pub struct IndexBackend {
pub kind: GraphqlKind,
pub prefix: Option<LitStr>,
pub options: Option<LitStr>,
pub question: Option<LitStr>,
pub engine: Option<LitStr>,
pub first: Option<NonZeroU32>,
}Expand description
ONE backend’s block inside a derived index’s graphql(…).
§Every name here is a SKIN’s, and none of them moves a Rust type
The axis law, as the model holds it: a name is the SKIN’s fact, so it lives in the block that
names the skin — one block per backend, and a second backend renames its own schema without
touching the first’s. The Rust family is re-based by the TOP-LEVEL rename, which never appears
in here.
§THE RECORD-SIDE PLACEMENT AUDIT, and the divergence it found
The round asked for this comparison and for the divergence to be aligned or written down. It is written down, because the two families differ in the fact that decides the placement.
A RECORD’s SDL renames sit at the DOMAIN level of its own graphql(…) — rename, plural,
output(rename = …), key/filter/order_by(rename = …), connection, edge. Its BACKEND
block accepts a rename too, and that word is parsed and never read: the record’s own note says
so out loud — kept because the grammar takes it, and unread because the emitted types are ONE
per record: the SDL is identical whichever Rust library serves it.
So the record’s placement is not a competing design; it is the same design under a premise this
family does not share. A record emits one set of wire types and asserts every backend renders
them identically, which makes a domain-level name correct and a backend-level one redundant. A
derived index’s ruled posture is the opposite — one skin per word, coexisting, kept apart by the
__<backend>_ ident prefix — and under that premise a domain-level SDL name would be a name
asserted across skins that are free to differ.
ALIGNING would mean moving the record’s live renames into its backend block, which is a change to the record’s grammar and to every declaration that writes one; it is out of this round’s scope and is recorded rather than done. What this family does NOT do is copy the record’s dead word: there is no domain-level SDL rename here, so nothing in this grammar is parsed and discarded.
Fields§
§kind: GraphqlKindThe backend this block dresses the declaration for.
prefix: Option<LitStr>prefix = "…" — the BASE this backend’s whole SDL type family is built on.
§The WORD is prefix because the seat is a STEM
rename names ONE thing verbatim; prefix re-stems a FAMILY. This seat names no type
literally: the emitted types are <base>FuseOptions, <base>FusedEvidence and their siblings,
so what a declaration writes here is composed with a suffix and is upper-camel-cased at the type
positions where GraphQL’s own convention is. There is no schema name equal to what an author
types at this position, which is the whole of why rename was the wrong word for it. A word
that names ONE type outright is a per-face override below, and those are carried verbatim under
their own rename rows.
rename at this position is refused BY NAME, pointing here — see ingraph-derive’s
index_backend, where the correction and its did-you-mean are written.
A LitStr and never an ident, on the 2026-08-13 ruling: this word names types in GraphQL’s
own namespace and nothing in Rust, so a Rust keyword is an ordinary word here.
options: Option<LitStr>options(rename = "…") — what this backend calls the package a client CONFIGURES a retrieve
with, whichever of the binding table’s shapes that is.
The FACE and not every options package: a tier’s Fusable::Options is the one a parent and a
client name, and the per-slot packages inside it are interior — re-based by the base above and
named by nothing here.
question: Option<LitStr>question(rename = "…") — the sister tree’s face, by the same table.
engine: Option<LitStr>engine(rename = "…") — what this backend calls the @oneOf selector over the typed engines.
first: Option<NonZeroU32>first = N — the CONTAINER tier’s SDL default for every query argument this skin publishes,
and the GraphQL-term spelling of the same number Numbers::limit carries in Rust vocabulary.
The bones/skin split ruled in Q6a: a container names limit and depth because those are the
words a Rust door is called with, and it names first HERE because that is the word a schema
publishes. Unwritten, the chain falls to the bones beside it — see
wire_first, which is the one place the whole order is spelled.
Implementations§
Source§impl IndexBackend
impl IndexBackend
Sourcepub const fn of(kind: GraphqlKind) -> Self
pub const fn of(kind: GraphqlKind) -> Self
A backend named and nothing else said about it — the graphql(async_graphql) spelling.