Look for metadata in package-local pkgdown.yml

Migrated from https://github.com/r-lib/pkgdown/issues/1366

Previously named: Private (Enterprise Hosted) packages not autolinking due to requirement to sign-in

Hi there.

My organization hosts an instance of GH enterprise.

The Process

The current trajectory for autolinking is as follows (correct me if I'm wrong)

Read URL from package DESCRIPTION, for example:

URL: https://pages.github.sydney.edu.au/speed-extract/mandrake/

Resolve {url}/pkgdown.yml , use the links given here to help configure autolinking.

For example, DESCRIPTION$URL gives:

https://pages.github.sydney.edu.au/speed-extract/mandrake/pkgdown.yml

pandoc: '2.10'
pkgdown: 1.5.1.9000
pkgdown_sha: ~
articles: []
last_built: 2020-07-22T05:05Z
urls:
  reference: https://pages.github.sydney.edu.au/speed-extract/mandrake//reference
  article: https://pages.github.sydney.edu.au/speed-extract/mandrake//articles

urls$reference gives https://pages.github.sydney.edu.au/speed-extract/mandrake//reference
urls$article gives https://pages.github.sydney.edu.au/speed-extract/mandrake//articles

pkgdown figures out the rest.

The problem with this process and enterprise hosted repos

This would be great, however, you will find that if you click on https://pages.github.sydney.edu.au/speed-extract/mandrake/pkgdown.yml

you will be given a challenge-response asking you to sign in.

curl says:

curl -I "https://pages.github.sydney.edu.au/speed-extract/mandrake/pkgdown.yml"

HTTP/1.1 302 Moved Temporarily
Server: GitHub.com
Date: Wed, 22 Jul 2020 06:27:49 GMT
Content-Type: text/html
Content-Length: 138
Location: https://github.sydney.edu.au/sessions/_auth_request_bounce?return_to=https%3A%2F%2Fpages.github.sydney.edu.au%2Fspeed-extract%2Fmandrake%2Fpkgdown.ym

So, pkgdown is never going to resolve those links.

Potential Fixes / Accomodations

I wonder, then, whether there are other ways pkgdown could accomodate linking enterprise hosted packages. A few jump out:

1 Allowing pkgdown.yml (The one deployed wherever our pkgdown site is deployed, eg https://pages.github.sydney.edu.au/speed-extract/mandrake/pkgdown.yml ) to be built into the package (Removing it from .Rbuildignore ) s.t if it is found, its url links are resolved automatically.
2 If pkgdown attempts to resolve DESCRIPTION$URL/pkgdown.yml , and fails, pkgdown can make its best-effort based off of DESCRIPTION$URL to set up some "standard" links, such as:
```
# base_url <- DESCRIPTION$URL
urls$reference <- file.path(base_url, "reference")
urls$article <- file.path(base_url, "articles")
```

These fall-backs might allow a best-effort linking for package's whose description$url field cannot be resolved publicly. I personally don't see too much harm in building a pkgdown.yml into installed R packages, so I think this could accomodate those of us who can't host all of our work publically.

Cheers!

r-lib / downlit