check for broken links

[thaJeztah]

carry of #12147 - rebase and trying to tweak configuration

carries / closes#12147
fixes#13690

[netlify]

✔️ Deploy Preview for docsdocker ready!

🔨 Explore the source changes: 9400eb5

🔍 Inspect the deploy log: https://app.netlify.com/sites/docsdocker/deploys/6166ee62b347f200089e1165

😎 Browse the preview: https://deploy-preview-13514--docsdocker.netlify.app

[thaJeztah]

Trying to make it ignore the anchor links for the API (because the API reference is generated client-side with JavaScript, so it won't be able to check them).

- ./_site/engine/swarm/manage-nodes/index.html
* linking to internal hash #operation/ServiceCreate that does not exist (line 316)
<a href="/engine/api/v1.31/#operation/ServiceCreate"></a>
* linking to internal hash #operation/ServiceCreate that does not exist (line 316)

Not sure what syntax to use to configure that (docs mention it accepts a string or regex, but so far it doesn't work) 🤔

• https://github.com/gjtorikian/html-proofer/blob/d64768a0f9fb9500e13665ac10f969a6c560c360/bin/htmlproofer#L52
• https://github.com/gjtorikian/html-proofer

[thaJeztah]

We need to remove the buildx excludes once they're fixed (docker/buildx#765 is merged, but only on master)

[thaJeztah]

Pushed a commit that manually fixes the buildx broken anchors instead (still have the "distribution" / "registry" anchor that's broken; no PR for that in upstream yet)

[usha-mandya]

This is great. Thank you @thaJeztah! Does the check for broken links appear as a standalone item for subsequent PRs, or built in as part of the build checks as per the current PR?

[thaJeztah]

It will be the same, but when clicking the details, you'll see it's a separate step.

Perhaps we could make it show as a separate step on the PR 🤔

[thaJeztah]

OK, I splitter it to a separate job. The only downside is that we're now building the image twice (I didn't find a way to have "steps" show up as a separate check, or how to split so that the build-cache of the "build" would be reused); I'm sure @crazy-max would have some tricks up his sleeve (we should also be using our build-action probably); perhaps something to leave for a follow-up cleanup.

[crazy-max]

@thaJeztah I think you can avoid using another action for that and simply add htmlproofer in the Dockerfile like:

...
FROM builderbase AS htmlproofer
COPY --from=current /usr/share/nginx/html /site
RUN htmlproofer /site --disable-external \
--internal-domains="docs.docker.com,docs-stage.docker.com,localhost:4000" \
--file-ignore="/^./_site/engine/api/.*$/,./_site/registry/configuration/index.html,./_site/engine/reference/commandline/buildx_build/index.html,./_site/engine/reference/commandline/buildx_create/index.html" \
--url-ignore="/^/docker-hub/api/latest/.*$/,/^/engine/api/v.+/#.*$/,/^/glossary/.*$/"

And add the dep in your Gemfile:

gem 'html-proofer'

Will result in your action job:

validate:
name: validate linksruns-on: ubuntu-18.04steps:
- name: print docker inforun: docker version && docker info
- uses: actions/checkout@v2
- name: htmlprooferrun: docker build --target=htmlproofer

or

validate:
name: validate linksruns-on: ubuntu-18.04steps:
- uses: actions/checkout@v2
- name: htmlprooferuses: docker/build-push-action@v2with:
target: htmlprooferoutputs: type=cacheonly

This way we are sandboxed and can reuse the same logic locally.

We could also use bake to remove overwhelming build commands:

# docker-bake.hcltarget"htmlproofer" {
target="htmlproofer"output=["type=cacheonly"]
}

validate:
name: validate linksruns-on: ubuntu-18.04steps:
- uses: actions/checkout@v2
- name: htmlprooferuses: docker/bake-action@v1with:
target: htmlproofer

Use locally: docker buildx bake htmlproofer.

[thaJeztah]

Haven't come round yet to make the changes suggested above, but chatted with @usha-mandya to get this in as-is, and I'll work on making the "dockerize" changes after that.

[usha-mandya]

LGTM. Thank you @thaJeztah. You are a ⭐