Obscurity Wiki

๐Ÿ“Ž Correction Cells

Thinking about what neighboring rules to use to detect links with [some text goes here](a-url.com) syntax requires more thinking. It's unlike the basic border-based recognition. This requires some error correction it seems.

Here is what I mean. Let's say you're a cell and your value is [ and your neighbors are [a and you already know that you're part of a paragraph. This means you are likely part of a link. You're still not 100% sure since it's hard to see what's ahead. A link's text could be hundreds of characters long.

What do you do? You assume you're a link until proven wrong.

This is where the concept of correction cells comes in. Within the condition/neighboring rule that colors the cell assuming it's part of a link, there is a counter condition that can turn the cell into a "correction cell". This would trigger a new type of overriding behaviour.

This makes me think more about general or universal cell taxonomy. This is probably what will stem from the work with experiments I'm doing here.

An optimization that could be applied here is ๐Ÿ”ฉ Fast-thinking, we can use the fast top-down scan of a whole section of markdown and get an approximation of where the link ends.