Web interface does not render composed unicode characters correctly

[ronisbr]

Description

All composed UTF-8 characters, like s̄_b, ṡ_b, etc., are not rendered correctly in Gitea.

Screenshots

See how s̄_b is being rendered. It even shows that there are a hidden unicode characters in this line.

Gitea Version

1.16.8

Can you reproduce the bug on the Gitea demo site?

Yes

Operating System

macOS

Browser Version

Safari 15.4

[wxiaoguang]

Could you provide a demo on try.gitea.io? Then people can work on it directly.

[ronisbr]

Done!

https://try.gitea.io/ronisbr/UTF8/src/branch/main/test.jl

[wxiaoguang]

Thank you. I did a quick test on GitHub https://github.com/wxiaoguang/playground/blob/main/test-utf8.jl

$ curl https://raw.githubusercontent.com/wxiaoguang/playground/main/test-utf8.jl | hexdump
0000000 b9e1 5fa1 0a62 cc73 5f84 0a62
000000c

$ curl https://try.gitea.io/ronisbr/UTF8/raw/branch/main/test.jl | hexdump
0000000 b9e1 5fa1 0a62 cc73 5f84 0a62
000000c

GitHub doesn't report the warning.

[ronisbr]

Yes, it always worked in github.

[zeripath]

Looks fine for me:

[zeripath]

GitHub doesn't report the warning.

Github is wrong to not report that there is something odd. ë and ë are not the same characters.

---

The problems stems from Safari's rendering. The code looks like:

<span class="n">s<span class="escaped-code-point" data-escaped="[U+0304]"><span class="char">̄</span></span>_b</span>

The <span class="escaped-code-point" data-escaped="[U+0304]"><span class="char"> is all inline and zero width so the overbar (U+0304) should still be being rendered over the s. Safari is incorrectly rendering this.

---

Now, how could we fix? Without having access to a Safari browser I'm not sure. Is there any way get Safari to just do the right thing with the spacing of the combining character here?

---

We do this splitting because it makes writing the escaping/unescaping extremely easy for combining marks:

gitea/modules/charset/escape.go

Lines 166 to 179 in ac88f21

	case unicode.Is(unicode.M, r):
	escaped.Escaped = true
	escaped.HasMarks = true
	if writePos < i {
	if _, err = output.Write(bs[writePos:i]); err != nil {
	escaped.HasError = true
	return
	}
	}
	if err = writeEscaped(output, r); err != nil {
	escaped.HasError = true
	return
	}
	writePos = i + size

In order to not do it we'd have to coalesce bytes that can be combined together and emit an escape block for them together eg.

<span class="n"><span class="escaped-code-point" data-escaped="s[U+0304]"><span class="char">s̄</span></span>_b</span>

That coalescing would require us to write the escaper to understand the rules for rendering of combining characters and have the state for handling these.

For example if we had: ̄ (that is [U020] + [U304]) Should that be coalesced as +[U304] or kept as [U304]? How about multiple combining characters e.g.: ē̂ e +U304 + U302? How about when/if we get round to properly dealing with ambiguous characters like с and С (which are not the same as c and C)? If these have combining characters do we coalesce or escape the ambiguous character separately?

---

If it is possible to get Safari just render the combining character in the right place that would be deeply helpful instead.

[ronisbr]

Hi @zeripath ! Thanks!

Indeed, in Chrome it works fine. It shows the warning (which should be OK) but the symbol is displayed correctly.

[zeripath]

I guess we should coalesce these.

[zeripath]

ugh it's even harder because we run this on already rendered data.

[silverwind]

If it is possible to get Safari just render the combining character in the right place that would be deeply helpful instead.

I don't think it's possible. Seems Safari's unicode composing does not work across tags, so all joined characters must be rendered as one continuous string, e.g. [char1][char2], not <span>[char1]</span>[char2].

Also, separate issue: This is another false-positive for the Bidi warning, that code seems to be way too aggressive in its detection.

[zeripath]

It's not a BIDI warning. It's a hidden character warning.

[silverwind]

So it's definitely a Safari bug: https://jsfiddle.net/9j0zc4su/

Safari rendering:

Firefox rendering:

[silverwind]

It's not a BIDI warning. It's a hidden character warning.

Okay, but there's nothing actually "hidden", is there?