feat(pbs): add Set and reusable identity hashing

This commit is contained in:
2026-10-25 12:15:13 +00:00
parent 632ba6d0d6
commit 8105604f92
59 changed files with 4190 additions and 34 deletions
+58 -1
View File
@@ -15,7 +15,7 @@ The Java tests generate PBX in the build-pipeline module's
`build/registry-conformance`. `run.py` creates a temporary Cargo project, copies
the runtime lockfile and references its bytecode/hal/vm/system crates by path.
It never modifies the runtime checkout. Add `--online` for an uncached machine;
by default Cargo uses only cached dependencies. Java 25, Rust and the runtime
by default Cargo uses only cached dependencies. Java 21, Rust and the runtime
checkout's dependencies are required.
The output identifies runtime revision, OS/architecture and load configuration.
@@ -304,3 +304,60 @@ Evidence prints consumer/provider source hashes, selected revision and clean sta
toolchain/platform/profile, lockfile/catalog hashes, capabilities, default ceilings
and lowered refusal limits. Keep new qualification separate from historical List/
Linked evidence; prior successful providers are not relabeled as the new target.
## PBS hashing, Identity and Set
`IdentityHashingPipelineTest` qualifies neutral 0x97/0x98 production, complete
concrete/Trait groups and both optimizer paths. It also emits malformed-category,
right-TypeId, unknown-TypeId, dead-code, partial-group and truncated fixtures.
Unknown global inputs in two neutral cases reach production runtime validation:
an absent/different left cannot hide the invalid complete right view. Refusal
retains all three input slots, charges seven cycles and executes zero successful
query steps; zero budget executes nothing.
`PbsIdentityHashingTest` selects all 30 scalar/String rows from the unchanged
118-row independent provider C XXH64 oracle. Both producer paths preserve exact
signed digest bits and attempted costs. Separate source fixtures qualify Struct,
Array, Fn, Random, complete contracts, mutation, aliases and discarded scalar and
reference results. The reference-query source gate ran before Set algorithm work.
The consumer oracle copy is checked against the selected provider on every
explicit harness invocation. Ordinary Gradle compilation uses local resources.
`PbsSetTest` exercises the complete source API, contextual/default/explicit
construction, optional/default diagnostics, nominal origin, shared complete values,
content opt-in, collisions, exact growth arithmetic, callbacks, aliases and Cursor.
Counter strategies demonstrate that rehash and capture do not call hash/equals.
SDK aliases, generic-function reexports and installed examples compile through
actual workspace builders; installed compilation uses only installDist JARs in an
external directory and writes mandatory PBX fixtures for runtime execution.
```sh
./gradlew :prometeu-compiler:prometeu-build-pipeline:test --tests '*IdentityHashingPipelineTest'
./gradlew :prometeu-architecture-tests:test --tests '*PbsIdentityHashingTest' --tests '*PbsSetTest'
./gradlew :prometeu-app:installedKnowledgeTest
CARGO_TARGET_DIR=/tmp/studio-pbx-target python3 tools/pbx-conformance/run.py --runtime ../runtime --identity-hashing-only
CARGO_TARGET_DIR=/tmp/studio-pbx-target python3 tools/pbx-conformance/run.py --runtime ../runtime --set-only
```
The Set quota corpus measures old/candidate coexistence and compares every
original-object instruction prefix with a successful run. It treats guard writes
separately from structural writes. Natural quota cases cover insertion, growth,
duplicate inclusion, removal, clear and capture with no claim that all failures
preserve the entire Set or that every theoretical failure boundary occurs at the
sampled limits. Current natural runs refuse before structural mutation; this is
reported explicitly by their summaries.
A **separate test-only instrumented source fixture** inserts a controlled Array
allocation after the real candidate-table store and before size assignment. A
lowered guest quota refuses that probe. The old size six, new table capacity 16,
seven reachable nodes and active guard remain, while the refused instruction
preserves its own entry state. This demonstrates the post-publication failure
boundary; it is not a production SDK allocation, callback, algorithm or claim about
natural refusal frequency. Its ordinary/optimized fixtures are named
`pbs-set-injected-refusal-growth*` and are generated from an in-memory SDK copy.
Real safepoints qualify source Set node/candidate release, removal/clear, snapshot
prefix reclamation and sticky final retention. Set GC tests do not assume an
iteration order. Add `--release` to qualify the Rust release profile independently
of the two producer variants; omit focused flags for the complete mandatory corpus.
All harness builds remain outside the runtime checkout, which is consumed unchanged.