Fix _include/media-url.html logic

[dolfs]

Type of change

• Bug fix (non-breaking change which fixes an issue)

Description

When baseurl in _config.yml is not empty, without this fix, it can be prepended more than once, which is incorrect.

This fix adds logic to prevent prepending the baseurl when it is already a prefix in the url, but correctly handles cases where it is a prefix, but the url does not follow this prefix with a path/url separator.

For example, when baseurl is set to /blog, but the url starts with /blogger it is not considered "prefixed" and /blog/ is prepended (resulting in /blog/blogger…, but if the url starts with /blog/it is considered "prefixed" and is not prepended again (keeping/resulting in/blog/…`)

A limitation is that if the user supplied a (partial) URL starting with /blog/ and truly intended to produce /blog/blog/, it won't work.

Additional context

Fixes #2639, and:

• Prevents occasional introduction of white space
• Check for pre-existing absolute url is made more robust
• Better handling of scenarios where site.baseurl ends in '/', or accidentally was set to just '/'

[cotes2020]

Hi there, thanks for the PR and for your time!

From the first review, I’ve identified two issues:

1. remove_last is a feature of Liquid 5.2.0, but Jekyll currently only supports Liquid 4.x.
2. There is a bug in this PR: when site.cdn is not empty, the output media URL results in /<baseurl><cdn>/<subpath>.

[dolfs]

Changes made.

[cotes2020]

Changes made.

Where is it? Did you forget to push?

[dolfs]

Oops, it should be there no. I am not very experienced with PRs.

[cotes2020]

After looking deeper into the related issue #2639, I don't think this change/PR is necessary. media-url.html is a snippet that is explicitly called, and whether the src parameter includes the baseurl is something the caller can control. Adding logic to detect the baseurl inside the snippet would not only overcomplicate it but, as you mentioned, we can't be certain if the src path just happens to match the baseurl.

A limitation is that if the user supplied a (partial) URL starting with /blog/ and truly intended to produce /blog/blog/, it won't work.

[cotes2020]

I've come up with a workaround: when using the link tag or other third-party filters, we can append a specific class (e.g., .remain) to the image. This acts as a flag to tell the media-url filter to skip any modifications.

Change media-url.html to:

{%- assign remain = include.remain | default: false -%}

{%- if url -%}
  {%- unless url contains '://' or remain -%}
    {%- comment -%}
      Handle src URL using existing logic.
      ...
    {%- endcomment -%}
  {%- endunless -%}

{{- url -}}

Usage:

![alt](/path/to/img)  <!-- output: <img src='[site.cdn]/[site.baseurl]/path/to/img'> -->

![alt]({% link /path/to/img %}){: .remain } <!-- output: <img src='/[site.baseurl]/path/to/img'> -->