[codetalcott]

Bumps the pin from Mojo 1.1.0 / max = "==26.6.0" to Mojo 1.2.0 / max = "==26.7.0" (stable channel, released 2026-10-22). pixi.toml and pixi.lock move together. No public Mojo signature changed, so @qkstat/retrieve and @qkstat/embed should compile unmodified once they are on 1.2.0.

What changed

• No hard errors on any compile surface. The one new diagnostic: x^ passed to a borrowed parameter now warns that the transfer has no effect, and every CI compile is --Werror. Four captures^ arguments to JsPromise.on_settled (which borrows captures) lose the ^. One of them was reported only by the compile-coverage target. Behaviour is unchanged, since nothing was moved before either.
• async/await → __async/__await (a warning that marks async unstable). Here they appear only in comments, which now note the new spelling.
• tests/cleanup_hook_observed.test.js has failed under the Guard Malloc recipe since 0.17.0. libgmalloc's banner, inherited by a spawned child, landed in the stderr the test asserts empty. The test now strips only those lines.
• Docs: the pin is updated in README, CONTRIBUTING and CLAUDE.md (plus a new spelling rule, and a regenerated docs/MOJO-RULES.md), and there is an Unreleased CHANGELOG entry. docs/toolchain-migrations.md gains a full 1.2.0 section.

Verified locally (darwin-arm64, Node 24.15.0)

• The addon, framework_coverage.mojo, all three examples and the FFI probe are warning-clean under --Werror.
• 850 tests and 12 GC finalizer tests pass; the full suite and async_stress (40 × 80) also pass under Guard Malloc with MALLOC_STRICT_SIZE=1 (banner confirmed).
• Codegen drift, the kitchen sink, and the codegen example (built and run) pass. The FFI probe reports 4/4, and the --Werror and keep-alive IR counterfactuals both hold.
• The docstring, compile-coverage, doc-sample, exports, rules-extract, platform, portability and GLIBCXX self-test gates pass.
• CLI, tutorial and host-mode e2e pass, including the pipeline.mojo collision, build-cache identity and -I roots.
• All 18 benchmarks are at 12–22% of ceiling. globalCacheActive() and asyncRuntimeInitOk() are both still true.

Expected CI notes

• m0-interop (non-required) will fail its gate step. No m0 on PyPI is gated on Mojo 1.2.0 yet (0.2.0 and the latest, 0.12.0, both list ["1.1.0"]). The pinned 0.2.0 source compiles on 1.2.0 and parity.mjs passes 7/7 against it. Move M0_VERSION once the re-gated m0 is published.
• Linux-only checks (glibc/GLIBCXX floor, ubuntu-latest tests) and the Deno and Bun 1.3.11 runtime gate can only be verified here. Locally, Bun 1.3.7 fails ownThread/ownThreadTeardown identically on main built with Mojo 1.1.0, so that comes from the old Bun.

🤖 Generated with Claude Code