nagayon-935 / kuroko / 29510272491

85%

main: 83%

Build Type

Pull #2

github

Committed by nagayon-935

Commit Message

fix: model logger's line buffer as terminal columns, not bytes

- change Logger.lineBuf from []byte (indexed by byte) to []rune (indexed
  by display column, one slot per terminal cell), matching how a real
  terminal's cell grid works; wide East Asian runes now occupy two
  consecutive cells (the rune itself, then a wideContinuation
  placeholder), decoded via the new internal/textwidth package
- decode incoming PTY bytes into full UTF-8 runes before writing them
  into lineBuf (accumulating partial multi-byte sequences across both
  processLine calls and Write() boundaries via the new runeBuf field),
  instead of writing raw bytes one at a time
- add renderLine to reassemble a column-indexed cell slice back into text,
  used wherever the committed line is handed to SplitPrompt/redact/file
  output

Previously, backspace (\b) and CSI C/D/P moved the cursor by one *byte*
per unit, but readline and every real terminal move it by one *display
column* per unit. A Japanese character is 2 columns wide but 3 UTF-8
bytes, so erasing or overwriting one via the column-correct number of
backspaces (2) landed mid-encoding and wrote invalid UTF-8 into the log.
The column-indexed cell model makes lineCol actually mean what readline's
own cursor movements assume it means.

Also fixes two comparisons (`len(l.savedLine) > end`) that implicitly
assumed lineCol was a byte count equal to the committed line's byte
length; they now compare rendered byte lengths directly
(`len(l.savedLine) > len(out)`), which is correct regardless of whether
the line contains multi-byte characters.

Pull Request Pull Request #2: fix: audit-driven hardening of PTY/signal handling, viewer scrolling, and logger multi-byte support

Coverage Stats

210 of 227 new or added lines in 7 files covered. (92.51%)

3 existing lines in 2 files now uncovered.

1861 of 2195 relevant lines covered (84.78%)

92.62 hits per line