[Feature] Unit test development tracking

[chenshengxin2026]

Summary

Tracking issue for unit test (UT) development across the simpler project. The goal is to build comprehensive offline-runnable unit tests for C++ runtime modules and Python utilities, enabling faster development iteration without requiring hardware.

Current Progress

Initial UT infrastructure and first batch of tests landed in PR #427 (add-unit-tests-cpp-and-python branch):

C++ Unit Tests (tests/ut/cpp/) — GoogleTest-based

Covering the a2a3/tensormap_and_ringbuffer runtime:

Test File	Module Under Test	Status
test_ring_buffer.cpp	pto_ring_buffer.h/cpp	2 intentional failures (expose known heap wrap-around & dep-pool entry-0 bugs)
test_scheduler_state.cpp	pto_scheduler.h/cpp	Pass
test_tensormap.cpp	pto_tensormap.h/cpp	Pass
test_shared_memory.cpp	pto_shared_memory.h/cpp	Pass
test_ready_queue.cpp	ready queue logic	Pass
test_core_types.cpp	core type definitions	Pass
test_submit_types.cpp	pto_submit_types.h	Pass
test_tensor.cpp	tensor.h	Pass
test_dispatch_payload.cpp	pto2_dispatch_payload.h	Pass
test_pto_types.cpp	pto_types.h	Pass

Python Unit Tests (tests/ut/) — pytest-based

Test File	Module Under Test	Status
test_elf_parser.py	python/elf_parser.py	12 tests, all pass
test_env_manager.py	python/env_manager.py	7 tests, all pass
test_task_interface.py	python/task_interface.py (nanobind bindings)	Pass
test_runtime_builder.py	python/runtime_compiler.py + examples/scripts/runtime_builder.py	Pass
test_chip_worker.py	src/common/worker/chip_worker.h/cpp	Pass

CI

• ut-cpp job added to .github/workflows/ci.yml — builds and runs C++ tests on ubuntu-latest (no hardware required)

Remaining Coverage Gaps

C++ — High Priority

• aicpu_build_graph runtime — pto_runtime2, pto_orchestrator, pto_scheduler, pto_shared_memory, pto_ring_buffer (shared source with tensormap_and_ringbuffer but different build config)
• host_build_graph runtime — runtime.h/cpp, aicore/aicpu executors
• Platform modules — memory_allocator, function_cache, host_regs, raii_scope_guard, runtime_compile_info (host-side, testable offline)
• Common modules — task_interface/task_args.h, data_type.h, arg_direction.h type correctness tests
• a5 architecture — verify tests also compile against a5 headers (shared runtime source)

Python — Medium Priority

• kernel_compiler.py — KernelCompiler class, toolchain selection, flag generation
• toolchain.py — GxxToolchain, Aarch64GxxToolchain, CCECToolchain flag/path logic
• examples/scripts/platform_info.py — architecture detection, platform discovery
• examples/scripts/run_example.py — argument parsing, config loading (unit-testable parts)
• examples/scripts/code_runner.py — execution flow logic

Infrastructure

• Python UT CI job (pytest on ubuntu-latest)
• Coverage reporting (optional)
• Fix the 2 known bug-exposing failures in test_ring_buffer (tracked separately as bug fixes)

Motivation / Use Case

Unit tests that run offline (no Ascend hardware) enable:

1. Faster CI feedback — catch regressions in seconds, not minutes
2. Safer refactoring — runtime internals (ring buffer, scheduler, shared memory) are complex and subtle
3. Bug detection — the initial batch already found 2 bugs in ring buffer logic
4. Developer onboarding — tests serve as executable documentation of module contracts

Proposed Approach

• C++ tests use GoogleTest with lightweight stubs for platform headers (no hardware dependency)
• Python tests use pytest with mocking for external dependencies (compilers, file system)
• Tests target module-level contracts, not integration behavior
• Bug-exposing tests are kept with correct expected values; they fail until the bug is fixed

Additional Context

• PR [WIP] Add: unit tests for C++ runtime modules and Python utilities #427: [WIP] Add: unit tests for C++ runtime modules and Python utilities #427
• Tests are under tests/ut/ (Python) and tests/ut/cpp/ (C++)
• C++ stubs live in tests/ut/cpp/stubs/ to isolate platform dependencies