Fix and rewrite markup anchor processing

[lunny]

Fix #29877

[wxiaoguang]

Why sometimes encode but sometimes not?

[lunny]

Why sometimes encode but sometimes not?

014c3a2

[silverwind]

Not sure this is right. It seems odd that we decode something that was previously encoded. I think removing all the decodes may just be enough too.

I'd certainly want to add some unit tests in vitest for sure.

[wxiaoguang]

I don't know whether it is right either .... does the author or approvers have more information to write into comment?

[silverwind]

needs deeper investigation, i will do later.

[silverwind]

As per https://stackoverflow.com/a/50141030/808699 and https://jsfiddle.net/silverwind/pe0twrf8/, it seems unnecssary to encode when setting a href attribute.

Not sure about the decode yet, but my intuition tells me it's unnecessary too.

[silverwind]

Fiddle updated: https://jsfiddle.net/silverwind/pe0twrf8/7/

My conclusion is

• never encode or decode when working with getAttribute and setAttribute, just use UTF-8.
• working with a.href would require at least decoding when reading but it should be forbidden in our code.
• use CSS.escape in CSS selectors

[silverwind]

Apparently it does make a difference whether a link is generated or pre-existing in HTML:

https://jsfiddle.net/silverwind/2wy39d7e/

So backend-generated links will appear escaped and need a decodeURIComponent when reading them from DOM. JS-generated links can insert as unicode into DOM, but I guess for consistency we should also encode them when writing them to DOM.

I'll work out something on this branch.

[silverwind]

I rewrote this file quite a bit now, the code is now easier to read and there is no more complex querySelector and only a single click listener that does it all.

Everything is working and I specifically tested with chinese links as well. @lunny can you try you test case on wiki as well?

[lunny]

It works but the links are still encoded which is different from Github's implementations. For modern web browsers, we don't need to encode them.

[silverwind]

It works but the links are still encoded which is different from Github's implementations. For modern web browsers, we don't need to encode them.

Yes, but it comes out encoded from the backend, I assume from the markup renderer. So it would need to be fixed there first. If you can, raise a PR and then I will adapt the JS in the same PR to support this.