From f97d6a0944537c253bf8f2db5d2cfc786508420b Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Thu, 12 Nov 2026 00:57:16 -0800 Subject: [PATCH] chore(quest): unblock pass of 2026-11-12 Rewrite the warm-gate quest for the merge queue, which replaced its quest land host steps; refresh the moq pin's and Cranelift's waits; and add an m0 quest for the merge queue's hosted-runner stall. Co-Authored-By: Claude Opus 5.5 --- quest/m0/checks/README.md | 2 +- quest/m0/checks/cache.md | 22 +++++++++++++++------- quest/m0/queue-context.md | 27 +++++++++++++++++++++++++++ quest/m1/ffi-refusals/moq-pin.md | 24 +++++++++++++++--------- quest/m3/cranelift-unwind.md | 6 +++--- 5 files changed, 61 insertions(+), 20 deletions(-) create mode 100644 quest/m0/queue-context.md diff --git a/quest/m0/checks/README.md b/quest/m0/checks/README.md index 7b4d1283b..2385edc85 100644 --- a/quest/m0/checks/README.md +++ b/quest/m0/checks/README.md @@ -2,7 +2,7 @@ ## Goal -The land gate and agents' local checks cost what moq's do: `just ci` runs only what a +The merge queue's build and agents' local checks cost what moq's do: `just ci` runs only what a change touches, builds warm, and runs the differential fixtures optimized, while a nightly `just gate` runs the full suite. Measured against the 2026-11-01 audit's baseline: about 33 CPU-hours a day in the gate (65 CPU-minutes a run) and 100 to 130 locally. diff --git a/quest/m0/checks/cache.md b/quest/m0/checks/cache.md index d11234f96..0000182f4 100644 --- a/quest/m0/checks/cache.md +++ b/quest/m0/checks/cache.md @@ -2,7 +2,7 @@ ## Goal -The land gate's `just ci` builds through mbx, as moq's does, so a run rebuilds only what +The merge queue's `just ci` builds through mbx, as moq's does, so a run rebuilds only what changed. ## Plan @@ -12,10 +12,18 @@ gate has. PR code can write to it, a risk moq already accepts; mbx keys each com its compiler, target, and inputs. Local worktrees share it too, where the dev shell already can. -Public API: none. - -## Waiting +The gate is no longer a `quest land` check: rust2 lands through GitHub's merge queue, whose +build (`.github/workflows/queue-build.yml`) runs on the host's `rust2-queue` runner. That +runner already has a persistent home and mbx's cache environment and job-boundary GC from +the host's `ci-runners.nix`, but nothing installs mbx or its Cargo shim, so the dev shell's +`mbx setup --status` finds no shim and the build is cold: a 2026-11-12 queue build logged +no `mbx` line and 379 `Compiling` lines. Install them in the queue build as moq's +`.github/actions/rust-cache` does on its `moq-queue` runner: `jdx/mr-boxington-action` at +mbx 1.29.2 or later (1.28.0 prints no real paths for files in a build script's `OUT_DIR`, +[jdx/mr-boxington#896](https://github.com/jdx/mr-boxington/issues/896)), then +`mbx setup --yes`, failing the job when no shim results, with its guard that only main's +merge queue writes the persistent store. [#426](https://github.com/moq-dev/rust2/pull/426) +already puts the shim ahead of the pinned toolchain in the dev shell. Done when a queue +build's log shows `mbx[cache]: object cache: ...`. -- Maintainer: move the land host's mbx pin from 1.28.0 to 1.29.2 or later, then rerun `mbx setup` wherever a shim exists, since a shim runs the mbx its `mbx-target` file names. Check: `mbx --version` on the land host prints 1.29.2 or later, and the shim's `mbx-target` names that store path. 1.28.0 prints no real paths for files in a build script's `OUT_DIR` ([jdx/mr-boxington#896](https://github.com/jdx/mr-boxington/issues/896)). -- Maintainer: give rust2's land instance moq's per-repository mbx profile, `id = 9`, `command = "just ci"`, `cpuQuota = 800`, `memoryHigh = "8G"`, `memorySwapMax = "1G"`, with `enable = lib.mkDefault false` until the next item. Check: the host config's `rust2` entry has these values. -- Maintainer: enable that profile once [#426](https://github.com/moq-dev/rust2/pull/426) has merged. Check: a gate run's log shows `mbx[cache]: object cache: ...`. +Public API: none. diff --git a/quest/m0/queue-context.md b/quest/m0/queue-context.md new file mode 100644 index 000000000..e90ee96a1 --- /dev/null +++ b/quest/m0/queue-context.md @@ -0,0 +1,27 @@ +# [S] The merge queue's required check waits on no GitHub-hosted runner + +## Goal + +A merge group's required context reports as soon as its self-hosted build finishes, however +busy GitHub-hosted runners are across moq-dev. + +## Plan + +`main`'s ruleset requires `Queue`, which `.github/workflows/queue.yml` reports from a job on +`ubuntu-latest` after `queue-build.yml` builds the group on the `rust2-queue` runner. Hosted +capacity is shared by every moq-dev repository, so when moq's and rind's runs fill it, the +group sits with a passed build until the queue's response window drops it. On 2026-11-12, +[#709](https://github.com/moq-dev/rust2/pull/709)'s group (run 43171064245) built in 14 +minutes and then waited for that runner for over 1h45m, while moq-dev/rind had 78 runs and +moq-dev/moq 63 queued; run 43095195569 from 2026-11-11 was still queued too. + +Make the required context come from the self-hosted build or from nothing that queues on a +hosted runner, keeping what queue.yml's comments guard: only `queue-build.yml` at `main` +reaches the `rust2-queue` runner group, a pull request still satisfies the requirement +without building, and a failed or cancelled build fails the group. Recommendation: require +the build's own context (`Queue build / Build`) instead of `Queue`, if a pull request's +skipped run of it satisfies the ruleset as the skipped `Queue` job does; changing `main`'s +ruleset is a maintainer step, so the PR gives its exact change. Check moq's and rind's queues +for the same job and report them as follow-ups for their trees. + +Public API: none. diff --git a/quest/m1/ffi-refusals/moq-pin.md b/quest/m1/ffi-refusals/moq-pin.md index f026d0e6d..4c8df9538 100644 --- a/quest/m1/ffi-refusals/moq-pin.md +++ b/quest/m1/ffi-refusals/moq-pin.md @@ -60,19 +60,25 @@ not reach by its defining path, such as `kio::Pending` and `lite::version::Versi whose defining path is itself private stays refused (rejected: (b) moq makes both reachable). This quest makes the exporter change, in `ffi/src/collect.rs`, before the bump. -Measured on 2026-11-11 at moq's `main` `d528b675`: the bump also has to fix 37 stale -`ffi/moq/ffi.toml` entries, such as excludes naming `bytes` and `kio`, which no longer -exist, and the new refusals the census lists there. +Measured on 2026-11-12 at moq's `main` `6936f0a`: `just ffi moq-cpp` fails the export with +36 errors, which the bump fixes in `ffi/moq/ffi.toml` or the exporter: 15 on the foreign +type `std::num::NonZero` (such as `Client::with_peer_hop`), 11 stale excludes naming nothing +exported (such as `bytes` and `kio`), and single refusals such as `qmux::Session`, +`CertificateDer`, a borrowing future from the non-async `Server::accept`, and borrowed +returns that are not `Clone`. ## Waiting - moq-dev/moq: `rind` builds for wasm32. moq-net depends on `rind`, whose md5 inline asm has no wasm32 arm, so `just ts moq ` fails at moq's `main`. Done when it passes there. - Still failing on 2026-11-11 at moq `d528b675`: `rind` 0.3.1, and 0.3.2 too, has an - unconditional `asm!` in `src/low/generic/md5.rs:22`, so the wasm32 build fails with E0658. - The fix belongs in `rind` itself, then a bump in moq. + The fix merged in rind's `main` ([moq-dev/rind#543](https://github.com/moq-dev/rind/pull/543)) + and ships in rind 0.4.0, whose release PR + ([moq-dev/rind#511](https://github.com/moq-dev/rind/pull/511)) the maintainer picked to merge + on 2026-11-12; then [moq-dev/moq#8432](https://github.com/moq-dev/moq/pull/8432) moves moq to + rind 0.4. On 2026-11-12 moq `6936f0a` still locks rind 0.3.1. - moq-dev/moq: moq-tokio builds without tripping `recursion_depth_exceeding_limit` (an overflow proving `Send` on `Instrumented<...>`), which `just ffi moq-cpp ` denies. - Done when it passes at moq's `main`. Still failing on 2026-11-11 at moq `d528b675`, at - `rs/moq-tokio/src/client.rs:722`, on the `Instrumented<...>` from - `tokio::spawn(panic::isolate(time::run(driver)).instrument(..))`. + Done when it passes at moq's `main`. Still failing on 2026-11-12 at moq `6936f0a`, at + `rs/moq-tokio/src/client.rs:735`, on the `Instrumented<...>` from + `tokio::spawn(panic::isolate(time::run(driver)).instrument(..))`, and in the moq-ffi facade + on `moq_net::Driver: Sync`. diff --git a/quest/m3/cranelift-unwind.md b/quest/m3/cranelift-unwind.md index 6b6e71f38..9c869947a 100644 --- a/quest/m3/cranelift-unwind.md +++ b/quest/m3/cranelift-unwind.md @@ -10,7 +10,7 @@ that ships with nightly, so moq's CI cannot build its tests with it: compile-time cost ([rustc_codegen_cranelift#1567](https://github.com/rust-lang/rustc_codegen_cranelift/issues/1567)). - It lacks the x86 intrinsic `llvm.x86.sse2.pmulhu.w`, which 31 hit - ([rustc_codegen_cranelift#171](https://github.com/rust-lang/rustc_codegen_cranelift/issues/171)). + ([rustc_codegen_cranelift#1712](https://github.com/rust-lang/rustc_codegen_cranelift/issues/1712)). It also cannot compile rind, moq's crypto, whose `naked_asm!` takes `sym` operands @@ -24,6 +24,6 @@ Cranelift, and delete this quest. ## Waiting - rust-lang/rustc_codegen_cranelift: [#1567](https://github.com/rust-lang/rustc_codegen_cranelift/issues/1567) - and [#171](https://github.com/rust-lang/rustc_codegen_cranelift/issues/171) are fixed in - the Cranelift backend nightly ships. Check the issues, or the + and [#1712](https://github.com/rust-lang/rustc_codegen_cranelift/issues/1712) are fixed in + the Cranelift backend nightly ships. Both still open on 2026-11-12. Check the issues, or the [build benchmark](/bench/build/README.md)'s moq report after a nightly pin bump.