Ligatures at the start of words break word selection

[w1ll-i-code]

Hi y'all.

I was trying out the example vello_editor and noticed some strange behavior: Most words are selected correctly on double-click, except the word finibus. In that case on double-click did not include the first letter of the word.

How to reproduce:

• write finibus into the editor. alternatively fixxxx or flxxxxx
• try selecting the text with a double-click

Here is what I have discovered so far:

• It's only a problem with that word, I tried selecting all other words in the example and it works fine.
• The problem is reproducible by typing finibus into another part of the editor and trying to select it again
• It seems to be a problem with the character sequence fi specifically, but I also noticed it on fl while testing
• This seems to only happen if the word starts with the characters, not if it's appears in or at the end of the word.
• The behavior does not appear with Fi
• Double-clicking on the f only selects that character

Tested with:

• Fedora 41, last updated on 2024-04-13
• Rust 1.86.0, both in debug and on release
• commit: d53a058

[dfrg]

Nice catch! The fi and fl sequences are common ligatures where those two characters are replaced with a single glyph and the internal cursor position is synthesized. Seems like we’re not detecting the appropriate word boundary in this case.

[dfrg]

Looking deeper, this is actually a flaw in swash where we discard cluster info for ligature components. Then in parley, we broadcast the info from the base character to all components. This effectively makes each component a word boundary when the base character is.

Since I’m working on a new shaper now, I’m disinclined to fix this in swash so I think the temporary hacky solution is to skip ligature components in parley when searching for word boundaries. This will be correct in all real world cases that come to mind (might be wonky for code font ligatures).