mcpp pack reports its build as planning: the pack's ninja drive has no progress report

[speak-agent]

Summary

mcpp pack builds before it packs. That build reports no progress: build_and_pack drives ninja with a default BuildOptions, whose progress is unset (src/pack/pipeline.cppm, the build of each configuration group). mcpp build attaches a progress::Build report (src/build/execute.cppm).

As a result, a pack whose build compiles shows only the planning heartbeat for as long as the compile runs. There is no Compiling line and no status row.

Measurement

Release v2026.10.1.1, job "build + upload (linux / x86_64)" (run 36790174544, job 110147157268). The step runs mcpp build --target x86_64-linux-musl with the bootstrap engine, then <fresh mcpp> pack --target x86_64-linux-musl --mode static. The fresh engine's fingerprint differs, so the pack recompiles mcpp:

23:42:52  Workspace building member 'mcpp'
23:42:52  Resolving toolchain / Resolved gcc@16.1.0 -> x86_64-linux-musl
23:43:52  Planning · 1:00
...
23:48:53  Planning · 6:00
23:49:30  Packing mcpp v2026.10.1.1 (static, stripped)

The 6 min 38 s are mostly the compile. The release of 2026.9.30.2 shows the same shape: run 36702587348, Planning · 1:00 to 6:00.

Expected

The pack's build is reported as mcpp build reports one: the Compiling lines and the status row, and Finished before Packing. The dispatch pass of a --format that a package provides should be reported the same way.

[speak-agent]

Fixed in 2026.10.1.2 (#754), following .agents/docs/2026-10-01-pack-drive-and-selection-independent-compile-design.md.

The pack was one symptom of a wider gap: each command assembled its own ninja options, and of eleven construction sites one attached the report and one passed the job count. Measured on 2026.10.1.1, [build] jobs = 3 reached ninja from mcpp build only; mcpp test and mcpp pack ran ninja's default.

• The backend takes -j from the plan and reclaims stale bmi_schedule = "on" tokens before the first drive of each build directory, for every command.
• A drive is reported whenever the command opened the report; mcpp test's own per-test drives state that they write their own lines.
• mcpp pack builds every configuration first, writes Finished before the first Packing line, reports the dispatch pass of a provided --format, and fills the global dependency cache.

Verification:

• e2e 871: a cold mcpp pack writes Compiling and Finished before Packing; its JSON output stays one document; under jobs = 1, no two compiles overlap under build, test or pack. It fails on 2026.10.1.1 and passes on 2026.10.1.2 under GCC 16 and clang 22.
• The release job of v2026.10.1.2 (run 36818095279, "build + upload (linux / x86_64)"), the measurement in this report: the fresh engine's pack recompiles mcpp and now writes the package lines, Building 251/414 · 1:36 to 411/414 · 5:36, and Finished release [optimized] in 5m49s before Packing mcpp v2026.10.1.2 (static, stripped).
• GalTranslPP PR 3 with 2026.10.1.2 (run 36824209051): both mcpp pack -p ... --format release steps state their builds and their dispatch passes, and finish in 83 s together.
• The sandbox script .agents/docs/2026-10-01-pack-drive-and-selection-verify.sh: sections 4 (the pack states its build) and 5 (jobs = 1 bounds mcpp test) pass on 2026.10.1.2 in an xlings subos --sandbox with the CN mirror. On 2026.10.1.1 section 5 fails there with overlapping compiles, and section 4 fails on the host with no Compiling line.