Skip to content

naja-scope v0.1.24

Latest

Choose a tag to compare

@github-actions github-actions released this 08 Nov 17:11
· 1 commit to main since this release
7fb6046

naja-scope 0.1.24 candidate

Requires najaeda 0.7.35 or newer. This candidate has not been published.

Content comparison

New read-only MCP/Python query get_content_hash(module=None) returns the
native SHA-256 of the top model, or a named elaborated model, including its
descendants. The response carries the hash format version and naja commit.
Hashes survive snapshot reload and reflect descendant wiring changes.
Computation is on demand, with no persistent cache or added module-card cost.

Names and IDs participate. Source metadata and primitive definitions are
excluded. Compare only within the same format version and track the engine
separately when primitive semantics matter; this is not functional equivalence
checking.

Engine upgrade

The published 0.7.35 wheel (c7ebdfc4f9f580684bf93998fcdc88e3009f2498)
adds raw timing/clock-domain queries, generated-clock handling, explicit reset
domains, and content hashing. This candidate exposes hashing. Timing/reset
analysis remains available through raw naja, pending a separately designed,
bounded naja-scope interface with explicit user constraints.

Validation

  • The complete local suite passes 155 tests, none skipped, including
    schematic integration and all three CVA6 regressions.
  • CVA6 v5.3.0 RV32 elaboration, fresh snapshot save/reload, and content-hash
    round trip pass. All three existing CVA6 regressions run against that fresh
    snapshot with unchanged assertions.
  • Complete default CVA6 RV64 and CORE-V-MCU demos pass unchanged, as do the
    bundled walkthrough and gate-level examples.
  • The raw nested dynamic packed-write reproducer and six upstream raw timing/
    reset checks pass with the published wheel.
  • CVA6 release/master and issue #3594 were rechecked on 2026-11-08. Master
    still fails only on the MMU InvalidExpression at cva6_mmu.sv:404.
    Retain v5.3.0; no RTL changes, blackboxing workaround or relaxed assertions.

See the engine validation and capability assessment
for details and remaining timing-analysis limits. Existing snapshots and user
working checkouts were preserved. No publication was performed.