Expand description
THE EMISSION PREFIX — the mark every name this crate mints into a declaration’s own scope wears, and the seat a domain framework building these models sets to its own. The EMISSION NAMESPACE: the mark every name this crate mints into somebody else’s scope wears, held as ONE value a front end sets rather than as a constant each mint re-spells.
§What the seat is FOR, and who the second occupant is
This crate is model in, tokens out — the derive builds a model today, and ingraph-compiler
will build one from a manifest. Both of those are ingraph’s own front ends and both want
__ig. The occupant this seat exists for is the third: a DOMAIN FRAMEWORK that depends on
ingraph-codegen directly, builds the same models, and emits them under its own name —
mediagraph-codegen is the first, and it passes __mg.
Without the seat that framework’s expansions would carry ingraph’s mark on every name they put
in an author’s scope, which is wrong twice over: the names read as another crate’s, and the
FENCE that makes them safe is a fence the other framework’s derive is not holding. A namespace
is minted in and refused in by the same party or it is not a namespace at all — see
EmissionPrefix::refused, which is the half a front end spends before it ever reaches a mint.
§MIXED GRAPHS ARE LEGAL, which is why this is a value and not a feature flag
One deployment may hold raw-ingraph records beside a platform’s, and their names are per
DECLARATION rather than per build: a __ig_key__ field on one record and a __mg_key__ field
on the next are two expansions of two models, and nothing downstream is entitled to assume a
single mark. A cfg or a crate-wide constant would have made that arrangement unrepresentable;
a field on the model makes it the ordinary case.
§WHAT IS AND IS NOT INSIDE THE SEAT
The seat covers every name that lands in a SHARED namespace — the declaring module’s scope, a database catalog, a wire face, or an emitted public API. Those are the names a second framework could stand beside, so those are the names a second framework must be able to move.
It does NOT cover a name minted inside an anonymous const _: () = { … } block or bound by a
let in a generated function body. Two expansions cannot see one another’s anonymous const, so
there is nothing there for a mark to separate — __ig_says_goodbye in
crate::emit and __ig_row in crate::aggregate are spelled once, collide with nothing,
and stay as they are. The tell is the FRAME: a name in the shared namespace wears __ at both
ends, and every one of those is threaded here.
Structs§
- Emission
Prefix - The mark a framework’s emitted names are minted under —
__igforingraphitself.