Module repl 

ReplSession — the replay-based interactive session behind largo repl (and any future notebook surface).

Architecture: replay, not a persistent interpreter

The session accumulates source text — ## definition blocks and Main statements — and on every eval composes them into a complete program and re-runs it through interpret_for_ui_with_args: the exact engine behind largo run --interpret (VM+JIT, prelude auto-import, the optimizer, and the debug shadow oracle). Only output lines past a high-water mark are surfaced, so each eval appears incremental. This gives zero semantic divergence between the REPL and real programs by construction, and makes source a valid, runnable .lg program at every moment (:save = a program).

A failing input rolls back: the offending chunk is popped and the high-water mark stays put, so the session never wedges and never duplicates output.

The replay caveat

Re-running the whole program re-executes side effects. Deterministic programs are unaffected (their prior output is suppressed by the high-water mark), but non-deterministic operations — current time, random numbers, file or network I/O — are re-evaluated on every eval, so a binding like Let t be the current time. drifts across lines. reset is the escape hatch; a seeded replay mode is the growth path.

Structs

ReplOutcome

The outcome of one REPL evaluation.

ReplSession

A persistent imperative REPL session. See the module docs for the replay architecture.