chore: Publish docs to arrow.apache.org

[Benjamin-Philip]

What issue does this PR close?

Closes #55.

What's changed

• Docs from main are served at https://arrow.apache.org/erlang/main/
• Released documentation is served at https://arrow.hexdocs.pm/ (currently
  inactive till we release)
• Non-main paths at https://arrow.apache.org/erlang/ redirect to HexDocs
• Preview documentation on a pull request is served on the fork's GitHub pages
• Duplicated workflows have been deleted

[Copilot]

Pull request overview

This PR updates CI workflows to publish generated documentation to arrow.apache.org/erlang/ (via an asf-site branch) in addition to GitHub Pages, and reorganizes doc building/deployment responsibilities across workflows.

Changes:

• Removes the docs-generation test job from Erlang CI.
• Refactors Docs workflow into a build job plus separate deploy jobs for GitHub Pages and ASF (asf-site) with job-scoped permissions/concurrency.
• Adds an ASF deployment job that publishes the built docs to the asf-site branch.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
.github/workflows/erlang-ci.yml	Removes the standalone doc generation test job from Erlang CI.
.github/workflows/docs.yml	Splits docs build/deploy into separate jobs and adds deployment to asf-site for arrow.apache.org publishing.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Benjamin-Philip]

I forgot to mention that when we publish a release to hex.pm (Erlang and Elixir's defacto package manager), we will also publish versioned docs to hexdocs.

See https://hex.pm/docs/rebar3-publish for the exact process.

With ASF site, GitHub Pages and HexDocs, there will be 3 mirrors for documentation.

[kou]

If we want to use hex.pm as our documentation site like we did for Go ( https://arrow.apache.org/go/ redirects to pkg.go.dev ), we can use only it.

[Benjamin-Philip]

I definitely don't want to break the existing pattern, so we can use Hexpm as our primary documentation site.

However, I still want to be able to serve docs until we cut a release. Also, Hexpm will only serve docs for releases much like pkg.go.dev or docs.rs. So, I think it might be useful to have an alternate site serving docs from main?

arrow-rs is serving from main on the ASF site, while docs.rs is handling serving all the versions. Maybe we could replicate this and shutdown GH Pages?

[Benjamin-Philip]

I definitely don't want to break the existing pattern, so we can use Hexpm as our primary documentation site.

However, I still want to be able to serve docs until we cut a release. Also, Hexpm will only serve docs for releases much like pkg.go.dev or docs.rs. So, I think it might be useful to have an alternate site serving docs from main?

arrow-rs is serving from main on the ASF site, while docs.rs is handling serving all the versions. Maybe we could replicate this and shutdown GH Pages?

@kou Thoughts?

[kou]

I'm OK that we use Hexpm for released documentations but I don't want to use https://arrow.apache.org/erlang/ for documentation for the main branch because users may access to https://arrow.apache.org/erlang/ .

We should provide documentations of released versions for users. Documentations for main (not released) should be only for Apache Arrow developers.

How about setting redirect to Hexpm for https://arrow.apache.org/erlang/ and putting the documentation for main to https://arrow.apache.org/erlang/main/ or something?

shutdown GH Pages?

Could you do #68 (comment) for it?

[Benjamin-Philip]

Let's serve https://arrow.apache.org/erlang/main/ and redirect the rest to
Hexpm. I'll create an issue in apache/infrastructure-asfyaml to take down the GH
Pages site. I'll also go ahead with using GH Pages in forks as a preview.

[Benjamin-Philip]

It looks like we'll need update dtolnay/rust-toolchain to pass ASF allowlist. I'll rebase once the dependabot PR comes along.

Googling around, I read you could trigger one from the Insights > Dependency graph > Dependabot tab?

[kou]

I can't find it in the tab.

It seems that our comment (# v1) is wrong:

arrow-erlang/.github/workflows/rust-ci.yml

Line 42 in 0be5ff1

- uses: dtolnay/rust-toolchain@29eef336d9b2848a0b548edc03f92a220660cdb8 # v1

Dependabot may not work because of it... Let's update it manually in a separated PR.

It seems that we need to use # stable based on:

• https://github.com/apache/infrastructure-actions/pull/1000/changes
• dtolnay/rust-toolchain@4be7066

[Benjamin-Philip]

My bad. I had put # v1 because that was the last release.

[Benjamin-Philip]

I tested this on httpd version:

Server version: Apache/2.4.68 (Fedora Linux)
Server built:   Jun  9 2026 00:00:00

I also aliased DocumentRoot to /erlang/ in my httpd.conf to better match the ASF site's conditions:

# Map DocumentRoot to /erlang

<Location "/">
    Require all denied
</Location>

Alias "/erlang" "/home/bphilip/GitHub/arrow-erlang/doc"

<Location "/erlang">
    Require all granted
</Location>

[Benjamin-Philip]

I'm copying .htaccess here because I don't want to overwrite any configuration on HexDocs's side in-case they use Apache.

Since I'm copying .htaccess here, I decided I'll copy .asf.yaml here as well.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 5 comments.

[Benjamin-Philip]

We can track closing down the existing GitHub Pages site in a separate issue and PR, since a fix upstream in ASF infrastructure may take time.

[Benjamin-Philip]

@kou, I've created an issue upstream: apache/infrastructure-asfyaml#109.

[kou]

+1

[Benjamin-Philip]

I'm thinking even if these docs get published at arrow.apache.org, I'll only update arrow-site to reference it after our first release.