xiaxinhang/pr-agent
1
0
Fork 0
mirror of https://github.com/qodo-ai/pr-agent.git synced 2025-07-03 04:10:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

enable ai_metadata

Browse Source
This commit is contained in:
mrT23
2024-09-07 17:25:05 +03:00
parent 24f7e8622f
commit 8706f643ef
32 changed files with 338 additions and 117 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
docs/docs/core-abilities/code_oriented_yaml.md Normal file
View File

@ -0,0 +1,2 @@
## Overview
TBD

47
docs/docs/core-abilities/compression_strategy.md Normal file
View File

@ -0,0 +1,47 @@
## Overview - PR Compression Strategy
There are two scenarios:
1. The PR is small enough to fit in a single prompt (including system and user prompt)
2. The PR is too large to fit in a single prompt (including system and user prompt)
For both scenarios, we first use the following strategy
#### Repo language prioritization strategy
We prioritize the languages of the repo based on the following criteria:
1. Exclude binary files and non code files (e.g. images, pdfs, etc)
2. Given the main languages used in the repo
3. We sort the PR files by the most common languages in the repo (in descending order):
* ```[[file.py, file2.py],[file3.js, file4.jsx],[readme.md]]```
### Small PR
In this case, we can fit the entire PR in a single prompt:
1. Exclude binary files and non code files (e.g. images, pdfs, etc)
2. We Expand the surrounding context of each patch to 3 lines above and below the patch
### Large PR
#### Motivation
Pull Requests can be very long and contain a lot of information with varying degree of relevance to the pr-agent.
We want to be able to pack as much information as possible in a single LMM prompt, while keeping the information relevant to the pr-agent.
#### Compression strategy
We prioritize additions over deletions:
- Combine all deleted files into a single list (`deleted files`)
- File patches are a list of hunks, remove all hunks of type deletion-only from the hunks in the file patch
#### Adaptive and token-aware file patch fitting
We use [tiktoken](https://github.com/openai/tiktoken) to tokenize the patches after the modifications described above, and we use the following strategy to fit the patches into the prompt:
1. Within each language we sort the files by the number of tokens in the file (in descending order):
- ```[[file2.py, file.py],[file4.jsx, file3.js],[readme.md]]```
2. Iterate through the patches in the order described above
3. Add the patches to the prompt until the prompt reaches a certain buffer from the max token length
4. If there are still patches left, add the remaining patches as a list called `other modified files` to the prompt until the prompt reaches the max token length (hard stop), skip the rest of the patches.
5. If we haven't reached the max token length, add the `deleted files` to the prompt until the prompt reaches the max token length (hard stop), skip the rest of the patches.
#### Example
![Core Abilities](https://codium.ai/images/git_patch_logic.png){width=768}

2
docs/docs/core-abilities/dynamic_context.md Normal file
View File

@ -0,0 +1,2 @@
## Overview - Asymmetric and dynamic PR context
TBD

2
docs/docs/core-abilities/impact_evaluation.md Normal file
View File

@ -0,0 +1,2 @@
## Overview - Impact evaluation 💎
TBD

62
docs/docs/core-abilities/index.md
View File

@ -1,52 +1,10 @@
## PR Compression Strategy
There are two scenarios:
1. The PR is small enough to fit in a single prompt (including system and user prompt)
2. The PR is too large to fit in a single prompt (including system and user prompt)
For both scenarios, we first use the following strategy
#### Repo language prioritization strategy
We prioritize the languages of the repo based on the following criteria:
1. Exclude binary files and non code files (e.g. images, pdfs, etc)
2. Given the main languages used in the repo
3. We sort the PR files by the most common languages in the repo (in descending order):
* ```[[file.py, file2.py],[file3.js, file4.jsx],[readme.md]]```
### Small PR
In this case, we can fit the entire PR in a single prompt:
1. Exclude binary files and non code files (e.g. images, pdfs, etc)
2. We Expand the surrounding context of each patch to 3 lines above and below the patch
### Large PR
#### Motivation
Pull Requests can be very long and contain a lot of information with varying degree of relevance to the pr-agent.
We want to be able to pack as much information as possible in a single LMM prompt, while keeping the information relevant to the pr-agent.
#### Compression strategy
We prioritize additions over deletions:
- Combine all deleted files into a single list (`deleted files`)
- File patches are a list of hunks, remove all hunks of type deletion-only from the hunks in the file patch
#### Adaptive and token-aware file patch fitting
We use [tiktoken](https://github.com/openai/tiktoken) to tokenize the patches after the modifications described above, and we use the following strategy to fit the patches into the prompt:
1. Within each language we sort the files by the number of tokens in the file (in descending order):
- ```[[file2.py, file.py],[file4.jsx, file3.js],[readme.md]]```
2. Iterate through the patches in the order described above
3. Add the patches to the prompt until the prompt reaches a certain buffer from the max token length
4. If there are still patches left, add the remaining patches as a list called `other modified files` to the prompt until the prompt reaches the max token length (hard stop), skip the rest of the patches.
5. If we haven't reached the max token length, add the `deleted files` to the prompt until the prompt reaches the max token length (hard stop), skip the rest of the patches.
#### Example
![Core Abilities](https://codium.ai/images/git_patch_logic.png){width=768}
## YAML Prompting
TBD
## Static Code Analysis 💎
TBD
# Core Abilities
PR-Agent utilizes a variety of core abilities to provide a comprehensive and efficient code review experience. These abilities include:
- [Local and global metadata](core-abilities/metadata.md)
- [Line localization](core-abilities/line_localization.md)
- [Dynamic context](core-abilities/dynamic_context.md)
- [Self-reflection](core-abilities/self_reflection.md)
- [Interactivity](core-abilities/interactivity.md)
- [Compression strategy](core-abilities/compression_strategy.md)
- [Code-oriented YAML](core-abilities/code_oriented_yaml.md)
- [Static code analysis](core-abilities/static_code_analysis.md)

2
docs/docs/core-abilities/interactivity.md Normal file
View File

@ -0,0 +1,2 @@
## Interactive invocation 💎
TBD

2
docs/docs/core-abilities/line_localization.md Normal file
View File

@ -0,0 +1,2 @@
## Overview - Line localization
TBD

56
docs/docs/core-abilities/metadata.md Normal file
View File

@ -0,0 +1,56 @@
## Overview - Local and global metadata injection with multi-stage analysis
(1)
PR-Agent initially retrieves for each PR the following data:
- PR title and branch name
- PR original description
- Commit messages history
- PR diff patches, in [hunk diff](https://loicpefferkorn.net/2014/02/diff-files-what-are-hunks-and-how-to-extract-them/) format
- The entire content of the files that were modified in the PR
In addition, PR-Agent is able to receive from the user additional data, like [`extra_instructions` and `best practices`](https://pr-agent-docs.codium.ai/tools/improve/#extra-instructions-and-best-practices) that can be used to enhance the PR analysis.
(2)
By default, the first command that PR-Agent executes is [`describe`](https://pr-agent-docs.codium.ai/tools/describe/), which generates three types of outputs:
- PR Type (e.g. bug fix, feature, refactor, etc)
- PR Description - a bullet points summary of the PR
- Changes walkthrough - going file-by-file, PR-Agent generate a one-line summary and longer bullet points summary of the changes in the file
These AI-generated outputs are now considered part of the PR metadata, and can be used in subsequent commands like `review` and `improve`.
This effectively enables chain-of-thought analysis, without doing any additional API calls which will cost time and money.
(3)
For example, when generating code suggestions for different files, PR-Agent can inject the AI-generated file summary in the prompt:
```
## File: 'src/file1.py'
### AI-generated file summary:
- edited function `func1` that does X
- Removed function `func2` that was not used
- ....
@@ ... @@ def func1():
__new hunk__
11 unchanged code line0 in the PR
12 unchanged code line1 in the PR
13 +new code line2 added in the PR
14 unchanged code line3 in the PR
__old hunk__
unchanged code line0
unchanged code line1
-old code line2 removed in the PR
unchanged code line3
@@ ... @@ def func2():
__new hunk__
...
__old hunk__
...
```
(4) The entire PR files that were retrieved are used to expand and enhance the PR context (see [Dynamic Context](https://pr-agent-docs.codium.ai/core-abilities/dynamic-context/)).
(5) All the metadata described above represent several level of analysis - from hunk level, to file level, to PR level, and enables PR-Agent AI models to generate more accurate and relevant suggestions.
## Example result for prompt with metadata injection
TBD

2
docs/docs/core-abilities/self_reflection.md Normal file
View File

@ -0,0 +1,2 @@
## Overview - Self-reflection and suggestion cleaning and re-ranking
TBD

2
docs/docs/core-abilities/static_code_analysis.md Normal file
View File

@ -0,0 +1,2 @@
## Overview - Static Code Analysis 💎
TBD

11
docs/mkdocs.yml
View File

@ -41,7 +41,16 @@ nav:
- 💎 Custom Prompt: 'tools/custom_prompt.md'
- 💎 CI Feedback: 'tools/ci_feedback.md'
- 💎 Similar Code: 'tools/similar_code.md'
- Core Abilities: 'core-abilities/index.md'
- Core Abilities:
- 'core-abilities/index.md'
- Local and global metadata: 'core-abilities/metadata.md'
- Line localization: 'core-abilities/line_localization.md'
- Dynamic context: 'core-abilities/dynamic_context.md'
- Self-reflection: 'core-abilities/self_reflection.md'
- Interactivity: 'core-abilities/interactivity.md'
- Compression strategy: 'core-abilities/compression_strategy.md'
- Code-oriented YAML: 'core-abilities/code_oriented_yaml.md'
- Static code analysis: 'core-abilities/static_code_analysis.md'
- Chrome Extension:
- PR-Agent Chrome Extension: 'chrome-extension/index.md'
- Features: 'chrome-extension/features.md'