[yu2971512385-ui]

clipToWidth measures the same line two ways. Its escape-sequence fallback asks lipgloss.Width, and padRight right below it does too, but the main path adds up runewidth.RuneWidth rune by rune. The answers differ for anything a terminal draws as one glyph out of several runes:

glyph	runes	lipgloss.Width	sum of RuneWidth
👨‍👩‍👧‍👦 family	7	2	8
👍🏽 thumb with a skin tone	2	2	4
🏳️‍🌈 rainbow flag	4	2	3

So a diff line holding one is clipped against one measure and padded against the other, and because the cut lands on a rune boundary rather than a glyph boundary it can leave a dangling zero-width joiner at the end of the line.

Over "+ " + glyph×8 + " end" at every width from 1 to 20:

                     cuts that land inside a glyph
                     shipped   this change
family                 15/20           0/20
skin tone              10/20           0/20
flag                    8/20           0/20
rainbow flag            7/20           0/20

The change

Step a grapheme cluster at a time with uniseg, already a direct dependency, and measure each with ansi.StringWidth — which this file already imports, and which is what lipgloss.Width calls underneath, so the clip and the pad now agree.

The early exit on the width budget is kept, so a huge line is still not walked to the end.

Checks

ASCII is unaffected. Over "", "a", "hello world", "x"×200 and "+ added a line of code" at every width from 0 to 30, both versions return byte-identical results.

TestClipToWidth_KeepsGlyphsWhole is added to diff_renderer_test.go: the result stays a prefix of the line, and the cut lands on a grapheme boundary.

What I ran, and what I did not

The sandbox I work in would not run go test ./... here. I put both versions of clipToWidth into a scratch module pinned to the go-runewidth v0.0.23, charm.land/lipgloss/v2 v2.0.3 and rivo/uniseg v0.4.7 this repo locks, and ran both over the cases above; the tables are that program's output. CI is the real check.

One thing worth knowing: lipgloss.Width and runewidth.StringWidth do not always agree — under these pins runewidth calls a flag one column where lipgloss v2 calls it two. That is why the width here comes from ansi.StringWidth rather than runewidth.StringWidth: it keeps one measure across the clip, the fallback and the pad.

Disclosure: written with AI assistance (Claude). The figures above are real output.

🤖 Generated with Claude Code