to FALSE so package "Qt5" is considered to be NOT

1 error(s), 0 warning(s) - exit 1

$ yer report error-reports/*.txt --html out/                 # static dashboard + report.json
$ yer analyze error-reports/*.txt --format sarif -o out.sarif # code-scanning annotations
$ yer summarize error-reports/err.txt | claude -p "Fix this build failure"
$ yer trend error-reports/*.txt --store .yer/trends.jsonl --record --fail-on-new  # regression gate

Documentation map

Doc	Purpose
docs/quickstart.md	Install and first run (dev + CI).
docs/ci.md	CI recipes: exit codes, SARIF upload, trend gating.
docs/releasing.md	Maintainer runbook: how to publish yer to PyPI.
docs/architecture.md	The five-stage pipeline and module layout.
docs/roadmap.md	Milestones and the development plan.
docs/data-format.md	Reference for the error-report.txt format.
docs/corpus.md	How to add to the sample corpus and derive fixtures.
docs/spec-driven-workflow.md	How specs, plans, and tasks are organized for Claude Code.
docs/specs/	Component specifications (SPEC-000…006).
tasks/	Actionable, per-milestone task lists.
CHANGELOG.md	Release history.
CLAUDE.md	Operating guide for Claude Code in this repo.

Design commitments (v1)

• Language: Python 3.11+. Core parse/analyze/models are stdlib-only; rendering adds Jinja2.
• Inputs: error-report.txt JSON files (globs, dirs, stdin).
• Outputs: terminal, canonical report.json, self-contained HTML, SARIF 2.1.0, and a local cross-run trend store.
• LLM: emits a structured summary only — no API calls in the core tool.
• CI-first: meaningful exit codes, machine-readable output, deterministic artifacts, and privacy redaction of host identity in everything shareable.

License

MIT © 2026 Rob Woolley.