Github

GitHubAPISettings

Bases: BaseSettings

repo_url

repo_url: str = Field(..., title='Github Repository URL', description='This should be a full url to the repository, e.g. `https://github.com/Mai0313/SWEBenchV2` or `Mai0313/SWEBenchV2`', frozen=False, deprecated=False)

max_page

max_page: int | None = Field(default=None, title='Max Page', description='Maximum number of pages to fetch for PRs', frozen=False, deprecated=False)

per_page

per_page: int | None = Field(default=None, title='Per Page', description='Number of PRs to fetch per page', frozen=False, deprecated=False)

GitHubPRExtractorBase

Bases: GitHubAPISettings

token

token: str | None = Field(default=None, validation_alias='GITHUB_TOKEN', description='GitHub API token for authentication', frozen=False, deprecated=False)

base_url

base_url: str = Field(default='https://api.github.com', validation_alias='GITHUB_API_BASE_URL', description='Base URL for GitHub API', frozen=False, deprecated=False)

repo_owner

repo_owner: str

repo_name

repo_name: str

headers

headers: dict[str, Any]

repo_url

repo_url: str = Field(..., title='Github Repository URL', description='This should be a full url to the repository, e.g. `https://github.com/Mai0313/SWEBenchV2` or `Mai0313/SWEBenchV2`', frozen=False, deprecated=False)

max_page

max_page: int | None = Field(default=None, title='Max Page', description='Maximum number of pages to fetch for PRs', frozen=False, deprecated=False)

per_page

per_page: int | None = Field(default=None, title='Per Page', description='Number of PRs to fetch per page', frozen=False, deprecated=False)

GitHubPRExtractor

Bases: GitHubPRExtractorBase

Methods:

Name	Description
get_rate_limit	Retrieve current GitHub API rate limit information.
get_merged_prs	Fetch all merged pull requests from the GitHub repository.
get_pr_files	Retrieve all files modified in a specific pull request.
get_file_content	Retrieve the content of a file at a specific commit SHA.
extract_pr_data	Extract complete training data for a single pull request.
extract_all_pr_data	Extract training data from all merged pull requests in the repository.

repo_url

repo_url: str = Field(..., title='Github Repository URL', description='This should be a full url to the repository, e.g. `https://github.com/Mai0313/SWEBenchV2` or `Mai0313/SWEBenchV2`', frozen=False, deprecated=False)

max_page

max_page: int | None = Field(default=None, title='Max Page', description='Maximum number of pages to fetch for PRs', frozen=False, deprecated=False)

per_page

per_page: int | None = Field(default=None, title='Per Page', description='Number of PRs to fetch per page', frozen=False, deprecated=False)

token

token: str | None = Field(default=None, validation_alias='GITHUB_TOKEN', description='GitHub API token for authentication', frozen=False, deprecated=False)

base_url

base_url: str = Field(default='https://api.github.com', validation_alias='GITHUB_API_BASE_URL', description='Base URL for GitHub API', frozen=False, deprecated=False)

repo_owner

repo_owner: str

repo_name

repo_name: str

headers

headers: dict[str, Any]

get_rate_limit

get_rate_limit() -> RateLimit

Retrieve current GitHub API rate limit information.

Makes a synchronous request to GitHub's rate limit endpoint to check current usage and remaining quota for API calls.

Returns:

Name	Type	Description
RateLimit	RateLimit	Current rate limit status including remaining calls and reset time.

Source code in src/swebenchv2/datamodule/github.py

def get_rate_limit(self) -> RateLimit:
    """Retrieve current GitHub API rate limit information.

    Makes a synchronous request to GitHub's rate limit endpoint to check
    current usage and remaining quota for API calls.

    Returns:
        RateLimit: Current rate limit status including remaining calls and reset time.
    """
    with httpx.Client(base_url=self.base_url, headers=self.headers, timeout=10) as client:
        response = client.get("/rate_limit")
        rate_limit = RateLimit(**response.json())
        logfire.info("Rate limit info", **rate_limit.rate.model_dump())
        return rate_limit

get_merged_prs

get_merged_prs() -> list[PullRequest]

Fetch all merged pull requests from the GitHub repository.

Synchronously retrieves all merged pull requests by paginating through the GitHub API, handling rate limits and filtering for merged PRs only.

Returns:

Type	Description
list[PullRequest]	list[PullRequest]: List of all merged pull request objects from the repository.

Source code in src/swebenchv2/datamodule/github.py

def get_merged_prs(self) -> list[PullRequest]:
    """Fetch all merged pull requests from the GitHub repository.

    Synchronously retrieves all merged pull requests by paginating through
    the GitHub API, handling rate limits and filtering for merged PRs only.

    Returns:
        list[PullRequest]: List of all merged pull request objects from the repository.
    """
    all_prs: list[PullRequest] = []
    page = 1

    with httpx.Client(base_url=self.base_url, headers=self.headers, timeout=10) as client:
        while True:
            if page > self._max_page:
                break

            response = client.get(
                url=f"/repos/{self.repo_owner}/{self.repo_name}/pulls",
                params={
                    "state": "closed",
                    "sort": "updated",
                    "direction": "desc",
                    "per_page": self._per_page,
                    "page": page,
                },
            )

            if response.status_code == 403:
                reset_time = int(response.headers.get("X-RateLimit-Reset", 0))
                if reset_time:
                    wait_time = reset_time - int(time.time()) + 1
                    logfire.info(f"Rate limit exceeded. Waiting {wait_time} seconds...")
                    time.sleep(wait_time)
                    continue

            if response.status_code != 200:
                logfire.error(f"Error fetching PRs: {response.status_code}")
                break

            response_list: list[dict[str, Any]] = response.json()

            # Check if we've reached the end of available data
            if not response_list:
                logfire.info(f"No more PRs found at page {page}. Stopping pagination.")
                break

            # Only keep merged PRs
            merged_prs = [
                PullRequest(**pr) for pr in response_list if pr.get("merged_at") is not None
            ]
            all_prs.extend(merged_prs)
            # Check Link header for pagination info
            link_header = response.headers.get("Link")
            if link_header and 'rel="next"' not in link_header:
                logfire.info("Reached last page based on Link header.")
                break
            page += 1
    logfire.info(
        f"Extracted {len(all_prs)} merged PRs from {self.repo_owner}/{self.repo_name}"
    )
    return all_prs

get_pr_files

get_pr_files(pr_number: int) -> list[FileData]

Retrieve all files modified in a specific pull request.

Fetches the list of files that were changed in the given pull request, including metadata about additions, deletions, and file status.

Parameters:

Name	Type	Description	Default
pr_number	int	The pull request number to fetch files for.	required

Returns:

Type	Description
list[FileData]	list[FileData]: List of file data objects representing all modified files.

Source code in src/swebenchv2/datamodule/github.py

def get_pr_files(self, pr_number: int) -> list[FileData]:
    """Retrieve all files modified in a specific pull request.

    Fetches the list of files that were changed in the given pull request,
    including metadata about additions, deletions, and file status.

    Args:
        pr_number (int): The pull request number to fetch files for.

    Returns:
        list[FileData]: List of file data objects representing all modified files.
    """
    with httpx.Client(base_url=self.base_url, headers=self.headers, timeout=10) as client:
        logfire.info(f"Fetching files for PR #{pr_number}")
        response = client.get(
            url=f"/repos/{self.repo_owner}/{self.repo_name}/pulls/{pr_number}/files"
        )
        if response.status_code != 200:
            logfire.error(f"Error fetching PR files: {response.status_code}")
            return []
        return [FileData(**file) for file in response.json()]

get_file_content

get_file_content(file_path: str, sha: str) -> str

Retrieve the content of a file at a specific commit SHA.

Fetches the raw content of a file from the repository at the specified commit, handling base64 decoding when necessary.

Parameters:

Name	Type	Description	Default
file_path	str	Path to the file within the repository.	required
sha	str	Git commit SHA to retrieve the file content from.	required

Returns:

Name	Type	Description
str	str	The decoded file content as a string, or empty string if not found.

Source code in src/swebenchv2/datamodule/github.py

def get_file_content(self, file_path: str, sha: str) -> str:
    """Retrieve the content of a file at a specific commit SHA.

    Fetches the raw content of a file from the repository at the specified
    commit, handling base64 decoding when necessary.

    Args:
        file_path (str): Path to the file within the repository.
        sha (str): Git commit SHA to retrieve the file content from.

    Returns:
        str: The decoded file content as a string, or empty string if not found.
    """
    with httpx.Client(base_url=self.base_url, headers=self.headers, timeout=10) as client:
        response = client.get(
            url=f"/repos/{self.repo_owner}/{self.repo_name}/contents/{file_path}",
            params={"ref": sha},
        )

        if response.status_code != 200:
            return ""

        file_info: dict[str, Any] = response.json()
        if file_info.get("encoding") == "base64":
            content = base64.b64decode(file_info["content"]).decode("utf-8")
            return content
        return file_info.get("content", "")

extract_pr_data

extract_pr_data(pr_info: PullRequest) -> TrainingData

Extract complete training data for a single pull request.

Processes a pull request to gather all modified files and their content before and after changes, creating a structured training data object.

Parameters:

Name	Type	Description	Default
pr_info	PullRequest	Pull request object containing metadata and references.	required

Returns:

Name	Type	Description
TrainingData	TrainingData	Complete training data including PR info, formatted question, and file changes.

Source code in src/swebenchv2/datamodule/github.py

def extract_pr_data(self, pr_info: PullRequest) -> TrainingData:
    """Extract complete training data for a single pull request.

    Processes a pull request to gather all modified files and their content
    before and after changes, creating a structured training data object.

    Args:
        pr_info (PullRequest): Pull request object containing metadata and references.

    Returns:
        TrainingData: Complete training data including PR info, formatted question, and file changes.
    """
    # Get PR modified files
    files_data: list[FileData] = self.get_pr_files(pr_number=pr_info.number)

    for file_info in files_data:
        if file_info.status == "removed":
            # File was deleted
            file_info.before_edit = self.get_file_content(
                file_path=file_info.filename, sha=pr_info.base.sha
            )
        elif file_info.status == "added":
            # File was added
            file_info.after_edit = self.get_file_content(
                file_path=file_info.filename, sha=pr_info.head.sha
            )
        else:
            # File was modified
            file_info.before_edit = self.get_file_content(
                file_path=file_info.filename, sha=pr_info.base.sha
            )
            file_info.after_edit = self.get_file_content(
                file_path=file_info.filename, sha=pr_info.head.sha
            )

    # Build training data
    question = f"PR#{pr_info.number}\nTitle:\n{pr_info.title}\nDescription:\n{pr_info.body}"
    training_data = TrainingData(pr_info=pr_info, question=question, files=files_data)
    return training_data

extract_all_pr_data

extract_all_pr_data(save_json: bool) -> ExtractionResult

Extract training data from all merged pull requests in the repository.

Orchestrates the complete extraction process by fetching all merged PRs and processing each one to create comprehensive training datasets.

Parameters:

Name	Type	Description	Default
save_json	bool	Whether to save the extraction results to a JSON file.	required

Returns:

Name	Type	Description
ExtractionResult	ExtractionResult	Complete extraction results with all PR training data and metadata.

Source code in src/swebenchv2/datamodule/github.py

def extract_all_pr_data(self, save_json: bool) -> ExtractionResult:
    """Extract training data from all merged pull requests in the repository.

    Orchestrates the complete extraction process by fetching all merged PRs
    and processing each one to create comprehensive training datasets.

    Args:
        save_json (bool): Whether to save the extraction results to a JSON file.

    Returns:
        ExtractionResult: Complete extraction results with all PR training data and metadata.
    """
    logfire.info(f"Extracting data from {self.repo_owner}/{self.repo_name}")
    # Get all merged PRs
    merged_prs = self.get_merged_prs()
    logfire.info(f"Found {len(merged_prs)} merged PRs")

    # Extract detailed data for each PR
    all_training_data: list[TrainingData] = []

    for pr_info in merged_prs:
        pr_data = self.extract_pr_data(pr_info=pr_info)
        all_training_data.append(pr_data)

    data = ExtractionResult(
        repository=f"{self.repo_owner}/{self.repo_name}",
        extracted_at=datetime.now().isoformat(),
        total_prs=len(all_training_data),
        prs=all_training_data,
    )
    if save_json:
        data.save_log()
    return data

AsyncGitHubPRExtractor

Bases: GitHubPRExtractorBase

Methods:

Name	Description
get_rate_limit	Retrieve current GitHub API rate limit information asynchronously.
get_merged_prs	Fetch all merged pull requests from the GitHub repository asynchronously.
get_pr_files	Retrieve all files modified in a specific pull request asynchronously.
get_file_content	Retrieve the content of a file at a specific commit SHA asynchronously.
extract_pr_data	Extract complete training data for a single pull request asynchronously.
extract_all_pr_data	Extract training data from all merged pull requests in the repository asynchronously.

repo_url

repo_url: str = Field(..., title='Github Repository URL', description='This should be a full url to the repository, e.g. `https://github.com/Mai0313/SWEBenchV2` or `Mai0313/SWEBenchV2`', frozen=False, deprecated=False)

max_page

max_page: int | None = Field(default=None, title='Max Page', description='Maximum number of pages to fetch for PRs', frozen=False, deprecated=False)

per_page

per_page: int | None = Field(default=None, title='Per Page', description='Number of PRs to fetch per page', frozen=False, deprecated=False)

token

token: str | None = Field(default=None, validation_alias='GITHUB_TOKEN', description='GitHub API token for authentication', frozen=False, deprecated=False)

base_url

base_url: str = Field(default='https://api.github.com', validation_alias='GITHUB_API_BASE_URL', description='Base URL for GitHub API', frozen=False, deprecated=False)

repo_owner

repo_owner: str

repo_name

repo_name: str

headers

headers: dict[str, Any]

get_rate_limit

get_rate_limit() -> RateLimit

Retrieve current GitHub API rate limit information asynchronously.

Makes an asynchronous request to GitHub's rate limit endpoint to check current usage and remaining quota for API calls.

Returns:

Name	Type	Description
RateLimit	RateLimit	Current rate limit status including remaining calls and reset time.

Source code in src/swebenchv2/datamodule/github.py

async def get_rate_limit(self) -> RateLimit:
    """Retrieve current GitHub API rate limit information asynchronously.

    Makes an asynchronous request to GitHub's rate limit endpoint to check
    current usage and remaining quota for API calls.

    Returns:
        RateLimit: Current rate limit status including remaining calls and reset time.
    """
    async with httpx.AsyncClient(
        base_url=self.base_url, headers=self.headers, timeout=10
    ) as client:
        response = await client.get("/rate_limit")
        rate_limit = RateLimit(**response.json())
        logfire.info("Rate limit info", **rate_limit.rate.model_dump())
        return rate_limit

get_merged_prs

get_merged_prs() -> list[PullRequest]

Fetch all merged pull requests from the GitHub repository asynchronously.

Asynchronously retrieves all merged pull requests by paginating through the GitHub API, handling rate limits and filtering for merged PRs only.

Returns:

Type	Description
list[PullRequest]	list[PullRequest]: List of all merged pull request objects from the repository.

Source code in src/swebenchv2/datamodule/github.py

async def get_merged_prs(self) -> list[PullRequest]:
    """Fetch all merged pull requests from the GitHub repository asynchronously.

    Asynchronously retrieves all merged pull requests by paginating through
    the GitHub API, handling rate limits and filtering for merged PRs only.

    Returns:
        list[PullRequest]: List of all merged pull request objects from the repository.
    """
    all_prs: list[PullRequest] = []
    page = 1

    async with httpx.AsyncClient(
        base_url=self.base_url, headers=self.headers, timeout=10
    ) as client:
        while True:
            if page > self._max_page:
                break

            response = await client.get(
                url=f"/repos/{self.repo_owner}/{self.repo_name}/pulls",
                params={
                    "state": "closed",
                    "sort": "updated",
                    "direction": "desc",
                    "per_page": self._per_page,
                    "page": page,
                },
            )

            if response.status_code == 403:
                reset_time = int(response.headers.get("X-RateLimit-Reset", 0))
                if reset_time:
                    wait_time = reset_time - int(time.time()) + 1
                    logfire.info(f"Rate limit exceeded. Waiting {wait_time} seconds...")
                    await asyncio.sleep(wait_time)
                    continue

            if response.status_code != 200:
                logfire.error(f"Error fetching PRs: {response.status_code}")
                break

            response_list: list[dict[str, Any]] = response.json()

            # Check if we've reached the end of available data
            if not response_list:
                logfire.info(f"No more PRs found at page {page}. Stopping pagination.")
                break

            # Only keep merged PRs
            merged_prs = [
                PullRequest(**pr) for pr in response_list if pr.get("merged_at") is not None
            ]
            all_prs.extend(merged_prs)
            # Check Link header for pagination info
            link_header = response.headers.get("Link")
            if link_header and 'rel="next"' not in link_header:
                logfire.info("Reached last page based on Link header.")
                break
            page += 1
    logfire.info(
        f"Extracted {len(all_prs)} merged PRs from {self.repo_owner}/{self.repo_name}"
    )
    return all_prs

get_pr_files

get_pr_files(pr_number: int) -> list[FileData]

Retrieve all files modified in a specific pull request asynchronously.

Asynchronously fetches the list of files that were changed in the given pull request, including metadata about additions, deletions, and file status.

Parameters:

Name	Type	Description	Default
pr_number	int	The pull request number to fetch files for.	required

Returns:

Type	Description
list[FileData]	list[FileData]: List of file data objects representing all modified files.

Source code in src/swebenchv2/datamodule/github.py

async def get_pr_files(self, pr_number: int) -> list[FileData]:
    """Retrieve all files modified in a specific pull request asynchronously.

    Asynchronously fetches the list of files that were changed in the given pull request,
    including metadata about additions, deletions, and file status.

    Args:
        pr_number (int): The pull request number to fetch files for.

    Returns:
        list[FileData]: List of file data objects representing all modified files.
    """
    async with httpx.AsyncClient(
        base_url=self.base_url, headers=self.headers, timeout=10
    ) as client:
        logfire.info(f"Fetching files for PR #{pr_number}")
        response = await client.get(
            url=f"/repos/{self.repo_owner}/{self.repo_name}/pulls/{pr_number}/files"
        )
        if response.status_code != 200:
            logfire.error(f"Error fetching PR files: {response.status_code}")
            return []
        return [FileData(**file) for file in response.json()]

get_file_content

get_file_content(file_path: str, sha: str) -> str

Retrieve the content of a file at a specific commit SHA asynchronously.

Asynchronously fetches the raw content of a file from the repository at the specified commit, handling base64 decoding when necessary.

Parameters:

Name	Type	Description	Default
file_path	str	Path to the file within the repository.	required
sha	str	Git commit SHA to retrieve the file content from.	required

Returns:

Name	Type	Description
str	str	The decoded file content as a string, or empty string if not found.

Source code in src/swebenchv2/datamodule/github.py

async def get_file_content(self, file_path: str, sha: str) -> str:
    """Retrieve the content of a file at a specific commit SHA asynchronously.

    Asynchronously fetches the raw content of a file from the repository at the specified
    commit, handling base64 decoding when necessary.

    Args:
        file_path (str): Path to the file within the repository.
        sha (str): Git commit SHA to retrieve the file content from.

    Returns:
        str: The decoded file content as a string, or empty string if not found.
    """
    async with httpx.AsyncClient(
        base_url=self.base_url, headers=self.headers, timeout=10
    ) as client:
        response = await client.get(
            url=f"/repos/{self.repo_owner}/{self.repo_name}/contents/{file_path}",
            params={"ref": sha},
        )

        if response.status_code != 200:
            return ""

        file_info: dict[str, Any] = response.json()
        if file_info.get("encoding") == "base64":
            content = base64.b64decode(file_info["content"]).decode("utf-8")
            return content
        return file_info.get("content", "")

extract_pr_data

extract_pr_data(pr_info: PullRequest) -> TrainingData

Extract complete training data for a single pull request asynchronously.

Asynchronously processes a pull request to gather all modified files and their content before and after changes, using concurrent requests for optimal performance.

Parameters:

Name	Type	Description	Default
pr_info	PullRequest	Pull request object containing metadata and references.	required

Returns:

Name	Type	Description
TrainingData	TrainingData	Complete training data including PR info, formatted question, and file changes.

Source code in src/swebenchv2/datamodule/github.py

async def extract_pr_data(self, pr_info: PullRequest) -> TrainingData:
    """Extract complete training data for a single pull request asynchronously.

    Asynchronously processes a pull request to gather all modified files and their content
    before and after changes, using concurrent requests for optimal performance.

    Args:
        pr_info (PullRequest): Pull request object containing metadata and references.

    Returns:
        TrainingData: Complete training data including PR info, formatted question, and file changes.
    """
    # Get PR modified files
    files_data: list[FileData] = await self.get_pr_files(pr_number=pr_info.number)

    semaphore = asyncio.Semaphore(10)

    async def get_file_content_with_semaphore(file_info: FileData) -> None:
        async with semaphore:
            if file_info.status == "removed":
                # File was deleted
                file_info.before_edit = await self.get_file_content(
                    file_path=file_info.filename, sha=pr_info.base.sha
                )
            elif file_info.status == "added":
                # File was added
                file_info.after_edit = await self.get_file_content(
                    file_path=file_info.filename, sha=pr_info.head.sha
                )
            else:
                # File was modified - need to get both before and after content
                before_task = self.get_file_content(
                    file_path=file_info.filename, sha=pr_info.base.sha
                )
                after_task = self.get_file_content(
                    file_path=file_info.filename, sha=pr_info.head.sha
                )
                file_info.before_edit, file_info.after_edit = await asyncio.gather(
                    before_task, after_task
                )

    # Execute all file content fetching tasks concurrently
    tasks = [get_file_content_with_semaphore(file_info) for file_info in files_data]
    await asyncio.gather(*tasks)

    # Build training data
    question = f"PR#{pr_info.number}\nTitle:\n{pr_info.title}\nDescription:\n{pr_info.body}"
    training_data = TrainingData(pr_info=pr_info, question=question, files=files_data)
    return training_data

extract_all_pr_data

extract_all_pr_data(save_json: bool) -> ExtractionResult

Extract training data from all merged pull requests in the repository asynchronously.

Orchestrates the complete extraction process by fetching all merged PRs and processing them concurrently to create comprehensive training datasets with optimal performance through async operations.

Parameters:

Name	Type	Description	Default
save_json	bool	Whether to save the extraction results to a JSON file.	required

Returns:

Name	Type	Description
ExtractionResult	ExtractionResult	Complete extraction results with all PR training data and metadata.

Source code in src/swebenchv2/datamodule/github.py

async def extract_all_pr_data(self, save_json: bool) -> ExtractionResult:
    """Extract training data from all merged pull requests in the repository asynchronously.

    Orchestrates the complete extraction process by fetching all merged PRs
    and processing them concurrently to create comprehensive training datasets
    with optimal performance through async operations.

    Args:
        save_json (bool): Whether to save the extraction results to a JSON file.

    Returns:
        ExtractionResult: Complete extraction results with all PR training data and metadata.
    """
    merged_prs = await self.get_merged_prs()
    logfire.info(f"Found {len(merged_prs)} merged PRs from {self.repo_owner}/{self.repo_name}")

    semaphore = asyncio.Semaphore(5)

    async def extract_single_pr(pr_info: PullRequest) -> TrainingData:
        async with semaphore:
            return await self.extract_pr_data(pr_info=pr_info)

    # Process all PRs concurrently
    tasks = [extract_single_pr(pr_info) for pr_info in merged_prs]
    all_training_data: list[TrainingData] = await asyncio.gather(*tasks)
    logfire.info(f"Successfully extracted {len(all_training_data)} PR datasets")

    data = ExtractionResult(
        repository=f"{self.repo_owner}/{self.repo_name}",
        extracted_at=datetime.now().isoformat(),
        total_prs=len(all_training_data),
        prs=all_training_data,
    )
    if save_json:
        await data.a_save_log()
    return data