cedarscript-integration-aider 0.0.11

Allows `Aider` to use CEDARScript as an edit format

CEDARScript Integration: Aider

CEDARScript Integration: Aider enables Aider to use CEDARScript as an edit format.

Table of Contents

• What is CEDARScript?
• Installation
• Usage
• Why Use CEDARScript?
• Performance Comparison
• Individual Test Analysis
• Detailed Analysis
• Contributing
• License

What is CEDARScript?

CEDARScript (Concise Examination, Development, And Refactoring Script) is a domain-specific language designed to improve how AI coding assistants interact with codebases and communicate their code modification intentions. It provides a standardized way to express complex code modification and analysis operations, making it easier for AI-assisted development tools to understand and execute these tasks.

How to use it

1. Install Aider, if you haven't.
2. Now, simply use the --edit-format switch and select cedarscript:

aider --model gemini/gemini-1.5-flash-latest --edit-format cedarscript

Why use CEDARScript?

TL;DR: You can get higher success rates when compared to other edit formats.

1. Higher Success Rates: Significantly improves the performance of AI models in code refactoring tasks.
2. Cost-Effective Performance: Enables more affordable models to compete with top-tier options.
3. Standardized Communication: Provides a consistent format for AI-code interaction in coding tasks.
4. Enhanced Accuracy: Reduces errors and improves the quality of AI-generated code modifications.

Performance Comparison

CEDARScript has shown remarkable improvements in AI model performance for code refactoring:

Model	Format	Pass Rate	Well-Formed Cases	Syntax Errors	Indentation Errors	Cost	Avg. Time per case
Gemini 1.5 PRO	CEDARScript	77.5%	86.5%	4	3	26.2	29
Gemini 1.5 Flash	CEDARScript	76.4%	94.4%	3	5	0.68	14.7
Claude 3.5 Sonnet	diff	64.0%	76.4%	n/a	n/a	n/a	n/a
Gemini 1.5 PRO	diff-fenced	49.4%	7.9%	21	93	28.3	110.1

Notable Achievements:

• Gemini 1.5 PRO with CEDARScript outperformed both its diff-fenced format and Claude 3.5 Sonnet.
• Most remarkably, the more cost-effective Gemini 1.5 Flash model, using CEDARScript, outperformed Claude 3.5 Sonnet.
  • It goes to show that even a more affordable model can surpass top-tier competitors when equipped with the right tools.

This suggests that CEDARScript can level the playing field, enabling more accessible AI models to compete with and even exceed the capabilities of more expensive options in certain coding tasks.

Individual Test Analysis

Overall Performance

	Count	%
Improved	39	43.8%
Worsened	7	7.9%
Stable	43	48.3%
Total tests	89	100%

Individual Test Diff

--- 2024-10-22-05-21-41--gemini-1.5-flash-refactoring-whole
+++ 2024-10-22-05-13-37--gemini-1.5-flash-refactoring-cedarscript-i0.0.9-e0.3.3
# ============= Failed Attempts per Test =============
# N >= 0: it eventually passed after N failed attempts
# N < 0 : All attempts failed and limit was reached

@@ Improved, now PASSED (36) @@
++analyzer_cli_DebugAnalyzer__make_source_table: -4 -> 0
++autodetector_MigrationAutodetector__trim_to_apps: -4 -> 0
++backends_ModelBackend_with_perm: -4 -> 0
++builtin_BuiltinVariable_call_setattr: -4 -> 0
++checks_BaseModelAdminChecks__check_ordering_item: -4 -> 0
++checks_BaseModelAdminChecks__check_raw_id_fields_item: -4 -> 0
++checks_ModelAdminChecks__check_action_permission_methods: -4 -> 0
++checks_ModelAdminChecks__check_inlines_item: -4 -> 0
++checks_ModelAdminChecks__check_list_display_item: -4 -> 0
++clustering_ops_KMeans__mini_batch_training_op: -4 -> 0
++codeeditor_CodeEditor___get_brackets: -4 -> 0
++config_AppConfig__path_from_module: -4 -> 0
++config_ConfigCLI__get_settings_vars: -4 -> 0
++coordinator_HERETransitDataUpdateCoordinator__parse_transit_response: -4 -> 0
++cuda_cpp_scheduling_CUDACPPScheduling__can_fuse_epilogue_impl: -4 -> 0
++dataframeeditor_DataFrameView_next_index_name: -4 -> 0
++diffsettings_Command_output_hash: -4 -> 0
++dim2_Dim2CompatTests_test_reductions_2d_axis0: -4 -> 3
++distribution_DistributionFiles_parse_distribution_file_SUSE: -4 -> 0
++doc_DocCLI_get_role_man_text: -4 -> 0
++figure_FigureBase_colorbar: -4 -> 0
++functional_Functional__conform_to_reference_input: -4 -> 0
++galaxy_GalaxyCLI_execute_list_collection: -4 -> 0
++kernel_SpyderKernel_get_fault_text: -4 -> 1
++main_widget_PylintWidget_parse_output: -4 -> 0
++methods_BaseMethodsTests_test_where_series: -4 -> 0
++ogrinspect_Command_add_arguments: -4 -> 0
++onnxfunction_dispatcher_OnnxFunctionDispatcher__get_aten_name: -4 -> 1
++operations_DatabaseOperations_last_executed_query: -4 -> 0
++polar_RadialTick__determine_anchor: -4 -> 0
++profile_analyzer_cli_ProfileAnalyzer__get_list_profile_lines: -4 -> 0
++split_cat_SplitCatSimplifier_replace_cat: -4 -> 0
++split_cat_SplitCatSimplifier_replace_split: -4 -> 0
++text_CountVectorizer__limit_features: -4 -> 0
++triton_TritonScheduling_define_kernel: -4 -> 0
++triton_TritonScheduling_generate_node_schedule: -4 -> 0

@@ Improved, minor (3) @@
+ checks_BaseModelAdminChecks__check_autocomplete_fields_item: 3 -> 0
+ dataloader_DataLoader__is_role: 2 -> 0
+ operations_OracleOperations_convert_extent: 1 -> 0

@@ Worsened, now FAILED (7) @@
--base_BaseHandler_adapt_method_mode: 0 -> -4
--feedgenerator_Atom1Feed_add_item_elements: 0 -> -4
--generic_bsd_GenericBsdIfconfigNetwork_parse_inet_line: 0 -> -4
--graph_drawer_FxGraphDrawer__stringify_tensor_meta: 0 -> -4
--group_batch_fusion_GroupLinearFusion_fuse: 0 -> -4
--inspectdb_Command_normalize_col_name: 0 -> -4
--introspection_DatabaseIntrospection__parse_column_or_constraint_definition: 0 -> -4

@@ Stable: PASSED (30) @@
=+autosave_AutosaveForPlugin_get_files_to_recover: 0
=+base_BaseHandler_check_response: 0
=+baseconv_BaseConverter_convert: 0
=+compile_utils_MetricsContainer__get_metric_object: 0
=+concat__Concatenator__clean_keys_and_objs: 0
=+config_NetworkConfig_parse: 0
=+csrf_CsrfViewMiddleware__set_csrf_cookie: 0
=+dumpdata_Command_add_arguments: 0
=+finders_FileSystemFinder_check: 0
=+gateway_Gateway_get_and_delete_all_sms: 0
=+getitem_BaseGetitemTests_test_get: 0
=+grad_scaler_GradScaler__unscale_grads_: 0
=+gradient_checker_GradientChecker__assertInferTensorChecks: 0
=+graph_MigrationGraph_iterative_dfs: 0
=+grpc_debug_server_EventListenerBaseServicer__process_tensor_event_in_chunks: 0
=+i18n_JavaScriptCatalog_get_paths: 0
=+inspectdb_Command_get_field_type: 0
=+inspectdb_Command_get_meta: 0
=+introspection_DatabaseIntrospection__get_column_collations: 0
=+load_v1_in_v2__EagerSavedModelLoader__extract_signatures: 0
=+makemessages_Command_add_arguments: 0
=+makemigrations_Command_add_arguments: 0
=+migrate_Command_add_arguments: 0
=+operations_DatabaseOperations_bulk_insert_sql: 0
=+operations_DatabaseOperations_check_expression_support: 0
=+reshaping_BaseReshapingTests_test_concat_mixed_dtypes: 0
=+schema_DatabaseSchemaEditor_quote_value: 0
=+shell_Command_python: 0
=+special_RunSQL__run_sql: 0
=+weather_NWSWeather__forecast: 0

@@ Stable: FAILED (13) @@
=-autodetector_MigrationAutodetector_check_dependency: -4 -> -4
=-checks_ModelAdminChecks__check_list_editable_item: -4 -> -4
=-common_methods_invocations_foreach_inputs_sample_func__sample_rightmost_arg: -4 -> -4
=-common_utils_TestCase_genSparseTensor: -4 -> -4
=-doc_DocCLI_display_plugin_list: -4 -> -4
=-generator_GenOpTestCase_out_variant_op_test_case_generator: -4 -> -4
=-options_ModelAdmin_message_user: -4 -> -4
=-patches__Curve__get_arrow_wedge: -4 -> -4
=-quiver_Barbs__make_barbs: -4 -> -4
=-reshaping_BaseReshapingTests_test_unstack: -4 -> -4
=-sharding_policies_MaxShardSizePolicy__add_partition: -4 -> -4
=-split_cat_SplitCatSimplifier_get_transform_params: -4 -> -4
=-symbolic_shapes_ShapeEnv_bind_symbols: -4 -> -4

# =============          TOTALS          =============
# IMPROVED: 39
#    Now PASSES: 36
#    Minor     : 3
# WORSENED: 7
#    Now FAILED: 7
#    Minor     : 0
# STABLE  : 43
#    PASSED: 30
#    FAILED: 13
# TOTAL  : 89

Detailed Analysis

This overview suggests that the CEDARScript edit format has had a significant positive impact on the task of method extraction, with improvements in nearly half of the tests and only a small percentage of tests worsening.

Improvements:

36 tests that previously failed now pass. This is a substantial improvement, indicating that CEDARScript is more effective in correctly extracting methods from classes across a wide range of codebases.

Notable improvements include:

• analyzer_cli_DebugAnalyzer__make_source_table: Suggests better handling of debug-related code refactoring.
• autodetector_MigrationAutodetector__trim_to_apps: Indicates improved capability in refactoring Django migration-related code.
• cuda_cpp_scheduling_CUDACPPScheduling__can_fuse_epilogue_impl: Shows better performance in handling complex CUDA-related code.
• triton_TritonScheduling_define_kernel and triton_TritonScheduling_generate_node_schedule: Demonstrates improved capability in refactoring GPU computing-related code.

Minor Improvements:

3 tests showed minor improvements, such as checks_BaseModelAdminChecks__check_autocomplete_fields_item, suggesting slight enhancements in handling Django admin-related code.

Regressions:

7 tests that previously passed now fail. While concerning, it's a relatively small number compared to the improvements.

Notable regressions include:

• feedgenerator_Atom1Feed_add_item_elements: Suggests potential issues with refactoring feed generation code.
• generic_bsd_GenericBsdIfconfigNetwork_parse_inet_line: Indicates challenges in refactoring network-related parsing code.
• introspection_DatabaseIntrospection__parse_column_or_constraint_definition: Shows difficulties in handling database schema introspection code.

Stability:

30 tests remained stable and passing, indicating that CEDARScript maintained performance in many areas, including various Django commands, database operations, and utility functions. 13 tests remained stable but failing, suggesting that some challenging areas were not addressed by either format. These include complex operations like symbolic_shapes_ShapeEnv_bind_symbols and reshaping_BaseReshapingTests_test_unstack.

Analysis by Domain:

• Web Frameworks (e.g., Django): Generally improved, with better handling of model admin checks, configuration, and database operations.
• Data Science and ML: Mixed results. Improvements in areas like clustering operations (clustering_ops_KMeans__mini_batch_training_op) and data frame handling, but persistent issues in some reshaping operations.
• System-level Operations: Some improvements (e.g., distribution_DistributionFiles_parse_distribution_file_SUSE) but also regressions (e.g., generic_bsd_GenericBsdIfconfigNetwork_parse_inet_line).
• GPU and High-Performance Computing: Significant improvements, especially in CUDA and Triton-related code.

Interpretation:

CEDARScript appears more effective in handling complex code structures, especially in areas related to web frameworks, data processing, and high-performance computing. It shows improved capability in understanding class contexts and correctly extracting methods across various domains. However, it may introduce new challenges in certain specific areas, possibly due to its different approach to code manipulation.

Areas for Further Investigation:

Understanding why certain tests regressed (e.g., feedgenerator_Atom1Feed_add_item_elements) could provide insights for improvement.

Analyzing the stable failing tests (e.g., symbolic_shapes_ShapeEnv_bind_symbols) to see if CEDARScript can be enhanced to address these persistent issues.

Conclusion:

• The introduction of CEDARScript appears to be a significant improvement for the task of extracting methods from classes.
• It shows particular strength in handling complex codebases, especially those related to web frameworks, data processing, and high-performance computing.
• However, care should be taken to address the areas where regressions occurred, particularly in system-level operations and certain parsing tasks.
• The consistent performance across various domains suggests that CEDARScript offers a more robust and versatile approach to code refactoring.

This analysis indicates that CEDARScript is a promising enhancement to Aider, offering more accurate and comprehensive method extraction capabilities across a wide range of codebases. However, it also highlights the need for continued refinement, especially in areas where regressions were observed.

Benchmark Metrics

Sonnet 3.5 + diff

- dirname: refac-claude-3.5-sonnet-diff-not-lazy
  model: claude-3.5-sonnet (diff)
  edit_format: diff
  pass_rate_1: 64.0
  percent_cases_well_formed: 76.4

Gemini 1.5 PRO + diff-fenced (leaderboard site)

- dirname: refac-gemini
  model: gemini/gemini-1.5-pro-latest
  edit_format: diff-fenced
  pass_rate_1: 49.4
  percent_cases_well_formed: 7.9

Gemini 1.5 PRO + diff-fenced (own tests)

- dirname: 2024-10-05-00-43-21--diff-fenced-Gemini-Refactoring
  test_cases: 89
  model: gemini/gemini-1.5-pro-latest
  edit_format: diff-fenced
  commit_hash: 772710b-dirty
  pass_rate_1: 18.0
  pass_rate_2: 21.3
  pass_rate_3: 24.7
  percent_cases_well_formed: 34.8
  error_outputs: 180
  num_malformed_responses: 180
  num_with_malformed_responses: 58
  user_asks: 128
  lazy_comments: 2
  syntax_errors: 21
  indentation_errors: 93
  exhausted_context_windows: 0
  test_timeouts: 0
  command: aider --model gemini/gemini-1.5-pro-latest
  date: 2024-10-05
  versions: 0.57.2.dev
  seconds_per_case: 110.1
  total_cost: 28.2515

Gemini 1.5 PRO + CEDARScript

- dirname: 2024-10-19-22-48-07--cedarscript-0.3.1-refactoring-gemini1.5pro
  test_cases: 89
  model: gemini/gemini-1.5-pro-latest
  edit_format: cedarscript-g
  commit_hash: 4da1e9b-dirty
  pass_rate_1: 77.5
  percent_cases_well_formed: 86.5
  error_outputs: 337
  num_malformed_responses: 19
  num_with_malformed_responses: 12
  user_asks: 12
  lazy_comments: 0
  syntax_errors: 4
  indentation_errors: 3
  exhausted_context_windows: 0
  test_timeouts: 0
  command: aider --model gemini/gemini-1.5-pro-latest
  date: 2024-10-19
  versions: 0.59.2.dev
  seconds_per_case: 29.0
  total_cost: 26.2374

Gemini 1.5 Flash + whole

- dirname: 2024-10-22-05-21-41--gemini-1.5-flash-refactoring-whole
  test_cases: 89
  model: gemini/gemini-1.5-flash-002
  edit_format: whole
  commit_hash: feb1c38
  pass_rate_1: 41.6
  pass_rate_2: 42.7
  pass_rate_3: 43.8
  pass_rate_4: 44.9
  percent_cases_well_formed: 100.0
  error_outputs: 0
  num_malformed_responses: 0
  num_with_malformed_responses: 0
  user_asks: 558
  lazy_comments: 28
  syntax_errors: 106
  indentation_errors: 20
  exhausted_context_windows: 0
  test_timeouts: 0
  command: aider --model gemini/gemini-1.5-flash-002
  date: 2024-10-22
  versions: 0.59.2.dev
  seconds_per_case: 231.1
  total_cost: 2.3894

Gemini 1.5 Flash + CEDARScript

- dirname: 2024-10-20-00-33-27--cedarscript-0.3.1-refactoring-gemini-1.5-flash
  test_cases: 89
  model: gemini/gemini-1.5-flash-latest
  edit_format: cedarscript-g
  commit_hash: 4da1e9b-dirty
  pass_rate_1: 76.4
  percent_cases_well_formed: 94.4
  error_outputs: 403
  num_malformed_responses: 13
  num_with_malformed_responses: 5
  user_asks: 21
  lazy_comments: 0
  syntax_errors: 3
  indentation_errors: 5
  exhausted_context_windows: 0
  test_timeouts: 0
  command: aider --model gemini/gemini-1.5-flash-latest
  date: 2024-10-20
  versions: 0.59.2.dev
  seconds_per_case: 14.7
  total_cost: 0.6757

functional_Functional__conform_to_reference_input

diff-fenced

    "cost": 0.33188854999999995,
    "duration": 27.793912172317505,
    "test_timeouts": 0,
    "commit_hash": "772710b-dirty",
    "num_error_outputs": 2,
    "num_user_asks": 3,
    "num_exhausted_context_windows": 0,
    "num_malformed_responses": 2,
    "syntax_errors": 0,
    "indentation_errors": 3,
    "lazy_comments": 0,

cedarscript

    "cost": 0.18178265,
    "duration": 11.176445960998535,
    "test_timeouts": 0,
    "commit_hash": "772710b-dirty",
    "num_error_outputs": 0,
    "num_user_asks": 1,
    "num_exhausted_context_windows": 0,
    "num_malformed_responses": 0,
    "syntax_errors": 0,
    "indentation_errors": 0,
    "lazy_comments": 0,

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License.