Skip to content
Merged
65 changes: 61 additions & 4 deletions docs/rfcs/0028-build-feedback-program.md
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
---
rfc: 0028
title: Build-feedback program — test-harness consolidation and workspace decomposition
status: drafted
status: specified
author: Jens Holdgaard Pedersen <jens@holdgaard.org>
drafting-assistance: Claude
created: 2026-07-06
Expand Down Expand Up @@ -129,12 +129,69 @@ runner-agnostic). CI keeps the exact same suite inventory.

## 5. Acceptance criteria

Written at the `specified` gate (docs/rfcs/README.md lifecycle).
Proposed scenarios accompany the drafting PR for sign-off.
Scenario ids `RFC0028.<m>`. Maintainer sign-off: 2026-07-06 (the
proposed scenarios accompanied the drafting PR, #383).

> **Scenario RFC0028.1 — consolidation preserves the test
> inventory.** Given the pre-consolidation
> `cargo test -p <crate> -- --list` inventory, When the crate's
> harness consolidation lands, Then the post-consolidation inventory
> lists the same set of tests, differing only by the harness's
> module-path prefix (`--list` prints test names as module paths; a
> file moving under `tests/it/` gains its module segment), And no
> test body changed in the move.

> **Scenario RFC0028.2 — isolation-sensitive tests stay isolated.**
> Given the slice-1 inventory of tests requiring process isolation
> (process-global installers, env-mutating, hardware-gated), Then
> those tests are not merged into a shared harness — they stay in
> dedicated integration-test binaries (grouped where they can safely
> share one), each annotated with the reason it cannot join the
> harness.

> **Scenario RFC0028.3 — the probe set improves.** Given the epic
> #382 probe set re-run after slices 1–2 — on the same machine and
> under the same conditions the baseline was captured (warm
> workspace, same toolchain; environment recorded next to the
> numbers in the epic) — Then the incremental-edit probe —
> `touch crates/ourios-core/src/lib.rs` (an mtime-only update,
> exactly as the epic's baseline measured it) followed by
> `cargo test -p ourios-querier --no-run` — drops below 30 s, And
> full-workspace `cargo test` wall time drops by at least 30%
> against the epic's baseline.

> **Scenario RFC0028.4 — the core split is behavior-free.** Given
> the `ourios-config` extraction, When the full workspace suite
> runs, Then results are identical pre/post split, And a
> `MinerConfig` edit no longer rechecks type-only core consumers.

> **Scenario RFC0028.5 — CI parity.** Given the consolidated
> harnesses (and nextest, if slice 5 adopts it), Then CI runs the
> identical suite inventory and stays green.

## 6. Testing strategy

Follows §5 at the `specified` gate.
Inventory diffs are the mechanism for RFC0028.1/RFC0028.5's name
half, and the PR diff is the mechanism for its no-body-change half:
a consolidation PR is restricted to file moves plus the mechanical
harness scaffolding (`tests/it/main.rs` `mod` lines, import-path
adjustments); the reviewer rejects any hunk inside a test function
body. For RFC0028.1's inventory, a
`cargo test -p <crate> -- --list` snapshot (scoped to the crate
being consolidated, matching RFC0028.1) is captured in each
consolidation
PR's description and diffed against the post-move run — the
reviewer checks the diff is a pure path-prefix rename. RFC0028.2 is
a committed list (the harness-exempt binaries and their reasons, in
the consolidating crate's `tests/` README or module docs).
RFC0028.3's probe numbers are recorded in epic #382 alongside the
baseline so the before/after is one table. RFC0028.4 is the full
suite run plus a
recheck-set spot check: a whitespace-only edit inside the
`MinerConfig` definition (`crates/ourios-core/src/config.rs` today;
its new home after the split), then `cargo build -vv` on a
type-only core consumer, asserting the build reports the consumer
`Fresh` (no `Compiling`/`Dirty` line for it).

## 7. Open questions

Expand Down