openharmonyinsight/openharmony-ci

OpenHarmony CI

Overview

Use the bundled script first. It resolves the latest DCP event from an OpenHarmony PR, summarizes per-job CI status, and fetches logs only for failed jobs unless explicitly told otherwise.

Permission Preflight

• Treat this workflow as network-required by default.
• Before running the bundled script for --pr, --pr-url, --event-id, artifact paths, or CI log URLs, request network permission up front.
• Mention the likely endpoints in the permission request when helpful: GitCode via oh-gc, dcp.openharmony.cn, and cidownload.openharmony.cn.
• Do not intentionally run a first attempt just to confirm that network is required. The failure is usually predictable from the workflow itself.

Primary Path

Run the bundled script:

python3 scripts/openharmony_ci.py --pr <pr_number> --repo <repo>

The script sets XDG_CACHE_HOME=/tmp/openharmony-ci-cache for oh-gc by default so it does not try to write under ~/.cache in sandboxed environments. If you already need a different cache location, set XDG_CACHE_HOME before invoking the script and the script will respect it.

Supported inputs:

• --pr <number>
• --pr-url <gitcode_pr_url>
• --event-id <dcp_event_id>

Useful flags:

• --log-mode auto Only fetch logs for failed jobs. This is the default and should remain the normal mode.
• --log-mode never Status only.
• --log-mode always Fetch logs for every job that exposes artifacts or log links.
• --json Machine-readable output.
• --download-dir <dir> Save downloaded log archives/files locally.

Expectations

• Prefer --pr or --pr-url when the task starts from a GitCode PR.
• The script reads openharmony_ci comments with oh-gc, extracts the newest DCP event id, then queries DCP APIs.
• In sandboxed or restricted environments, request network access before invoking the script because the primary path depends on external services.
• In restricted environments, prefer the bundled script over raw oh-gc calls because it already redirects the oh-gc cache to a writable temporary directory.
• Judge pass/fail from DCP event data, not labels alone.
• If there are multiple failed jobs, report all of them.
• Treat jobs with a start time but no end result as running, not failed.
• On failures, follow artifact listings to logs automatically. build.log may redirect to another URL, and build.log.zip may actually be a tar archive.
• In the final answer, report both the DCP overall result and the per-job failures. If overall=success but some jobs are marked skip or failed, say that explicitly instead of flattening it into a single verdict.
• If a failed job shows skip build or a similar reason, include that reason verbatim and avoid implying that the whole PR failed unless the DCP overall result is also failed.

Fallback

If the script cannot be used, fall back to the manual workflow:

1. Extract the DCP event id from the latest openharmony_ci build-start comment.
2. Query https://dcp.openharmony.cn/api/codecheckAccess/ci-portal/v1/event/<event_id>.
3. If a failed job has Artifacts, query https://dcp.openharmony.cn/api/dataService/ci-portal/v1/files?directoryUrl=<ArtifactsPath>.
4. Download the chosen log from https://cidownload.openharmony.cn/<path>.

Quick Examples

python3 scripts/openharmony_ci.py --pr 82764 --repo openharmony/arkui_ace_engine
python3 scripts/openharmony_ci.py --pr-url 'https://gitcode.com/openharmony/arkui_ace_engine/pull/82764' --json
python3 scripts/openharmony_ci.py --event-id 69c51ede64650f998b1d01a4 --log-mode auto
XDG_CACHE_HOME=/tmp/custom-oh-gc-cache python3 scripts/openharmony_ci.py --pr 82764 --repo openharmony/arkui_ace_engine