activeinferenceinstitute/geo_infer/geo-infer-test

GEO-INFER-TEST

Instructions

Core Capabilities

• Unified test runner: run_unified_tests.py for all 45 modules
• Category filtering: unit, integration, system, performance
• Module filtering: Test individual or groups of modules
• Result reporting: JUnit XML + HTML reports per module
• Fixtures: Shared spatial test data and coordinate generators

Usage

# Run all tests
uv run python GEO-INFER-TEST/run_unified_tests.py
 
# Run specific module
uv run python GEO-INFER-TEST/run_unified_tests.py --module MATH
 
# Run by category
uv run python GEO-INFER-TEST/run_unified_tests.py --category integration
 
# Canonical runner
uv run python GEO-INFER-TEST/run_unified_tests.py --module MATH

Pytest Markers

unit, integration, system, performance, geospatial, api, slow, fast

Examples

# Spatial test fixture: generate realistic test coordinates
import numpy as np
def make_test_coordinates(n=100, center=(45.5, -122.6), spread=0.1):
    """Generate n random lat/lng pairs around a center point."""
    lats = np.random.normal(center[0], spread, n)
    lngs = np.random.normal(center[1], spread, n)
    return list(zip(lats.clip(-90, 90), lngs.clip(-180, 180)))
coords = make_test_coordinates(200)

# Property-based test example with Hypothesis
from hypothesis import given, strategies as st
@given(
    lat=st.floats(min_value=-90, max_value=90),
    lng=st.floats(min_value=-180, max_value=180),
    resolution=st.integers(min_value=0, max_value=15)
)
def test_h3_roundtrip(lat, lng, resolution):
    """The cell center re-encodes to the same cell at the same resolution."""
    import h3
    cell = h3.latlng_to_cell(lat, lng, resolution)
    rlat, rlng = h3.cell_to_latlng(cell)
    assert h3.latlng_to_cell(rlat, rlng, resolution) == cell

# Run specific test categories
uv run python GEO-INFER-TEST/run_unified_tests.py --module MATH
uv run python GEO-INFER-TEST/run_unified_tests.py --module SPACE
uv run python GEO-INFER-TEST/run_unified_tests.py --module ACT

Agent interfaces

Use uv run --frozen geo-infer catalog TEST to discover public source APIs, then geo-infer describe before calling a selected target. Scripts use geo_infer_api.agent_runtime.AgentSession; the optional stdio MCP server is geo-infer-mcp. See requests, native objects and composition. The adapter retains this module's dependencies and authorization requirements.

Guidelines

• Use real objects in tests
• Property-based tests (Hypothesis) in ≥10 modules
• Retain the runner's immutable attempt receipts, selection and JUnit files

under .geo-infer-test-results/runs/; discovery does not prove execution.

Integrations

• EXAMPLES → Example code treated as integration tests
• OPS → CI/CD test results feed monitoring
• All modules → Unified test runner spans all 45 modules

Profile actual backend workloads

Use geo-infer-profile --module geo_infer_api.agent_benchmarks --output .geo-infer-test-results/profiles/agent-run --timeout 120 for tracing. Select --python python3.15 --engine sampling only in a target environment with its dependencies and OS inspection permission. Script and MCP target is geo_infer_test.profiling:profile_command; check receipt success, source_bound and source_unchanged as well as transport ok. Retain failed runs. Read profiling controls and native evidence for sampler modes, runtime/JIT facts, artifact custody and timing interpretation. Read retained evidence through geo-infer-profile --report RECEIPT --sort-by own or geo_infer_test.profiling:summarize_profile in Python, CLI or MCP. The report verifies sealed bytes without rerunning targets; preserve historical success and source identity. Use an independently retained receipt hash through --receipt-sha256 / expected_receipt_sha256 when authenticity matters. Trusted local receipts remain required; artifact hashes alone do not authenticate a rewritten manifest. Sampling ranks weighted stacks, tracing ranks function time/calls, and overlapping cumulative times are not additive percentages. Use PerformanceMonitor.section for balanced tracer ownership and the bounded LoadTester for concurrent local workloads.

Await BenchmarkRunner.run_async or LoadTester.run_async for async functions, callable objects and wrappers returning awaitables. Measurements include awaited completion. Async load owns at most concurrency workers and drains them on cancellation; ordinary request errors count in metrics. Benchmark failures propagate. Synchronous run rejects async work instead of timing coroutine creation. Async methods do not offload blocking synchronous functions.

TestDiscoverer includes async test definitions and reports malformed files, actual file totals, ROOT categories, extra roots and cross-module imports. Discovery metadata alone does not prove tests executed; retain runner receipts and JUnit results for qualification.

Render with docxology/template

After a clean publication build, use the explicit optional template lane in scripts/render_manuscript_pdf.py --template-root PATH --template-python PYTHON --template-revision FULL_COMMIT. The separate interpreter must already have template's dependencies. The template checkout must be clean. Script, CLI and MCP target geo_infer_test.manuscript_rendering:render_template_pdf consumes the prepared combined Markdown and records template source hashes, runtime, inputs, owned-process evidence and the final PDF hash. This runs trusted local code and does not install dependencies or replace publication verification. See template rendering and cover checks.

Linux GRIB/GIS workloads

After installing the complete frozen workspace, use the explicit native I/O startup profile for Linux runtime commands and programmatic Python consumers:

uv run --frozen python GEO-INFER-TEST/native_io_profile.py -- python -c 'import xarray; print(xarray.backends.list_engines())'

The profile configures native library lookup before the child process starts. Its description is advisory; real GRIB/GeoTIFF, backend-discovery and projection tests provide the scoped interoperability checks. Retain full unit/coverage results separately. Other platforms keep their declared native profiles.