Markdown link check runs on entire repo

[breedx-splk]

When the required build checks for a PR run, one of these checks is a markdown link checker. This is useful to help ensure that we don't amass a bunch of broken links in the docs. However, the markdown link checker looks like it runs across the entire repo. This can result in PRs failing their checks even when the broken link exists in an unrelated pice of markdown. This can cause small PRs to appear broken even when they are not, and it can extend time-to-merge.

One example of this has been #215, where a link in the http docs caused the client session.id PR to fail verifications.

Suggestions:

• Only run the markdown link check against markdown files that have changed in the PR
• Instead of running on PRs, only run on a schedule across the whole repo and open an issue when the checker fails

[AlexanderWert]

Only run the markdown link check against markdown files that have changed in the PR

What if a file A.md was moved to A-new.md in that PR, and another untouched markdown file B.md referenced the old file A.md but has not been updated in that PR? By running the check only on changed files would miss this broken link, right?

[trask]

In the Java repos, we retry failures (file-by-file) a few times to help mitigate sporadic failures: https://github.com/open-telemetry/opentelemetry-java-instrumentation/blob/main/.github/scripts/markdown-link-check-with-retry.sh

[breedx-splk]

Only run the markdown link check against markdown files that have changed in the PR

What if a file A.md was moved to A-new.md in that PR, and another untouched markdown file B.md referenced the old file A.md but has not been updated in that PR? By running the check only on changed files would miss this broken link, right?

Definitely not a use case I had considered (repo-internal breakage)....I've only ever seen it break on external outbound links 🤣 .

I have no idea how to do this, but if the tools somehow support internal/external distinction this could be useful. Always check/validate internal links but only external links in changed files. I'm sure this is an unreasonable request. 😁

[breedx-splk]

Oh, I see that the github action version (not npm) has this check-modified-files-only option.

How about this compromise (and I'm happy to PR this if that helps):

• PR checks are relaxed to only check the modified files
• Entire repo is checked nightly and issue created on failure

Thoughts?

[joaopgrassi]

I have no idea how to do this, but if the tools somehow support internal/external distinction this could be useful. Always check/validate internal links but only external links in changed files. I'm sure this is an unreasonable request.

Maybe a simple check to see if the link is a "path" or a "http link" is enough to distinguish? Run checks on "path" links for the entire repo, but only run "http checks" for the modified files?

[pyohannes]

Maybe a simple check to see if the link is a "path" or a "http link" is enough to distinguish? Run checks on "path" links for the entire repo, but only run "http checks" for the modified files?

That would be easily possible by having two separate markdown-link-check configs, adding one that ignores every link starting with http[s]:. Then duplicating the check, and running the one with the old config only on files changed in the PR.

And this would be needed in addition:

• Instead of running on PRs, only run on a schedule across the whole repo and open an issue when the checker fails