Previous-generation container support
The loader accepting previous-generation binaries is a test-material decision, not a product direction (see SCOPE.md) - but it needs doing, because homebrew for the current target is scarce and a loader with nothing to load cannot be debugged.
Housekeeping
Small, known, and deliberately not urgent. Recorded so they are choices rather than oversights.
memmap2is declared at workspace level and used by no crate. The other three this entry used to name -serde_json,ash,directories- all found consumers.cargo-machetedoes not flag workspace-level entries, so this one will not trip the gate on its own.- Public API with no consumer.
HandleAllocator,GuestResult,Sink,CountingSink,StubReturn::as_raw,SymbolDbFile,Registry::len,SymbolDb::len,Protection, andRegionare declared shape awaiting their phase. Listed so the count can be watched going down rather than up. Two entries once on it,SinkandCountingSink, were the whole of a crate nothing depended on, and went with it (D211). - The provenance guard still excludes
tests/fixtures/synthetic/, a path that does not exist and will not until roadmap phase 0. Harmless, and worth deleting rather than carrying if phase 0 slips much further - an exclusion for a directory nobody has seen is indistinguishable from a mistake. - CI has never executed. The workflow YAML is
unverified as CI, though every gate it runs is now a
./bin/orbistounverb that has been run locally, which is most of the risk.pages.ymlis the exception in the other direction:./bin/orbistoun sitereproduces its steps exactly. - The macOS release artifact cannot run a guest.
release.ymlbuildsaarch64-apple-darwin, whereenter_processisunimplemented!()- guest x86-64 code cannot execute natively on ARM, and principle 12 rules out an execution-backend abstraction on purpose. The analysis commands (symbols,imports,knows,questions,worklist) are genuinely useful there, so the build is worth keeping - but it is offered on the site with no caveat, andrunpanics rather than refusing.