Status: Accepted
Date: September 15, 2026
Related: ADR-298 (NK1117, the same rule for a value),
ADR-296 D17 (the emitter writes a name for a name, which is how an
unknown one would reach rustc intact), ADR-285 D14 (OFFERED,
Part I 2.2’s own set, which this reads)
fn main() {
let x: Widgit = 3
}lowered as written is let x: Widgit = 3;, and comes back as rustc’s
“cannot find type Widgit in this scope” — Part III C.1’s
class, about a file nobody wrote.
A value nothing declares is NK1117 (ADR-298) — “nothing
declares q“. Without the same for a type, C.1’s rule holds for one half of
this language’s names and not the other.
NK1135.At the name, in this language’s words, with the misspelling named.
OFFERED is Part I 2.2’s own set and already answers “does as name a type
this language has”; the ledger’s types map is every type this program and
std declare, under both the qualified key and its last segment, because a
library writes the module in front and a caller writes the suffix; and the
declaration around the name contributes its type parameters. A list maintained
beside those would be a second thing to forget, and forgetting it refuses a
correct program.
What is left is a short table of names the emitter writes that no ledger
declares — Part I 2.2’s numeric surface beyond OFFERED’s four, the collections
the prelude provides, and the hulls.
http::Response names a package’s type, and whether this build can see that
package is ADR-286 D11’s, with a message of its own. Saying
“nothing declares it” about a name a dependency does declare would be
Part III C.4 — a correct program refused — so this
walk is silent about every name with a :: in it.
dsl, views and traits.It asks about a written name rather than about a value’s type, so it needs
the item tree and neither the scope stack nor the inference. Every position a
type can stand in is read: a parameter, a result, a struct’s field, an enum
variant’s payload in both shapes, a trait method’s signature, an impl
method’s, a let’s annotation, and every argument nested inside any of those.
A refusal moves from the backend into this compiler. A specification block
or a program that names a type nothing declares was already rustc’s
“cannot find … in this scope”, about the generated file; it is now refused at
the name, in this language’s words, which is the whole of what C.1 asks for.
A type name the language does not have — List[Fortune] for Vec[Fortune] —
is caught the same way.
std does not publish. That is
spec-promises.md’s kind of question rather than a
defect in the compiler; the page is the owner’s.fn f[T: Summarize] names Summarize where a
trait stands, not where a type does, and a trait nothing declares is
ADR-295’s question rather than this one’s.as position. NK1122 already refuses a cast that leaves the eight
types (ADR-285 D14), and two refusals on one token would be two
messages about one mistake.