post-processing: consolidate project-to-versions implementations

[jameslamb]

Followup to #654

And earlier version of these changes were already approved in #657 ... I'll repurpose that PR to just be about adding the "Inactive Projects" section.

This project does some post-processing to ensure that links like /stable, /latest, /legacy, and /nightly point to the correct versions. I misunderstood how that worked and so missed it in #654, resulting in some broken and incorrect links.

This consolidates duplication implementations of the same "what versions of the API docs should be built?" logic to help reduce the risk of such mistakes in the future.

Notes for Reviewers

Ugh MORE PRs?

If this works well, there should be only 1 more PR to fully restore the changes from #654 and finish rapidsai/build-planning#197.

Then 1 more with docs on all this (#664 (comment)).

Then I'll stop fiddling with this and move on 😅

How to review this

I know this looks like a lot, but I think the end state is actually simpler than the current state, even if the diff is kind of gross.

Real-world purpose of each of the files in the diff:

• ci/download_from_s3.sh = download rendered HTML docs from the S3 bucket rapids-upload-docs posts to,
• ci/update_symlinks.sh = implement paths like /stable via local symlinks, e.g. create a symbolic link _site/api/cudf/stable pointing to local directory ./_site/api/cudf/25.08
  • this is how links like https://docs.rapids.ai/api/cudf/stable/ and https://docs.rapids.ai/api/cudf/latest/ end up both pointing to 25.08 docs
• ci/customization/customize_doc.py = populate custom release selectors (dropdowns that say things like "stable (25.10)")
• ci/customization/customize_docs_in_folder.sh = figure out which files to invoke customize_doc.py for, and invoke it

On main, those files have their own (and slightly different!) implementations of "figure out which versions of docs to build for which projects". That leads to issues like those that led to #658

This proposes centralizing that in one place that's re-used by all of them:

• (new!) ci/get-projects-to-versions.sh = parse the configuration files in _data/ and determine which projects to build and what types + versions of docs to build for them

How I tested this

Check that this new script determines the expected set of projects and versions:

ci/get-projects-to-versions.sh

ci/customization/projects-to-versions.json (click me)

{
  "cucim": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "cudf": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "cudf-java": {
    "legacy": "25.06",
    "stable": "25.08"
  },
...
  "rapidsmpf": {
    "nightly": "25.10"
  },
  "rmm": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  }
}
``

<details><summary>full output (click me)</summary>

```text
Skipping: cuspatial | nightly | 25.10
Skipping: cuspatial | stable | 25.08
Skipping: cuproj | nightly | 25.10
Skipping: cuproj | stable | 25.08
Skipping: cusignal | legacy | 25.06
Skipping: cusignal | nightly | 25.10
Skipping: cusignal | stable | 25.08
Skipping: cudf-java | nightly | 25.10
Skipping: rapidsmpf | legacy | 25.06
Skipping: rapidsmpf | stable | 25.08
Skipping: libcuspatial | nightly | 25.10
Skipping: libcuspatial | stable | 25.08
Skipping: libcuproj | nightly | 25.10
Skipping: libcuproj | stable | 25.08
{
  "cucim": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "cudf": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "cudf-java": {
    "legacy": "25.06",
    "stable": "25.08"
  },
  "cugraph": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "cuml": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "cuproj": {
    "legacy": "25.06"
  },
  "cusignal": {},
  "cuspatial": {
    "legacy": "25.06"
  },
  "cuvs": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "cuxfilter": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "dask-cuda": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "dask-cudf": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "kvikio": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "libcudf": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "libcuml": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "libcuproj": {
    "legacy": "25.06"
  },
  "libcuspatial": {
    "legacy": "25.06"
  },
  "libkvikio": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "librmm": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "libucxx": {
    "legacy": "0.44",
    "nightly": "0.46",
    "stable": "0.45"
  },
  "raft": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "rapids-cmake": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  },
  "rapidsmpf": {
    "nightly": "25.10"
  },
  "rmm": {
    "legacy": "25.06",
    "nightly": "25.10",
    "stable": "25.08"
  }
}

Tested this locally end-to-end (including really downloading all the API docs from S3) following the docs added in #659

All looked good! API docs links worked, and selector dropdowns were correctly populated and their links worked. For example, for rapidsmpf (which should only have nightly docs):

and cuml (which should have nightly, stable, and legacy):

[netlify]

✅ Deploy Preview for docs-rapids-ai ready!

Name	Link
🔨 Latest commit	cca5af1
🔍 Latest deploy log	https://app.netlify.com/projects/docs-rapids-ai/deploys/689ab6a01ede8f0008468441
😎 Deploy Preview	https://deploy-preview-666--docs-rapids-ai.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[jameslamb]

This indentation change was already approved, merged, and proved working in #654 ... but it needs to show up here again because that was reverted (for other, unrelated reasons) in #658

[jameslamb]

I started the changes in this function to account for projects which don't have requested versions of all of nightly, stable, and legacy. For example, for rapidsmpfwe only hostnightly` API docs today:

docs/_data/docs.yml

Lines 170 to 179 in 5c2a36d

	rapidsmpf:
	name: RapidsMPF
	path: rapidsmpf
	desc: 'RAPIDS Multi-Process Foundation (rapidsmpf) is a collection of multi-GPU, distributed memory algorithms written in C++ and exposed to Python.'
	ghlink: https://github.com/rapidsai/rapidsmpf
	versions:
	# enable or disable links; 0 = disabled, 1 = enabled
	legacy: 0
	stable: 0
	nightly: 1

Because the version number we're comparing here is just {major}.{minor} (doesn't need to account for hotfix versions like 25.10.01), I think we can rely on always having exact matches.

And so don't need the r_versions class and its version-comparison logic.

[jameslamb]

Thanks @ajschmidt8 ! I'll merge this so we can get some feedback from a deployment.

We're getting close, I swear 😅

[jameslamb]

The deployment succeeded!

https://github.com/rapidsai/docs/actions/runs/16921632721/job/47948902232

And the API docs (https://docs.rapids.ai/api/ and everything linked from it) look right... correct versions and labels, working and correct selectors on individual projects' pages.

Almost there!