openai / mle-bench Source Code Overview |
Analysis scope, overview of main, test, generated, deployment, build, and other code.
Source Code Analysis Scope
Files includes and excluded from analyses
- 11 extensions are included in analyses: json, py, md, yaml, txt, sh, gitignore, gitattributes, dockerignore, toml, mjs
- 2 criteria are used to exclude files from analysis:
- exclude files with path like ".*/git[-]history[.]txt" (Git history) (1 file).
- exclude files with path like ".*/git[-][a-zA-Z0-9_]+[.]txt" (Git data exports for sokrates analyses) (0 files).
- The source code folder contains 1,016 files:
- 915 files are included in analyses.
- 101 files are excluded from analyses:
Overview of Analyzed Files
Basic stats on analyzed files
Intro
For analysis purposes we separate files in scope into several categories: main, test, generated, deployment and build, and other.
- The main category contains all manually created source code files that are being used in the production.
- Files in the main category are used as input for other analyses: logical decomposition, concerns, duplication, file size, unit size, and conditional complexity.
- Test source code files are used only for testing of the product. These files are normally not deployed to production.
- Build and deployment source code files are used to configure or support build and deployment process.
- Generated source code files are automatically generated files that have not been manually changed after generation.
- While a source code folder may contain a number of files, we are primarily interested in the source code files that are being written and maintained by developers.
- Files containing binaries, documentation, or third-party libraries, for instance, are excluded from analysis. The exception are third-party libraries that have been changed by developers.
Main Code
All manually created or maintained source code that defines logic of the product that is run in a production environment.
- The following criteria are used to filter files:
- files with paths like ".*".
- 382 files match defined criteria (14,775 LOC, 100.0% vs. main code):
- 212 *.py files (12,470 LOC)
- 169 *.yaml files (2,273 LOC)
- 1 *.mjs files (32 LOC)
- " *.py" is biggest, containing 84.4% of LOC.
- " *.mjs" is smallest, containing 0.22% of LOC.
Test Code
Used only for testing of the product. Normally not deployed in a production environment.
Generated Code
Automatically generated files, not manually changed after generation.
Build and Deployment Code
Source code used to configure or support build and deployment process.
- The following criteria are used to filter files:
- files with paths like ".*[.]git[a-z]+".
- files with paths like ".*/[.]gitattributes".
- files with paths like ".*[.]sh".
- files with paths like ".*/requirements[a-zA-Z0-9._-]*[.]txt".
- files with paths like ".*/package[.]json".
- files with paths like ".*/package[-]lock[.]json".
- files with paths like ".*/[.]gitignore".
- files with paths like ".*/[.]dockerignore".
- files with paths like "(.*/)?[.]dockerignore".
- files with paths like ".*/setup[.]py".
- files with paths like ".*/pyproject[.]toml".
- 16 files match defined criteria (417 LOC, 2.8% vs. main code):
- 8 *.sh files (194 LOC)
- 3 *.gitignore files (126 LOC)
- 1 *.toml files (50 LOC)
- 1 *.py files (39 LOC)
- 2 *.gitattributes files (5 LOC)
- 1 *.dockerignore files (3 LOC)
- " *.sh" is biggest, containing 46.52% of LOC.
- " *.dockerignore" is smallest, containing 0.72% of LOC.
Other Code
- The following criteria are used to filter files:
- files with paths like ".*[.]txt".
- files with paths like ".*[.]json".
- files with paths like ".*[.]md".
- files with paths like ".*/README[.][a-z0-9]+".
- files with paths like ".*/[Ee]xamples/.*".
- 510 files match defined criteria (14,248 LOC, 96.4% vs. main code):
- 167 *.md files (8,833 LOC)
- 99 *.txt files (4,042 LOC)
- 241 *.json files (1,315 LOC)
- 1 *.py files (48 LOC)
- 1 *.sh files (8 LOC)
- 1 *.gitignore files (2 LOC)
- " *.md" is biggest, containing 61.99% of LOC.
- " *.gitignore" is smallest, containing 0.01% of LOC.
Analyzers
Info about analyzers used for source code examinations.
- *.py files are analyzed with PythonAnalyzer:
- All basic standard analyses supported (source code overview, duplication, file size, concerns, findings, metrics, controls)
- Advanced code cleaning (empty lines and comments removed for LOC calculations, additional cleaning for duplication calculations)
- Unit size analysis
- Conditional complexity analysis
- Basic heuristic dependency analysis
- *.yaml files are analyzed with YamlAnalyzer:
- All basic standard analyses supported (source code overview, duplication, file size, concerns, findings, metrics, controls)
- Advanced code cleaning (empty lines and comments removed for LOC calculations, additional cleaning for duplication calculations)
- No unit size analysis
- No conditional complexity analysis
- No dependency analysis
- *.mjs files are analyzed with DefaultLanguageAnalyzer:
- All basic standard analyses supported (source code overview, duplication, file size, concerns, findings, metrics, controls)
- Basic code cleaning (empty lines removed for LOC calculations and duplication calculations)
- No unit size analysis
- No conditional complexity analysis
- No dependency analysis
2026-06-24 10:36