feat(gust): exec-provider componentizes — WIT-type the taskdisp seam (unblocks gale#224)

[avrabe]

Unblocks gale#224 (the fused OS as one component exporting gust:os).

The blocker

Four of the five gust:os providers already componentize cleanly. exec-provider did not:

exec-provider  core 4334 B  ->  FAILS: module requires an import interface named `env`

It carried one raw import, env.poll_task, and three raw #[no_mangle] exports, and it had no WIT world. Meanwhile interface taskdisp { poll-task: func(id: u32) -> u32; } already existed and is that import — and spawn-provider and timer-provider both import taskdisp while nothing exported it. The hole sat at the centre of the composition graph, which is why gale#224 could not close.

This is a seam defect, not a design one.

The change

• wit-os/gust-os.wit: add interface exec (admit / poll-round / state) and world exec-provider { import taskdisp; export exec; }
• exec-provider: move onto the same wit_bindgen::generate! pattern the other four providers use

Verified

wasm-tools component new   -> OK, 7097 B
wasm-tools validate        -> valid component
wasm-tools component wit   -> import gust:os/taskdisp@0.1.0
                              export gust:os/exec@0.1.0

No import outside gust:os/* / gust:hal/*. build-gustos-components.sh still passes for the existing providers; gust_exec_probe still runs green under qemu (both tasks Done, hi-prio first).

No semantic drift

The executor's algorithm, data structures and numeric behaviour are untouched. The lib.rs diff deletes no code — only three comment lines change. The Guest impl forwards to the existing extern "C" bodies rather than restating the marshalling, so the two surfaces cannot diverge, and the #[no_mangle] symbols remain exported because build.rs asserts them and the probe links them.

One consequence, documented not worked around

Re-dissolving today's source flips the object's symbol contract: poll_task becomes defined (the forwarder) and poll-task becomes the undefined seam — the same shape os-ts-cm3.o already has. Whoever regenerates exec-cm3.o must rename the probe's stub to #[export_name = "poll-task"], as gust_os_ts_probe already did. Both directions were checked (current stub -> duplicate-symbol link error; renamed stub -> probe green).

The checked-in exec-cm3.o was deliberately not regenerated: that would bump it off the synth 0.42.0 build whose measurements and defect narrative RESULTS.md documents, which deserves its own reviewable change. The tree is green as committed, and the requirement is recorded in RESULTS.md and at the probe's stub so it cannot be hit silently.

Part of the v0.6.0-rc Research Day cut (M1).

🤖 Generated with Claude Code

https://claude.ai/code/session_011QG86sovTbfnPNY9SfhSmo

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!