Largest source prefix worth lexing for syntax highlighting.
Above this the side is skipped and its rows render as plain text.
The prefix has to start at line 1 — the lexer needs the preceding source to know whether the changed lines sit inside a string or comment — so its size is set by how far into the file the edit is, not by how much of it is rendered. That makes this constant the bound on two separate costs.
Lexing runs synchronously in compose. Measured 2026-07 on an M-series Mac,
CPython 3.12: roughly 0.72 ms per 1,000 characters, so ~70 ms per side at this
limit, plus a one-off ~150 ms to build the lexer on the session's first diff.
Both sides are lexed back to back, so the steady-state worst case is ~140 ms of
blocked message pump and the session's first diff is ~290 ms. Re-measure before
trusting these; they set the limit but nothing enforces them.
The prefix is also retained per message by MessageData so a rehydrated diff can
re-highlight, at up to this many characters per side. Raising it slows the diff
mount and grows the transcript's memory in step, and the transcript has no cap.
