docs(adr): ADR-0005's release-please evidence is out of date — the action and CLI take an API URL, and the provider tracker has moved #1536

New issue

Open

opened 2026-10-03 14:53:01 +02:00 by claude · 0 comments

claude commented

2026-10-03 14:53:01 +02:00

Member

Upstream issue and PR numbers below are written without a hash, so this forge does not link them to this repository's own issues.

Measured

The two statements, in docs/adr/0005-wheel-reinvention-check.md at origin/main 8086700:

:42  The [README](https://github.com/googleapis/release-please) contains no custom-API-URL config,
     no `baseUrl` / `GH_HOST` env var, no self-hosted-or-enterprise mentions, no Gitea/Forgejo/GitLab support.
:44  The only relevant tracker entry is [release-please issue #1021 "Add the ability to change git provider
     and send notifications about events"](https://github.com/googleapis/release-please/issues/1021),
     open since August 2021, labeled `needs design`. **4+ years without movement.** ...

:42, "no custom-API-URL config": literally true of the library README only

The release-please README at main (edce3d8) still has no such setting. A grep for api-url, enterprise, gitea, forgejo and gitlab finds nothing. But the section's own heading names googleapis/release-please-action, and two other surfaces do have the setting:

The action documents github-api-url and github-graphql-url inputs:
- action.yml:33-40 at github.com/googleapis/release-please-action@0b6b3fc018/action.yml, with defaults ${{ github.api_url }} and ${{ github.graphql_url }};
- the README table at lines 80-81: "Override the GitHub API URL."
The CLI has --api-url and --graphql-url (src/bin/release-please.ts:168-177 at github.com/googleapis/release-please@edce3d805e/src/bin/release-please.ts).
Both existed on the ADR's date. The last change to the action's action.yml before 2026-06-25 is 46dfc01 (2026-04-22), and it carries both inputs at :33 and :37. The CLI at 8c5e2ae (2026-04-09) carries both options at :168 and :173.

Both settings are GitHub-shaped: a REST URL plus a GraphQL URL, the GitHub Enterprise case. They are no route to Forgejo. This instance (Forgejo 15.0.7+gitea-1.22.0) answers GET /api/graphql with 404, the same as a made-up path (read 2026-10-03).

:44, "4+ years without movement": not true by the ADR's date

Upstream issue 1021 is still open, with labels type: feature request and needs design and 10 comments. Around it (GitHub API, read 2026-10-03):

Date	Upstream event
2025-07-14 to 2025-12-04	Four comments on issue 1021: 2025-07-14, 2025-11-24, and two on 2025-12-04
2025-12-04 → 2026-04-09	PR 2631, "feat: adds Provider & GitLab support", opened and then closed unmerged: https://github.com/googleapis/release-please/pull/2631
2026-04-09	PR 2729, "feat: create Scm abstraction", merged. The `Scm` interface is at `src/scm.ts:79`. https://github.com/googleapis/release-please/pull/2729
2026-06-09	PR 2770, "feat: remove dep on code-suggester", merged. On issue 1021 (2021-12-23) the maintainer had named code-suggester as the blocker. https://github.com/googleapis/release-please/pull/2770
2026-08-06	Issue 2870, "Different git forge providers possible?", opened. It is open, with 0 comments and labels `type: question`, `priority: p3`. https://github.com/googleapis/release-please/issues/2870

At edce3d8 only two classes implement Scm: GitHub (src/github.ts:136) and LocalGitHub (src/local-github.ts:68). No src/ file is a Gitea, Forgejo or GitLab provider.

The verdict is unaffected

The first event trigger (:178) is that upstream issue 1021 resolves AND a Forgejo provider ships. Neither half is met. Falsifier (A) at :193 already says that activity is not provider availability.

One more sentence to reconcile. :173 says that if release-please "ever lands provider abstraction", the calculus shifts. An abstraction has now landed (PR 2729), but without a provider other than GitHub. Read alone, the sentence suggests the calculus has shifted. :178's trigger says it has not.

Why it matters

ADR-0005 is the build-or-adopt record that the toolkit sends readers to. #1531's review leaned on it to answer "what about git-cliff?". Its §Falsifier says the evidence is verifiable by anyone. A reader who checks :42 against the action, or :44 against the tracker, finds the evidence wrong and has reason to doubt the verdict, although the verdict survives. AGENTS.md section 8 asks for claims about external systems to be verified at source and dated.

Acceptance criteria

:42 says that the library README has no endpoint setting, and that the action (github-api-url, github-graphql-url) and the CLI (--api-url, --graphql-url) do. It says both are GitHub-shaped, and links to the sources pinned at a SHA.
:44 no longer says "without movement". It names PRs 2631, 2729 and 2770 and issue 2870, with dates and URLs, and says upstream issue 1021 is still open.
:173 agrees with :178: an abstraction has landed, and the trigger needs a Forgejo provider.
The correction follows ADR-0005's own amendment form: a dated Amendment … (#N) bullet in the header, like :7 and :8, with the original wording kept as history or marked as amended in place.
The verdict and event trigger 1 are restated as unmet, citing a pinned SHA where no non-GitHub Scm implementation exists.
Every new upstream claim carries its read date (AGENTS.md section 8, lifecycle caveat). Upstream numbers are written so that this forge does not link them to local issues.

Out of scope

Re-running the wheel-reinvention evaluation. That includes the calendar backstop at :180, "re-evaluate at the next major version bump (v0.5.0 or v1.0.0)". v0.5.0 is long past, and whether that re-evaluation ran is a separate question.
The other tool entries (git-cliff, actions/forgejo-release, the semantic-release family, release-notes-assistant). They were not re-checked here.
GitHub portability for the toolkit itself.

Anchor

Found 2026-10-03 while reviewing PR #1531 / issue #1529, when release-please's commit-note behaviour was compared for the commits-as-default question. Upstream state was read from the GitHub API and from raw files at the SHAs above on the same day.

ADR-0005's release-please entry rests on two statements that no longer match upstream. Both were already inexact on the ADR's own date, 2026-06-25. The verdict still holds. The evidence for it needs correcting. *Upstream issue and PR numbers below are written without a hash, so this forge does not link them to this repository's own issues.* ## Measured **The two statements**, in `docs/adr/0005-wheel-reinvention-check.md` at origin/main `8086700`: ```text :42 The [README](https://github.com/googleapis/release-please) contains no custom-API-URL config, no `baseUrl` / `GH_HOST` env var, no self-hosted-or-enterprise mentions, no Gitea/Forgejo/GitLab support. :44 The only relevant tracker entry is [release-please issue #1021 "Add the ability to change git provider and send notifications about events"](https://github.com/googleapis/release-please/issues/1021), open since August 2021, labeled `needs design`. **4+ years without movement.** ... ``` ### :42, "no custom-API-URL config": literally true of the library README only The release-please README at main (`edce3d8`) still has no such setting. A grep for api-url, enterprise, gitea, forgejo and gitlab finds nothing. But the section's own heading names `googleapis/release-please-action`, and two other surfaces do have the setting: - **The action** documents `github-api-url` and `github-graphql-url` inputs: - `action.yml:33-40` at https://github.com/googleapis/release-please-action/blob/0b6b3fc0186a2f7118bfd88fab9ea481e1839504/action.yml, with defaults `${{ github.api_url }}` and `${{ github.graphql_url }}`; - the README table at lines 80-81: "Override the GitHub API URL." - **The CLI** has `--api-url` and `--graphql-url` (`src/bin/release-please.ts:168-177` at https://github.com/googleapis/release-please/blob/edce3d805ef3ac964d1ba2b29b0f42905f2fa412/src/bin/release-please.ts). - **Both existed on the ADR's date.** The last change to the action's `action.yml` before 2026-06-25 is `46dfc01` (2026-04-22), and it carries both inputs at :33 and :37. The CLI at `8c5e2ae` (2026-04-09) carries both options at :168 and :173. Both settings are GitHub-shaped: a REST URL plus a GraphQL URL, the GitHub Enterprise case. They are no route to Forgejo. This instance (Forgejo `15.0.7+gitea-1.22.0`) answers `GET /api/graphql` with 404, the same as a made-up path (read 2026-10-03). ### :44, "4+ years without movement": not true by the ADR's date Upstream issue 1021 is still open, with labels `type: feature request` and `needs design` and 10 comments. Around it (GitHub API, read 2026-10-03): | Date | Upstream event | |---|---| | 2025-07-14 to 2025-12-04 | Four comments on issue 1021: 2025-07-14, 2025-11-24, and two on 2025-12-04 | | 2025-12-04 → 2026-04-09 | PR 2631, "feat: adds Provider & GitLab support", opened and then closed unmerged: https://github.com/googleapis/release-please/pull/2631 | | 2026-04-09 | PR 2729, "feat: create Scm abstraction", merged. The `Scm` interface is at `src/scm.ts:79`. https://github.com/googleapis/release-please/pull/2729 | | 2026-06-09 | PR 2770, "feat: remove dep on code-suggester", merged. On issue 1021 (2021-12-23) the maintainer had named code-suggester as the blocker. https://github.com/googleapis/release-please/pull/2770 | | 2026-08-06 | Issue 2870, "Different git forge providers possible?", opened. It is open, with 0 comments and labels `type: question`, `priority: p3`. https://github.com/googleapis/release-please/issues/2870 | At `edce3d8` only two classes implement `Scm`: `GitHub` (`src/github.ts:136`) and `LocalGitHub` (`src/local-github.ts:68`). No `src/` file is a Gitea, Forgejo or GitLab provider. ### The verdict is unaffected The first event trigger (:178) is that upstream issue 1021 resolves AND a Forgejo provider ships. Neither half is met. Falsifier (A) at :193 already says that activity is not provider availability. **One more sentence to reconcile.** :173 says that if release-please "ever lands provider abstraction", the calculus shifts. An abstraction has now landed (PR 2729), but without a provider other than GitHub. Read alone, the sentence suggests the calculus has shifted. :178's trigger says it has not. ## Why it matters ADR-0005 is the build-or-adopt record that the toolkit sends readers to. #1531's review leaned on it to answer "what about git-cliff?". Its §Falsifier says the evidence is verifiable by anyone. A reader who checks :42 against the action, or :44 against the tracker, finds the evidence wrong and has reason to doubt the verdict, although the verdict survives. AGENTS.md section 8 asks for claims about external systems to be verified at source and dated. ## Acceptance criteria - [ ] :42 says that the library README has no endpoint setting, and that the action (`github-api-url`, `github-graphql-url`) and the CLI (`--api-url`, `--graphql-url`) do. It says both are GitHub-shaped, and links to the sources pinned at a SHA. - [ ] :44 no longer says "without movement". It names PRs 2631, 2729 and 2770 and issue 2870, with dates and URLs, and says upstream issue 1021 is still open. - [ ] :173 agrees with :178: an abstraction has landed, and the trigger needs a Forgejo provider. - [ ] The correction follows ADR-0005's own amendment form: a dated `Amendment … (#N)` bullet in the header, like :7 and :8, with the original wording kept as history or marked as amended in place. - [ ] The verdict and event trigger 1 are restated as unmet, citing a pinned SHA where no non-GitHub `Scm` implementation exists. - [ ] Every new upstream claim carries its read date (AGENTS.md section 8, lifecycle caveat). Upstream numbers are written so that this forge does not link them to local issues. ## Out of scope - Re-running the wheel-reinvention evaluation. That includes the calendar backstop at :180, "re-evaluate at the next major version bump (v0.5.0 or v1.0.0)". v0.5.0 is long past, and whether that re-evaluation ran is a separate question. - The other tool entries (git-cliff, actions/forgejo-release, the semantic-release family, release-notes-assistant). They were not re-checked here. - GitHub portability for the toolkit itself. ## Anchor Found 2026-10-03 while reviewing PR #1531 / issue #1529, when release-please's commit-note behaviour was compared for the commits-as-default question. Upstream state was read from the GitHub API and from raw files at the SHAs above on the same day.

claude added the

labels

2026-10-03 14:53:01 +02:00

claude referenced this issue

2026-10-03 14:55:14 +02:00

feat(prep): a release without fragments gets its notes from feat/fix commit subjects (#1529) #1531