CompileLoop
An open-source Java tool that uses local code models to automatically fix Maven build compilation errors. It applies and verifies patches in a temporary copy, presenting only patches that pass verification to developers.
Timeline (last 90 days)
CompileLoop is an open-source Java tool that takes a project with a failed Maven build, sends the compilation errors to a local code model, obtains fix suggestions, applies the patch in a temporary copy and rebuilds to verify, leaving the original source code unchanged.
In the benchmark tests, the model could generate semantically correct fixes, but the line counts in the unified diff headers were wrong, causing the patch tool to refuse to apply them.
CompileLoop borrows the idea of git apply --recount, recalculating line counts from the diff body and using deterministic code to correct fields that can be deterministically computed, rather than letting the AI retry.
In the AI coding toolchain, the executable artifacts produced by the model (patches, commands, file contents) should serve as the basis for verification, while natural language explanations (summaries, justifications) should not be treated as a source of truth by tool logic.
CompileLoop added duplicate patch detection (generating fingerprints using Java String.hashCode()) and error-unchanged detection (stopping after failing twice in a row with the same diagnostic information) as early termination conditions.
In the benchmark's 60 runs, the early termination conditions were triggered 27 times, including 21 detections of lack of progress and 6 cases of model provider unavailability.
DeepSeek-Coder 6.7B often correctly identified the fix in its summary, but the generated diff contained context lines that did not match the source file, causing the patch to be rejected as unsafe to apply.
The most time-consuming part of the CompileLoop project is the engineering infrastructure around the model, including the process runner that executes Maven, the parser that parses Maven output, the patch validator, the temporary workspace application test, and the repair engine that records attempts and detects stalled progress.
The CompileLoop project currently has 233 tests, CI runs on Ubuntu and Windows, and the test suite uses a simulated Ollama server and scripted model responses, so it can run without downloading local models.
CompileLoop's temporary copy is not a secure sandbox, build scripts can execute arbitrary commands, and container isolation is an important next step.
1 more timeline events
All Facts (11)
CompileLoop is an open-source Java tool that takes a project with a failed Maven build, sends the compilation errors to a local code model, obtains fix suggestions, applies the patch in a temporary copy and rebuilds to verify, leaving the original source code unchanged.
50%UnverifiedIn the benchmark tests, the model could generate semantically correct fixes, but the line counts in the unified diff headers were wrong, causing the patch tool to refuse to apply them.
50%UnverifiedCompileLoop borrows the idea of git apply --recount, recalculating line counts from the diff body and using deterministic code to correct fields that can be deterministically computed, rather than letting the AI retry.
50%UnverifiedIn the AI coding toolchain, the executable artifacts produced by the model (patches, commands, file contents) should serve as the basis for verification, while natural language explanations (summaries, justifications) should not be treated as a source of truth by tool logic.
50%UnverifiedCompileLoop added duplicate patch detection (generating fingerprints using Java String.hashCode()) and error-unchanged detection (stopping after failing twice in a row with the same diagnostic information) as early termination conditions.
50%UnverifiedIn the benchmark's 60 runs, the early termination conditions were triggered 27 times, including 21 detections of lack of progress and 6 cases of model provider unavailability.
50%UnverifiedDeepSeek-Coder 6.7B often correctly identified the fix in its summary, but the generated diff contained context lines that did not match the source file, causing the patch to be rejected as unsafe to apply.
50%UnverifiedThe most time-consuming part of the CompileLoop project is the engineering infrastructure around the model, including the process runner that executes Maven, the parser that parses Maven output, the patch validator, the temporary workspace application test, and the repair engine that records attempts and detects stalled progress.
50%UnverifiedThe CompileLoop project currently has 233 tests, CI runs on Ubuntu and Windows, and the test suite uses a simulated Ollama server and scripted model responses, so it can run without downloading local models.
50%UnverifiedCompileLoop's temporary copy is not a secure sandbox, build scripts can execute arbitrary commands, and container isolation is an important next step.
50%UnverifiedCompileLoop's developer LambdaJedi uses Claude for architecture discussions and code review.
50%