[ryansolid]

Requested by Ryan · project thread

What

Before this PR, the spec covered onSettled only in its owned form, effect () { … } (EF-4). It had nothing for Solid's onFlush (solidjs/solid#4292), and W-3 and AC-1 ruled it out ("to land something now, make it a guess or a separate event"). This PR adds both as functions of the language, the way until is (AC-6): ordinary Solid calls with static checks and nothing extra at run time. Both are marked taken by default 2026-10-27.

• EF-7 onSettled(() => B), called in imperative scope, runs B once the change the calling run belongs to has committed. That is the change its writes open, or the action that adopts them (AC-1). This is Solid's unowned form.
  • In reactive scope it is IO_IN_REACTIVE; the owned form there is effect ().
  • A cleanup return is SETTLED_CLEANUP_UNOWNED at compile time.
• EF-8 onFlush(() => B) runs B as a separate run, and so a separate instant, right after the current one. It never waits on a hold, so B's writes commit on their own unless they join a hold (H-2, D-8).
  • W-3 is amended: a run still can't split itself, but onFlush schedules another run.
  • AC-1 and 07 §5 now point at onFlush for writes that should land now.
• Diagnostics: a closure with no parameters is required for both (HOOK_CALLBACK, new). Solid's ON_FLUSH_IN_OWNED_SCOPE warning becomes a static error. spec/99 is updated.

ref/ and the crate

• The interpreter (runtime.ts) and the checker (check.ts) implement EF-7 and EF-8.
• The oracle, compiled, server and types runners lower both calls to Solid's own onSettled / onFlush. On the server they throw if called, since no imperative run happens there (SV-2a).
• Scenario tables get a (flush) row, chained like (effect).
• Main (with Rust crate: front half, resolved model (CP-6), lowering through compile_program, and @reactivescript/compiler #6) is merged in, and the same checks and lowering are ported to compiler/src/check.rs and compiler/src/emit.rs.

Scenarios

• S-flush-1: a menu closes now while the next page loads. The variant without onFlush waits for the page.
• S-settled-1: Solid's save-button toast example, plus a variant where nothing is held.

Runs (ref/SOLID_NEXT 0847122)

runner	result
interpreter	99/99
oracle	99/99 (9 recorded divergences, unchanged)
compiled	99/99, and 99/99 on the crate's output (RX_CRATE=1)
ssr	92 pass, 0 fail, 7 skip
types	137/137
npm --prefix ref run crate	286/286 agree
crate -- --lower	166/166 alike

The oracle runs Solid's own functions and agrees with the spec's frames on both new scenarios.

🤖 Generated with Claude Code

https://claude.ai/code/session_0117BoWNyLJmSs7V2omSfCrh