field_set_by_name.rs sits at exactly 2000/2000 lines — split it before the next change

[proggeramlug]

#7401 brought the file back under the cap, but only just: it is now at exactly 2000 of 2000 allowed lines. The next line anyone adds turns lint — a required check — red on main again.

It was at 1990 before 2026-08-04, so it has been one small change away from this for some time. #7401 bought roughly nothing in headroom; it unblocked the gate, it did not fix the problem.

Why this matters more than a style cap

While lint is red, it is a required context that everyone bypasses. That is the gate-theatre pattern in its purest form: the check is required, it is failing, and nothing stops. A file parked at 100% of the cap guarantees a recurrence.

Suggested split

js_object_set_field_by_name is the bulk of it and has several clearly separable concerns:

• the shape-transition cache fast path
• the keys.is_null() / first-key creation path
• the overflow-map arms
• the class-static mirror interaction (eight call sites)
• the descriptor/frozen/sealed guards

Any two of those moving to a sibling module buys enough headroom that the cap stops being a tripwire.

Related

The eight refresh_roots_after_alloc!() sites came from #7381/#7383. Their rationale now lives once at the macro rather than at each call site, which is the right shape and should be preserved by any split.