Hermes Agent
ed23524817
Preserve source locations, validate malformed JSON, and add opt-in resource limits. Include executable error-guide coverage, learner pilot materials, native Windows acceptance helpers, and updated CI/documentation.
3.5 KiB
Testing Claro
Run from the repository root. Rebuild rather than trusting an old executable. The developer gate below is exercised on Linux; Windows sanitizer checks may skip and do not establish native Windows or Win98 SE acceptance.
Full developer gate
make -s all
python3 tools/validate_v1_17.py
python3 tools/validate_compiler_warnings.py
python3 tools/validate_ci_workflow.py
python3 tools/validate_documentation_status.py
validate_v1_17.py is the current full gate despite its historical filename.
It compiles current source and runs strict C89, VC6 source/build settings, golden
fixtures, built-in validation, First Hour, introductory syntax/behavior, editor
metadata, deterministic task helpers, package security, typecheck diagnostics,
reliability, version consistency and documentation-status checks.
The reliability runner includes leak-enabled ASan/UBSan learning coverage. Read each exit status and skip; a disabled sanitizer or platform skip is not a pass for that boundary. No test suite proves arbitrary-program safety.
Small changes and teaching examples
./claro --version
./claro test
./claro doctor
./claro validate
python3 tools/validate_learning_syntax.py
python3 tools/validate_first_hour.py
A successful built-in golden run ends with PASS: 0 failure(s). The learning
checkers execute input-dependent examples with fixed input and compare output;
a syntax-only check is not sufficient. CLARO_EXE can select a specific actual
executable through a launcher, including a sanitizer or Wine build.
Run ./claro check lessons/01_hello.claro for a focused syntax check. Still test
interaction and recovery with learners using the pilot checklist.
Target and archive acceptance
Use distribution verification for standard source/Linux x64/ MinGW Windows x64 archives. It rebuilds or executes extracted payloads, checks manifests, and records exact artifact results. Source builds are not evidence that a shipped binary is correct.
For VC6 / Windows 98 SE, build98.bat builds the x86 executable;
run claro98.exe test and claro98.exe validate. Modern Python developer tooling
is not a Windows 98 prerequisite. The legacy build ZIP uses its own checksum
manifest and report rather than the generic distribution validator.
Wine results must be labeled Wine results. Claim native Windows/Win98 support only after running the target tests there. Release status records the dated scope; the checklist governs publication.
New focused checks
python3 tools/validate_error_locations.py
python3 tools/validate_error_guide.py
python3 tools/validate_learner_pilot.py
python3 tools/validate_resource_guards.py
They are also part of the full gate. validate_learning_ownership.py runs them
with leak detection enabled, including negative paths. Pilot code checks are
not learner observations. Native acceptance requires
real target logs; the Wine-rejecting probe prevents silently substituting a
compatibility-layer result. Current six-track status
separates implemented work from blocked human/hosted acceptance.
Historical tools
Older validate_rc*.py and versioned validators remain for history and focused
compatibility checks. They are not alternative current release entry points.
CI coverage is checked from executable run steps, not comments. Hosted CI
success must be verified separately from these local commands.