DIDACT (​​Dynamic Integrated Developer ACTivity) / DevScript

- https://blog.research.google/2023/05/large-sequence-models-for-software.html
- https://files.sri.inf.ethz.ch/website/papers/icml21-tfix.pdf

From https://chat.openai.com/c/dfb3d49c-778d-4fec-a7f6-1794404f64d7


1. **Debugging & Repair Tasks:**
- **Build Repair**: Correcting issues that cause build failures, ensuring that the codebase compiles and deploys correctly.
- **Build Error Prediction**: Anticipating potential compilation errors before the build process to allow developers to make preemptive fixes.
- **TFix**: Referring to the TFix tool from ETH Zurich's SRI lab, which automatically fixes bugs based on test failures. TFix uses machine learning to suggest changes that would make failing tests pass.

2. **Code Review Tasks:**
- **Comment Prediction**: Predicting comments that code reviewers might leave, potentially highlighting areas of the code that require attention or improvement.
- **Tip Prediction**: Offering suggestions for best practices or improvements, which can be especially helpful for less experienced developers or in educational settings.
- **Comment Repair**: Responding to reviewers’ comments by making appropriate changes in the code to address the raised concerns.
- **Pre-Submit Cleanup**: Refactoring or cleaning up the code before submission to ensure it meets quality standards and is ready for review.
- **Readability Renaming**: Suggesting more readable and understandable names for variables, functions, and other code elements to improve code clarity.
- **Code Review Autocompletion**: Automating the generation of responses or fixes based on code review feedback, which can speed up the review process.

3. **Code Editing:**
- **Edit Prediction**: Predicting the next edit a developer is likely to make, given the current code context, to speed up the development process.
- **Span Denoising**: Cleaning up segments of code that are unclear, potentially problematic, or overly complex to improve code quality.
- **Variable Renaming**: Suggesting or applying more descriptive or convention-compliant names to variables to enhance code readability and maintainability.
- **History-Augmented Code Completion**: Using the historical context of changes to provide more accurate code completion suggestions.
- **Prompting & Chaining**: Guiding the model with a series of prompts to generate code or edits, creating a sequence of actions that represent a more complex development activity.