Expand description
The ENGLISH PLURAL at expansion time — the word a list face is named from when a declaration wrote none.
One entity is Reel and its page is reels, and nothing in heck knows that: casing is a
mechanical transform over the characters a name already has, and a plural is a fact about the
LANGUAGE. So the word comes from here, and the casing helpers keep their own job — this module
never changes a letter’s case, and no caller of it may pluralise by hand.
§The dependency lives on the EXPANSION side, and that is the whole of its reach
ingraph-codegen is reached only through ingraph-derive, which is a proc-macro crate — so this
crate and everything it names are built for the HOST and linked into the compiler’s macro tier,
never into a consumer’s binary. A declaring crate pays for the pluraliser once at build time and
ships not one byte of it. The dep tree is what says so rather than this sentence:
cargo tree -p ingraph -e normal reaches pluralizer only under the (proc-macro) node, and a
build with default-features = false — the derive off — has no path to it at all.
§WHICH crate, and the measurement that chose it
Three carry an English plural: pluralizer 0.5.0, cruet 1.0.0 (the maintained Inflector fork)
and inflection-rs 0.2.2. inflections is in the family by name only — it is a CASING library
with no plural in it at all, which is also why taking it would have meant a second casing
authority beside heck.
All three were run over this repository’s own entity names and over a wider English corpus. The two failure modes are not equally bad, and that is what decided it:
pluralizer 0.5 | cruet 1.0 | inflection-rs 0.2 | |
|---|---|---|---|
Splice (a corpus record) | Splices | Splice | Splice |
Slice | Slices | Slice | Slices |
Child / Life / Knife | Children / Lives / Knives | Childs / Lifes / Knifes | Children / Lives / Knives |
Person / Datum / Analysis | People / Data / Analyses | Persons / Datums / Analyses | People / Data / Analyses |
Media / Audio (mass nouns) | Media / Audio | Medias / Audios | Media / Audios |
| last release | 2025-01-17 | 2026-04-03 | 2025-05-16 |
| dependencies | once_cell, regex | regex | deunicode, lazy_static, regex |
| licence | MIT OR Apache-2.0 | BSD-2-Clause | MIT |
cruet is the more recently released of the two, and it is the one this framework cannot use: a
wrong plural that DIFFERS from the singular is silent — childs reaches a schema, no rule can
see that it is wrong, and a client reads it forever. A word the pluraliser cannot inflect comes
back UNCHANGED, which is loud: it is exactly the degenerate case the list face refuses by name,
and the declaration answers it with graphql(plural = "…"). So the crate whose failures are the
loud kind is the one that fits a design built on refusing silence — and pluralizer also has the
better irregulars by a wide margin, which is the same argument one level down.
Version 0.5 and no patch floor, per the deps law: 0.minor IS a 0.x crate’s compatibility
axis.
§What this answers, and what it deliberately does not
- Multi-word names pluralise their LAST word, and by construction rather than by a split
written here: the rules are anchored at the END of the word, so
MediaAsset→MediaAssetsandVideoChild→VideoChildrenfall out of asking the whole name. Nothing tokenises a camel-cased name, so no second casing authority appears. - Case is not this layer’s business. A few irregulars come back with a lower-case initial
(
Foot→feet,Man→men) because the crate’s table stores the replacement in lower case. That never reaches a schema: a root field is spelled with a lower initial anyway, sofeetis the field name either way. Measured, and pinned in this module’s tests. - A degenerate answer is not refused here.
Media→Media,Series→Series,Dimensions→Dimensions: the word has no distinct plural, or already is one. This function reports what English gave it; whether that is usable is the CALLER’s question, asked where the word is spent — seeRecordModel::degenerate_plural.
Functions§
- of
- The English plural of one word, as the list face’s default name is built from.