unsafe-partial-stdlib
Referencereference to a partial stdlib eliminator
restriction · warning · stable · since 1.0 · fix: never · off by default
The stdlib’s partial eliminators raise on the case their type admits
but their contract excludes: List.hd and List.tl raise Failure on
[], List.nth raises on short lists and negative indices, Option.get
raises Invalid_argument on None, and Result.get_ok/get_error
raise on the other constructor. Each has a total spelling — a match,
List.nth_opt, Option.value, or Result.get_ok' with its carried
message.
(* bad *) Option.get (find_user id)(* good *) match find_user id with Some u -> … | None -> …Why restrict this? Partiality is a legitimate tool — a raise in a
failing test is the test doing its job, and field review measured
exactly that population (261 findings on one real workspace, 246 of
them in test code, the sampled remainder contract-true). A partial
eliminator is a finding only where the house has banned the idiom in
favor of total spellings, so the rule is house policy in the
restriction tier — off even under --select all, cherry-picked by
exact name, with [@litany.expect] as the per-call justification
trail.
Fires once per reference that resolves to a listed Stdlib declaration
— saturated calls, partial applications, and first-class uses alike, and
through aliases and open. Shadowed modules, values merely named hd,
the _opt/value siblings, and the Not_found retrieval protocol
(List.find, Hashtbl.find, Sys.getenv, …) deliberately do not fire.
No fix: each remedy restructures control flow.

