Share harness passes between worktrees and cap harnesses machine-wide

A pass is keyed by the content of the harness's dependencies, so it is
also valid in any worktree with the same files. Passes are published to
parity-shared/ beside the shared vanilla data, and a worktree adopts one
whose key matches its own files instead of running the harness: an
agent's gate only runs what its own changes affect, instead of the whole
suite from scratch. Every runner on the machine also shares
PARITY_MACHINE_JOBS slots, so several agents' gates can't overload it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Ludovic Marie
2026-10-11 16:42:52 +02:00
parent bcfa2f6974
commit d061bdf2fb
3 changed files with 76 additions and 7 deletions
+3 -1
View File
@@ -22,7 +22,9 @@ Vanilla oracle runs are reused as well. `tools/parity_shim.py` stands in for `ja
Probes build into `.local/parity-target/` with release optimisation but no LTO and 16 codegen units. On ludodesktop that rebuilds every `horizon-world` probe in 3 seconds after a one-line change, against 122 seconds with the release profile. Rust gives these settings identical results (no fast-math; overflow checks as in release), and the full harness suite passes under them. Probes that harnesses build in the dev profile get `opt-level = 3` too, keeping their overflow checks and debug assertions. Harnesses must locate binaries through `cargo metadata` or `lab.cargo_target()`, never a hard-coded `target/`.
Parallel harnesses are admitted by memory: each harness's recorded peak is reserved, within three quarters of the memory available at the start. `-j` caps the count (default: a third of the CPU threads).
Passes are shared between worktrees. A harness's key depends only on the content of its dependencies, not on where they are, so every pass is also written to `parity-shared/` beside the shared vanilla data (`PARITY_SHARED` overrides it). A worktree whose files match those a harness passed with, in any worktree, counts that pass without running the harness: after merging the integration branch, an agent's gate only runs the harnesses its own changes affect.
Every runner on the machine shares `PARITY_MACHINE_JOBS` slots (default: half the CPU threads), so several agents' runs together can't overload it. Parallel harnesses are also admitted by memory: each harness's recorded peak is reserved, within three quarters of the memory available at the start. `-j` caps the count (default: a third of the CPU threads).
## The gate