Severity: LOW Status: Fixed
On the client-side redraw path (Term.drawLine, used by full redraws), when
the last content on a line is a promoted emoji pair (base + U+FE0F with
variation-selector-always-wide on), the trailing clearToEOL was issued
from end_col + 1 — the padding cell’s column. On terminals that render
base+VS16 as one wide glyph occupying two cells (Kitty, and other hosts the
option exists for), that erase starts inside the glyph: the right half of
the emoji is blanked at every redraw.
The same geometry existed before the option landed (option-off base+VS16 cells produced the identical EL position), so this is a pre-existing defect surfaced by the feature analysis rather than a regression of the option commit itself.
drawLine skips padding cells when painting and finds the last non-padding
content column (end_col). The erase start is computed as
const next_col = end_col + char_width.charWidth(s.grid.getCell(end_col, ly).char);
For a promoted base charWidth is 1, but in grid space the content extends
through the padding cell at end_col + 1; one-past-the-last-drawn-column
is therefore end_col + 2. The formula only accounted for charWidth-based
wide pairs, not for promotion padding.
next_col, if the cell at next_col is a padding cell,
advance next_col by one so the EL begins after the promoted pair. This
matches the grid model (option on ⇒ the pair occupies two cells).src/tty/tty.zig:
"drawLine clearToEOL skips the padding cell of a promoted VS16 pair (bug #447)":
builds a line ending in a promoted pair followed by one empty column and
asserts the cursor is positioned at pair_end + 2 (not on the padding
column) before the erase.