How token rewriting behaves

The syntax-rewriting macros operate structurally on token trees rather than by editing source strings.

Groups are recursive

Every ordinary parenthesized, braced, and bracketed token group is visited recursively. A transformation therefore works at any nesting depth without assuming that its input is a Rust item, expression, type, or even valid Rust syntax yet.

Attributes are opaque

Outer and inner attributes are copied as complete opaque regions. Tokens inside #[...] and #![...] are never rewritten. This keeps derive helpers, cfg expressions, and other attribute-specific syntax under the control of the attribute that owns them.

Excluded macros are opaque

An excluded invocation is copied together with its complete delimited input. Exclusion is exact, including raw identifiers. Declarative macro definitions can also be excluded without rewriting their bodies.

Macro exclusions are shared through a private configuration envelope when several rewriting macros are chained by expand!.

What this does not promise

Token rewriting does not make arbitrary text valid Rust. The final transformed stream still has to parse and type-check normally. Diagnostics after a large transformation may also point at the macro invocation rather than at an exact character in the out-of-line source.