Test mixing direct api_key and ${VAR} in same config

Test sanitization with mix of valid and invalid values

Test that MLIR file exists and is valid

Test that the MLIR file exists and is readable

Test that LLMModelConfig temperature is None by default

Test that LLMModelConfig top_p is None by default

Test that models inherit temperature/top_p from parent config

Test database with multiple islands

Multi-line changes should show actual SEARCH/REPLACE content

Multiple diff blocks should be numbered

Test that children inherit their parent's island at time of creation

Test that multiple migration rounds don't create exponential program growth

Test that retries work correctly with actual errors (not timeouts)

Test that calling run() multiple times doesn't pollute the database

Test that negative patience is allowed (but probably shouldn't be used)

Nested relative paths should resolve correctly

Test no validation when cascade evaluation is disabled

Test that no program ID appears in multiple islands simultaneously

Test that no programs with _migrant suffixes are created

Test that migration is skipped with single island

Test that programs are never assigned to multiple islands

Test that initial program is not added when database already has programs at iteration 0

Test that non-OpenAI API bases don't trigger reasoning model logic

Test that when novelty checking is disabled, adding programs from async context works fine (this is the workaround from issue #313).

Test that novelty checking works correctly when called from within an async context (this was the bug in issue #313). Expect

Test that novelty checking also works correctly when called from a synchronous (non-async) context. Expected behavior: Shoul

Null template_dir should remain None

Test that OpenAILLM stores and uses reasoning_effort from config

Test that children are added to the same island as their parents

Test that early_stopping_patience defaults to None (disabled)

Test pending artifacts storage and retrieval

Test per-dimension bin configuration

Test that per-island MAP-Elites works correctly with migration

Test population size limit enforcement

Test that callable evaluator can be executed in a subprocess

Test _prepare_evaluator with callable function

Test _prepare_evaluator with existing file

Test _prepare_evaluator with code string

Test _prepare_evaluator raises error for invalid code string

Test _prepare_program with existing file

Test _prepare_program with list of lines

Test _prepare_program with code string

Test _prepare_program doesn't add duplicate markers

Test processing dict results

Test processing EvaluationResult objects

Test that _process_evaluation_result handles artifacts correctly

Test that _process_evaluation_result handles dict results correctly

Test that ProcessParallelController correctly passes file suffix

Test that programs are added to their island's specific feature map

Test that program generations are tracked correctly

Create a test program file

Test that programs are assigned to correct islands

Test that programs_as_changes_description defaults to False

Test error handling when programs directory is missing

Test PromptConfig with custom changes description values

Test PromptConfig defaults for changes description

Test the proposed fix: using a method that doesn't modify shared state This simulates what the fix would look like

Test with real example config file

Test a scenario similar to the reported bug

Test that reasoning_effort can be loaded from YAML config at LLM level

Test that reasoning_effort can be specified per model

Test that model configs inherit reasoning_effort from parent LLM config

Test that model-level reasoning_effort overrides LLM-level

Test that reasoning_effort is included in API call parameters

Test various model names to ensure correct reasoning model detection

Test rebuilding with only primary model

Test rebuilding models with both primary and secondary models

Test CLI override of config file models

Test rebuilding with custom weights

Test that secondary model with zero weight is excluded

Test that rebuilding preserves shared configuration

Relative paths in template_dir should resolve relative to config file

Test that all artifacts are included using .items() without prioritization

Test that all artifacts are included using .items()

Test that security filter redacts potential tokens

Test artifact truncation for large content

Test graceful shutdown request

Test that programs_as_changes_description requires diff_based_evolution

Test that missing environment variable raises ValueError

Test that strings without ${VAR} are returned unchanged

Test that None is returned unchanged

Test that _resolve_env_var resolves ${VAR} syntax

Test that resume correctly continues from checkpoint

Test basic evolution run

Test run_evolution with cleanup=False

Test run_evolution with real optillm server - basic program evolution

Test run_evolution with Config object

Test safe decoding of bytes artifacts

Test safe decoding of invalid bytes

Test safe decoding of string artifacts

Test sampling from the database

Test that sampling from empty island falls back to available programs

Test that sampled parent is actually from the requested island

Test that different islands return different programs

Test that inspiration sampling respects island boundaries

Test that NaN in metrics is sanitized to None

Test that -inf in metrics is sanitized to None