[BUG] Segment._split_cells doesn't handle non-unit characters well

[rodrigogiraoserrao]

The classmethod Segment._split_cells uses a heuristic to start the search for the cut point that is too eager.
Because the algorithm implemented doesn't backtrack, one can construct example strings for which the cuts have the wrong sizes.

The two examples below show instances of the wrong result being computed.

s = Segment("🦊🦊🦊\n\n\n\n\n\n")
print(Segment._split_cells(s, 3))  # (Segment('🦊🦊🦊\n '), Segment(' \n\n\n\n'))

s = Segment("🦊🦊🦊abcdef")
print(Segment._split_cells(s, 3))  # (Segment('🦊🦊 '), Segment(' abcdef'))

[Mr-Neutr0n]

I took a look at this and opened a fix in #3974.

The root cause is that _split_cells used a heuristic (pos = int((cut / cell_length) * len(text))) to estimate the character index, then adjusted by +1/-1 with special-case handling for out_by == ±1. With mixed character widths (double-width emoji alongside zero-width or single-width chars), the estimate could land far from the actual cut point, and the incremental correction logic could misbehave.

The fix replaces this with a linear scan that accumulates cell widths character by character, so it always finds the correct split point regardless of the width distribution. The double-width-character-straddling-the-cut-point logic (replacing with two spaces) is preserved.

[TomJGooding]

It looks like this old issue might have since been resolved?

Running the example above with the latest Rich (v14.3.2) now returns a different result:

(Segment('🦊 '), Segment(' 🦊\n\n\n\n\n\n'))
(Segment('🦊 '), Segment(' 🦊abcdef'))

The segments are missing a fox, but this is the expected behaviour according to the docstring:

Note that splitting a double-width character, may result in that character turning into two spaces.

[TomJGooding]

@Manny7717 Verified AI slop comment that adds nothing to this discussion. Consider deleting.

This issue is resolved and should just be closed - if nothing else to stop the noise from AI slop.

@rodrigogiraoserrao Sorry for the ping, but would you mind closing this issue? I think it's clear Will won't be looking at this anytime soon.

[rodrigogiraoserrao]

Closed as per Tom's suggestion.

[github-actions]

I hope I helped!

Consider sponsoring my work on Rich. I give tech support for free, in addition to maintaining Rich and Textual.

If you like using Rich, you might also enjoy Textual.

Will McGugan