redundant-conversion-roundtrip
Referenceconversion immediately undone by its inverse
perf · warning · stable · since 1.0 · fix: sometimes · on by default
Converting a value out of its type and immediately back does
allocation and parsing work to end where it started:
int_of_string (string_of_int n) is n.
(* bad *) int_of_string (string_of_int n)(* good *) nFires only on compositions proved identities by the stdlib’s contracts,
with both conversions resolving to their Stdlib declarations and the
inner application feeding the outer directly: int_of_string of
string_of_int, bool_of_string of string_of_bool, Char.chr of
Char.code, of_string of to_string for Int32/Int64/Nativeint,
and Bytes.to_string of Bytes.of_string. Pipeline spellings collapse
and fire. The reverse directions deliberately do not: string_of_int (int_of_string s) canonicalizes ("+3", "0x10" change),
Char.code (Char.chr n) deletes a range check, Bytes.of_string (Bytes.to_string b) must stay a fresh mutable copy, and the lossy float
pair is excluded; shadowed conversions and work between the two also
stay clean. The fix — the inner argument itself — is Safe for the
unboxed pairs (int, bool, char) and withheld for the boxed and copying
pairs, where replacement changes physical identity.

