Module declarations 

Project declarations: the closed vocabulary a project uses to name constructs the parser cannot see.

A \bea/\eea delimiter pair defined in a sibling .sty, an environment that behaves like align but has no built-in counterpart, a verbatim environment built by machinery no definition scan can follow — these are facts about the document that its text does not carry, and the inferred environment-alias scan (crate::semantic::define) cannot reach them (issue #109). This module is the type those facts arrive in.

It is the one sanctioned input to the parse that is not the text (AGENTS.md decision #12). What keeps that admissible is the safety property that a declaration names a spelling, never a pairing: every shape gate still runs, so a declared \bea whose \eea is unreachable demotes to a plain command exactly as an inferred one does. Config widens what is recognized and can never force a tree the text does not support, which is what makes a wrong declaration a no-op rather than a corruption.

Three shape rules, recorded here because they are what keep the vocabulary from growing into a query language:

1. Keyed by category, then name. One dedicated map per syntactic category (Declarations::environments today; commands and, if the shortverb case is ever taken, characters later), and never a scalar knob inside a name map — a category-wide switch would collide with a construct of that name, so it belongs in a sibling section.
2. like never crosses categories. It means “copy the curated built-in entry of the same kind”, and a genuinely cross-category relation gets its own key instead (EnvironmentDecl::begin/EnvironmentDecl::end, the command spellings that stand in for an environment’s delimiters).
3. Deserialization validates nothing. Every rule is checked in one later pass, Declarations::resolve, so that a failure can be reported against the key the user wrote (environments.myenv.like) rather than swallowed by a deserializer that only knows it was handed a string.

The type lives in this crate, not in the CLI, because the parse is what consumes it and because three front ends must be able to produce the same value: badness.toml, the dprint plugin’s own config (sandboxed, no filesystem), and eventually a % badness-env comment directive. Serde is a hard dependency here (the signature database is JSON), so — unlike badness-formatter’s FormatStyle — the derives need no feature gate and the CLI can deserialize straight into these types instead of maintaining a mirror that could drift. The wire spellings are therefore public API, pinned by the tests at the bottom of this file.

Structs

CommandName

A control-word name as written in a declaration, stored without the leading backslash — the spelling every signature and ParseCtx map is keyed by.

DeclarationError

A rule Declarations::resolve rejected, with the dotted key of the entry that broke it (environments.myenv.like) so the CLI can point at the line the user wrote.

Declarations

Every declaration a project makes, as authored — unresolved and unvalidated.

EnvironmentDecl

One [environments.<name>] entry: what the environment named by the key behaves like, and which command spellings stand in for its delimiters.

ResolvedDeclarations

A project’s declarations, checked and projected into signature data by Declarations::resolve.

Enums

DeclarationErrorKind

Why a declaration was rejected. Each variant is a rule from AGENTS.md decision #12 or its architecture section.

Type Aliases

EnvironmentDecls

The name-keyed [environments] map. A type alias so the CLI’s Config can name the field’s type without restating the key type.