Allow `reflow=true` to be influenced by `ignore-link-urls` setting

[macintacos]

Problem or motivation

To understand why I would like this, I think there needs to be some context.

I'm building a tool as a fun side project called caret (https://github.com/macintacos/caret), that basically facilitates presenting a plan from an agent in a web UI that a user can interact with and approve / request changes / reject from the web UI instead of the terminal.

To facilitate some features, I am using rumdl to essentially format the plan so that it reflows at natural line breaks, to help with telling the agent a specific place where a comment was inserted by the user. It's been working great so far.

There's a feature I want to add though - I basically want to update the plan presentation such that it "hides" links (but still allows them to be clickable). This basically entails hiding the [ and ] characters, and everything in side of the ().

However, if I do that, sometimes the plans that get generated just look weird, simply because I'm hiding the excess text, but that text still "exists", so rumdl needs to format it.

At first, I thought that ignore-link-urls would solve the issue, but it appears that it's only respected as a lint option, not a fmt option.

Take the following markdown:

This is a piece of [text](https://google.com/lol/alskjdhflkajshdfljkahsdfljkhasdkjfhasdkjfhasdjklfhlkajsdhfkjlashdflkjashdflkjahdskfjlhasdjfha) that should be on a single line

Running rumdl with ignore-link-urls=true and reflow=true still formats the line:

$ rumdl fmt \
      --config 'MD013.reflow=true' \
      --config 'MD013.ignore-link-urls=true' \
      .tmp/rumdl-test.md
.tmp/rumdl-test.md:1:1: [MD013] Paragraph could be normalized to use line length of 90 characters [fixed]

Fixed: Fixed 1/1 issues in 1 file (8ms)

This is a piece of
[text](https://google.com/lol/alskjdhflkajshdfljkahsdfljkhasdkjfhasdkjfhasdjklfhlkajsdhfkjlashdflkjashdflkjahdskfjlhasdjfha)
that should be on a single line

Proposed solution

It would be incredible for my use case if ignore-link-urls (or some other option) allowed me to reformat text as if the text within the parentheses and the brackets weren't there at all. Then, I think my plan view would look exactly the way I would want visually, without needing to come up with a hack on top of rumdl itself to reformat the plans nicely.

Alternatives considered

No response

Feature category

Auto-fix / formatting

Example markdown

Contribution

• I would be willing to submit a PR to implement this feature

Additional context

No response

[rvben]

Thanks for the writeup, the caret context helped.

I ran your exact line through prettier (--prose-wrap always) and mdformat (--wrap 80) at 80 columns. Both produce exactly what rumdl produces, byte for byte: the link alone on its own line. Formatters measure the source text. Overflow that can't be fixed gets left long or forgiven by the linter, not measured away.

So ignore-link-urls is a lint exemption, not a measurement rule, and reflow measures the markdown as written. docs/md013.md now says so explicitly. That was the part genuinely missing.

For caret I'd wrap in your renderer. You know which spans you're hiding; rumdl only sees [text](url).

A rendered-width mode isn't out of the question, but it'd have to be opt-in. ignore-link-urls defaults to true, so folding it into reflow would silently re-wrap documents for everyone using reflow. Happy to discuss if that would unblock you.

[macintacos]

@rvben appreciate the updates to the documentation, that helps.

I would be perfectly happy with the behavior I'm describing being a net-new opt-in option if I'm honest. If I had to do this myself I'm not even really sure how I would approach it - I guess chop up the prose at every link "boundary" and stitch it back together afterwards? The UI unfortunately "needs" the be the same as the plan that is presented, so the less hacks I have to do on the underlying plan prose to get what I want, the better off I'll be.

Totally understandable if you don't think it's a good idea to add this new option though, it is your project after all 😄

[rvben]

Let's do it. Your instinct about chopping prose at link boundaries is roughly where I landed, but there is a cleaner framing that avoids the hard part.

I am not going to add a "rendered width" mode. Defining visible width honestly means answering it for autolinks, bare URLs, image alt text, reference links, and link reference definitions, and several of those have no obviously right answer. That is a lot of new spec surface for your use case.

Instead: rumdl check already has this concept. ignore-link-urls (default true) forgives a line whose only excess is an inline link destination, and code-spans = false does the same for an unbreakable code span. reflow just doesn't consult either one, so it measures raw source while the checker measures an exempted length. So the feature is "let reflow measure lines the way check already measures them", behind one opt-in that defaults off.

That has a property I like: it cannot desync fmt from check, because both sides read the same exemption. Turn it on with ignore-link-urls = true and reflow leaves your link line long and check forgives it. Set ignore-link-urls = false and reflow wraps it and check reports it. No implied options, and it fixes the code-span case for free.

Default stays as it is, so nobody's reflowed documents move. Queued behind an open reflow correctness bug (#748); I don't want to add a second measurement path while the first one still mis-wraps nested emphasis.