display: end an escape sequence where the specification says it ends

[berrym]

Second slice of duplicate-path item 6, and the more interesting one: consolidating onto the shared function exposed a bug in the shared function.

What the copy did

composition_engine.c carried its own width walk beside a call to the canonical one 300 lines away in the same file. The copy:

• enumerated terminators m K J H A B C D G f s u, so a sequence ending in h or l never terminated — and \033[?25l / \033[?25h is the cursor hide/show pair a shell emits constantly. Everything after one measured as zero width.
• carried a standing TODO: Proper UTF-8 width calculation and counted one column per character.

Deleted. Both sites in that file now call the shared computation.

What auditing the terminators then found

The shared walk had the same class of error. It ended a sequence at the first letter — which looks equivalent, because most final bytes are letters. But ECMA-48 defines the final byte as 0x40–0x7E, and \033[3~ ends in ~.

So the shared function measured the remainder of such text as zero width too. Consolidating onto it without checking would have spread that case rather than fixing it.

The core now implements the specified rule: consume the introducer, and for CSI/OSC run to the first byte in 0x40–0x7E; a two-character escape ends at once. This is exactly what completion_menu_renderer already did — the correct implementation existed, in the copy nobody had consolidated onto.

Curation

No reference shell is the authority on how lush measures its own prompt, so this is not a consensus question. The question is whether lush can beat the heuristic it had — and it can, because the specification defines the answer exactly. Every hand-rolled enumeration in this tree disagreed with it in a different way, which is the argument for having one.

Verification

tests/unit/test_screen_buffer.c pins the terminators that broke each copy: SGR colour, cursor hide/show (l, h), erase-line K, clear-screen J, cursor-position H, multi-parameter SGR, and \033[3~ — the non-letter final byte the shared function itself failed until this change.

Full suite 197/197. ASan 65/65.

[codecov]

Codecov Report

❌ Patch coverage is 80.00000% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/display/screen_buffer.c	75.00%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!