Function visit_unbounded 

pub fn visit_unbounded(
    fields: &'static [Field],
    visit: &mut dyn FnMut(UnboundedField) -> ControlFlow<()>,
) -> ControlFlow<()>

Report EVERY member that makes fields unbounded, in declaration order.

first_unbounded answers “which member costs the bound” one member at a time, and that is the wrong shape for the caller that actually asks. phase-403 W0 made an unbounded type a BUILD ERROR, and a stock ROS type is routinely unbounded in several places at once — nav_msgs/Odometry has header.frame_id and child_frame_id, and sensor_msgs/PointCloud2 has four. Naming only the first turns “bound your types” into cap, rebuild, discover the next one, repeat, once per member, with a whole codegen run between each step. One build should name everything that needs a bound.

THE walk. first_unbounded is expressed on top of this rather than beside it: two walks of one schema is the shape the sizes-header mirror defect keeps taking (issues 0088 -> 0268), and “the first thing this reports” is exactly what “the first unbounded member” means, so there is nothing left for a second implementation to say.

visit returns ControlFlow so a caller that wants only the first can stop the walk rather than let it enumerate a whole type and discard all but one answer. The return value is Break iff the visitor broke.

&mut dyn FnMut rather than impl FnMut: the walk recurses, and a recursive generic function cannot name the closure type it would instantiate itself with. no_std and allocation-free either way — nothing is collected here, which is what lets an all-members form exist at all on a target with no allocator. The COLLECTING is the caller’s, and only codegen (std) does it.