Tests for MIOpen subcmd modules

[reidkwja]

Motivation

Adding tests for subcmd modules, improving test coverage for the overall MITuna source code. The coverage report identified significant gaps in four key subcommand modules:

• tuna/miopen/subcmd/export_db.py (47% coverage)
• tuna/miopen/subcmd/import_configs.py (70% coverage)
• tuna/miopen/subcmd/load_job.py (72% coverage)
• tuna/miopen/subcmd/merge_db.py (49% coverage)

These modules handle critical database export/import operations, and the missing coverage represented untested error paths, edge cases, and conditional branches.

Technical Details

The tests cover the following areas:

export_db.py (tests/test_export_db_branches.py):

• Filename generation variants (OpenCL/HIP, num_cu encoding, kernel/perf/find db extensions)
• Golden version validation requiring arch parameter
• Non-convolution config type handling in fin_net_cfg_job
• Duplicate solver entry detection in add_entry_to_solvers
• FDB entry trimming to top 4 results in build_miopen_fdb
• Kernel db extension handling (.o, .mlir.o) and -mcpu injection
• Multi-CU skew handling in build_miopen_fdb_skews and export_kdb
• PDB building and ordering

import_configs.py (tests/test_import_configs_branches.py):

• Query construction variants (tag/recurrent combinations)
• Batch list parsing and iteration
• Tag-only vs insert+tag processing paths
• IntegrityError handling in add_model/add_frameworks
• Missing framework/model ID validation in add_benchmark
• Argument validation in check_import_benchmark_args
• Multiple subcommand flag paths (print_models, add_model, add_framework, add_benchmark)

load_job.py (tests/test_load_job_branches.py):

• Algorithm-to-solver mapping in arg_solvers
• Missing tag name validation
• Config query filtering by tag and command type
• Compose query with tunable/dynamic/batch_norm filters
• Empty results error logging
• Tag error handling in run_load_job

merge_db.py (tests/test_merge_db_branches.py):

• Argument parsing requiring one of perf_db/find_db/bin_cache
• Input validation in target_merge (None/empty checks)
• Float parsing for sort keys
• copy_only short-circuit paths
• SQLite binary cache duplicate handling
• File list filtering (excluding non-gfx, DB_ALIAS entries)

All tests use mocking/monkeypatching to avoid database dependencies and test logic in isolation.

Test Plan

1. Created four new test modules under tests/:

   • test_export_db_branches.py (12 tests)
   • test_import_configs_branches.py (10 tests)
   • test_load_job_branches.py (6 tests)
   • test_merge_db_branches.py (8 tests)

2. Integrated tests into CI pipeline (vars/utils.groovy):

   • Added to pytestSuite1 with coverage tracking
   • Tests run with python3 -m coverage run -a -m pytest

3. Fixed test failures:

   • Adjusted test_build_miopen_pdb for actual deduplication behavior
   • Fixed IntegrityError construction with required SQLAlchemy parameters
   • Enhanced load_job stubs with SQLAlchemy-compatible column/table attributes
   • Updated merge_db tests to use valid filename patterns and non-aliased entries

Test Result

All 36 new tests pass successfully in CI:

-  test_export_db_branches.py: 12/12 passed
-  test_import_configs_branches.py: 10/10 passed
-  test_load_job_branches.py: 6/6 passed
-  test_merge_db_branches.py: 8/8 passed

These tests target previously uncovered lines/branches, improving coverage for the four subcommand modules.