Skip to content

tfs mount_is_home: generalize the whole-tree exec-cache trigger beyond the literal java_home annotation key #548

Description

@ronaldtse

Finding

crates/tfs/src/context.rs::mount_is_home (the spec 22 §6 whole-tree
exec-cache materialization trigger) reads the LITERAL key
identity.annotations.java_home from the mounted image's manifest. The
mechanism is runtime-neutral — "the mounted tree is an interpreter home,
materialize it whole" — but the key name is java-specific.

Why it matters now

The truffleruby feedstock (tebako-packages/truffleruby,
TODO.truffleruby/02) ships a RUBY home whose only working visibility is
exec-cache (the native launcher resolves its home exe-relative and dies
materialized alone — "Loading libjli failed", rc 255; no -Xhome /
--home / TRUFFLERUBY_HOME override exists, probed 2026-09-06). To get
the whole-tree trigger it must declare annotations: {java_home: "/"}
on a ruby home — a lying L1 key, documented as a deviation in the
feedstock's manifest template.

jruby (TODO.jruby/02) and every future interpreter-home runtime hit the
same key.

Proposal

Generalize the trigger to a neutral annotation — e.g.
identity.annotations.home: "/" — keeping java_home as a back-compat
alias (already-shipped openjdk images carry it; spec 18's evolution law
applies). One match arm in mount_is_home, the schema note in
docs/spec/schemas/payload-manifest.yaml, a spec 22 §6 sentence, parity
tests.

Until this lands, interpreter-home runtimes declare java_home and
document the deviation (the truffleruby feedstock's
manifests/runtime.yaml does).

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions