Skip to content

Testing and Contributing

Cavanshirpro edited this page Nov 7, 2026 · 2 revisions

Testing and Contributing

The release baseline is 0.1.3; follow the roadmap for subsequent development. Changes to original material use LGPL-3.0-only and retain SPDX/dependency notices.

Local checks by change type

Change Relevant checks
Container/profile/runtime CTest, conformance vectors and schema checks
Ownership/buffers/callbacks/codecs ASan/UBSan build and affected lifecycle tests
JS/TS Native addon tests plus ESM/CommonJS TypeScript checks
Java/Kotlin JNI build and Java/Kotlin smoke tests
Python Native C ABI integration and facade tests
Package/version/license Version/license checks and actual packaged metadata/content checks
Performance Same-machine baseline/candidate experiment plus correctness regressions
ctest --test-dir build --output-on-failure
python tools/check_versions.py
python tools/check_licenses.py
python tests/schema.py
python tests/test_registry.py
npm --prefix javascript run typecheck

Native language tests need matching explicit/bundled libraries. Follow Installation and Builds rather than using an old prebuild with a newly edited facade.

What CI proves

The core workflow builds desktops, runs sanitizers/fuzzing and prepares source/ facade assets. The language workflow builds five desktop targets, executes APIs, installs platform wheels and assembles npm/Maven distributions. Packaging checks actual LGPL metadata, license texts and project sources inside archives.

0.1.3 implementation commit cca0c717309857d2223421e547b29bd4a077527e passed all thirteen jobs in its core and language runs. Its Windows Unicode regression, five Node tests, 19 Java/JNI checks, Kotlin checks and installed Python wheel all passed. This does not claim native plugin activation on Windows or untested mobile/device support.

Regression and benchmark expectations

Preserve complete hashing, signature coverage, permissions, limits, extraction policy and callback ownership when optimizing. Performance tests should not assert a machine-specific millisecond threshold. Record inputs, cache state, compiler/build mode, sample count and memory scope. The current validation benchmark retains raw results and distinguishes whole-process RSS from payload scratch buffers.

Bug reports should include versions, OS/CPU/runtime, operation/error code, expected/actual result and a minimal reproduction. See CONTRIBUTING.md, test report and Performance and Limits.

Clone this wiki locally