[Proposal] Future improvments for code hightlight diff

[wxiaoguang]

Follows

• Improve code diff highlight, fix incorrect rendered diff result #19958

Background

Gitea's code diff is based on highlighted code at the moment. #19958 only does some improvements, there are still some bad cases for rendering if the diff breaks in a highlighted span. See the screenshot, the diff result is inconsistent.

(more hints, in the second screenshot, the color of bbbb is different, it's also caused by that the diff breaks in the span)

Possible solutions

Use "semantic change expanding" algorithm

Expand the spans with semantics to other diff sections?

Use a general token parser

More aggressively, use placeholders for all tokens (make aaaabbbb a token), then there will be no break in tokens?

Some other diff algorithms

(TODO)

Some references

• https://github.com/dandavison/delta
• Git's [diff].algorithm = patience

[silverwind]

Regarding diff algorithms, there's also histogram which is an improvement over patience. This article gives a good summary. I think we may opt to use these newer diff algorithms by default in the future.

[wxiaoguang]

I am not working on this, so close it. Feel free to take it.