Beagle SCM

JS-089: back-edge require cycle — views/bro/bro.js requires core/loop.js

core/loop -> views/bro/bro -> core/loop is a require cycle broken only by a LAZY in-body require. A leaf view re-entering the whole dispatcher — driving the loop and capturing fd-1 via a monkey-patch of io.writeAll — is a layering inversion: the BRO view should consume a sub-run capability, not own one. Pairs with JS-081 (the fd-1 patch). Method: Issues.

Input

Context

• bro.js:83 const loop = require("core/loop.js"); inside a function body — the cycle is only "OK" because it's deferred.
• bro.js:84-88 then monkey-patches io.writeAll to capture fd 1 (chunks.push(b.data().slice())) around the sub-run.
• bro.js:79-82 the comment itself flags the trap: requiring the be/loop.js ENTRY shim self-runs and recurses (JAB-028) — so it must reach for the core/loop.js MODULE, deepening the inversion.

Goals

• Expose a re-entry / sub-run capability through ctx, e.g. ctx.driveSubRun(argv) returning the captured hunks, so bro consumes an INJECTED callback instead of requiring core/loop.
• Removing the cycle ALSO removes the fd-1 monkey-patch (JS-081).

Constraints

• Byte-identical be bro … output: the captured-hunks stream must match the current io.writeAll-capture path exactly.
• ctx.driveSubRun lives at the loop edge (core owns re-entry), so bro never imports core — the arrow points down only.

WIP

Design decisions

• loop.js builds ctx.driveSubRun once at dispatch and threads it into ctx; bro calls it, captures the returned hunks, no patch.
• Keep the self-run-shim guard reasoning in a comment at the seam so a future caller doesn't reintroduce the entry-shim require.

TODOs

• Repro-first: a test/js/bro case asserting captured-hunk bytes for a nested bro run, against the current monkey-patch path.
• Add ctx.driveSubRun(argv) at the loop edge; thread into ctx.
• Rewrite bro.js:83-88 to call ctx.driveSubRun; drop the require + the io.writeAll patch.

Blockers and bummers

• ← JS-081 (the fd-1 capture monkey-patch is removed together).

• 2026-07-07 audit: STILL OPEN (line refs shifted) — views/bro/bro.js:105 in-body require("core/loop.js") + the io.writeAll global patch (:106-118) persist; no ctx.driveSubRun seam exists. JSQUE-020's opts2.reentry marker mitigates the nested pager (see JS-081) but does NOT break the cycle or remove the monkey-patch.
• 2026-07-10 revision: still OPEN — bro.js:108 in-body require("core/loop.js") + io.writeAll patch :109-123 (2nd require :176); no ctx.driveSubRun seam anywhere in be/

Outcome

• (pending)