diffrag 0.2.3__tar.gz → 0.3.0__tar.gz

{diffrag-0.2.3 → diffrag-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: diffrag
-Version: 0.2.3
+Version: 0.3.0
 Summary: AI-powered git diff analysis and code review with RAG context
 Keywords: code-review,git,ai,llm,rag,openai
 Author: Eduardo Nunez

{diffrag-0.2.3 → diffrag-0.3.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "diffrag"
-version = "0.2.3"
+version = "0.3.0"
 description = "AI-powered git diff analysis and code review with RAG context"
 readme = "README.md"
 license = { file = "LICENSE" }

{diffrag-0.2.3 → diffrag-0.3.0}/src/diffrag/ai/openai_compat.py

@@ -1,7 +1,9 @@
 """OpenAI-compatible REST client (works with OpenWebUI, Ollama /v1, OpenAI, etc.)."""
 
+import asyncio
 import json
 import logging
+import random
 
 import httpx
 
@@ -10,6 +12,9 @@ from .base import ChatMessage, CompletionResponse, EmbeddingResponse
 
 log = logging.getLogger(__name__)
 
+_MAX_RETRIES = 4
+_RETRY_BASE_DELAY = 1.0
+
 
 class OpenAICompatClient:
     """HTTP client for any OpenAI-compatible ``/chat/completions`` + ``/embeddings`` API.
@@ -86,37 +91,55 @@ class OpenAICompatClient:
         content = ""
         model = self._model
         prompt_tokens = completion_tokens = 0
-        try:
-            async with (
-                httpx.AsyncClient(timeout=self._timeout) as client,
-                client.stream("POST", url, json=payload, headers=self._headers()) as response,
-            ):
-                if not response.is_success:
-                    await response.aread()
-                response.raise_for_status()
-                async for line in response.aiter_lines():
-                    if not line.startswith("data: "):
-                        continue
-                    chunk_str = line[6:]
-                    if chunk_str == "[DONE]":
-                        break
-                    try:
-                        chunk = json.loads(chunk_str)
-                    except json.JSONDecodeError:
-                        continue
-                    model = chunk.get("model", model)
-                    for choice in chunk.get("choices", []):
-                        content += choice.get("delta", {}).get("content", "")
-                    if usage := chunk.get("usage"):
-                        prompt_tokens = usage.get("prompt_tokens", 0)
-                        completion_tokens = usage.get("completion_tokens", 0)
-        except httpx.HTTPStatusError as exc:
-            raise AIClientError(
-                f"API request to {url} failed with status {exc.response.status_code}: "
-                f"{exc.response.text[:200]}"
-            ) from exc
-        except httpx.RequestError as exc:
-            raise AIClientError(f"Network error reaching {url}: {exc}") from exc
+        for attempt in range(_MAX_RETRIES + 1):
+            content = ""
+            model = self._model
+            prompt_tokens = completion_tokens = 0
+            try:
+                async with (
+                    httpx.AsyncClient(timeout=self._timeout) as client,
+                    client.stream("POST", url, json=payload, headers=self._headers()) as response,
+                ):
+                    if not response.is_success:
+                        await response.aread()
+                    response.raise_for_status()
+                    async for line in response.aiter_lines():
+                        if not line.startswith("data: "):
+                            continue
+                        chunk_str = line[6:]
+                        if chunk_str == "[DONE]":
+                            break
+                        try:
+                            chunk = json.loads(chunk_str)
+                        except json.JSONDecodeError:
+                            continue
+                        model = chunk.get("model", model)
+                        for choice in chunk.get("choices", []):
+                            content += choice.get("delta", {}).get("content", "")
+                        if usage := chunk.get("usage"):
+                            prompt_tokens = usage.get("prompt_tokens", 0)
+                            completion_tokens = usage.get("completion_tokens", 0)
+                break  # success
+            except httpx.HTTPStatusError as exc:
+                if exc.response.status_code == 429 and attempt < _MAX_RETRIES:
+                    delay = float(
+                        exc.response.headers.get("retry-after", _RETRY_BASE_DELAY * 2**attempt)
+                    )
+                    delay += random.uniform(0, delay * 0.1)
+                    log.warning(
+                        "Rate-limited (429). Waiting %.1f s before retry %d/%d",
+                        delay,
+                        attempt + 1,
+                        _MAX_RETRIES,
+                    )
+                    await asyncio.sleep(delay)
+                else:
+                    raise AIClientError(
+                        f"API request to {url} failed with status {exc.response.status_code}: "
+                        f"{exc.response.text[:200]}"
+                    ) from exc
+            except httpx.RequestError as exc:
+                raise AIClientError(f"Network error reaching {url}: {exc}") from exc
 
         return CompletionResponse(
             content=content,

{diffrag-0.2.3 → diffrag-0.3.0}/src/diffrag/cli/main.py

@@ -97,6 +97,7 @@ def _build_reviewer(
         similarity_top_k=settings.review.similarity_top_k,
         max_prompt_tokens=settings.review.max_prompt_tokens,
         verbosity=verbosity,
+        concurrency=settings.review.concurrency,
     )
     return reviewer, None
 

{diffrag-0.2.3 → diffrag-0.3.0}/src/diffrag/config/settings.py

@@ -117,18 +117,25 @@ class ReviewSettings:
         max_prompt_tokens: Approximate token budget per diff chunk before splitting.
         verbosity: Detail level of generated reviews. One of ``"brief"``,
             ``"standard"`` (default), or ``"detailed"``.
+        concurrency: Maximum number of file reviews sent to the AI concurrently.
+            Higher values reduce wall-clock time but increase peak API load.
     """
 
     base_branch: str = "main"
     similarity_top_k: int = 5
     max_prompt_tokens: int = 6000
     verbosity: str = "standard"
+    concurrency: int = 3
 
     def __post_init__(self) -> None:
         if self.verbosity not in ("brief", "standard", "detailed"):
             raise ConfigurationError(
                 f"Invalid verbosity {self.verbosity!r}. Choose 'brief', 'standard', or 'detailed'."
             )
+        if self.concurrency < 1:
+            raise ConfigurationError(
+                f"Invalid concurrency {self.concurrency!r}. Must be 1 or greater."
+            )
 
 
 @dataclass

{diffrag-0.2.3 → diffrag-0.3.0}/src/diffrag/data/config.example.toml

@@ -51,3 +51,7 @@ similarity_top_k = 5
 max_prompt_tokens = 6000
 # Detail level of generated reviews: "brief" | "standard" | "detailed"
 verbosity = "standard"
+# Maximum number of file reviews sent to the AI simultaneously.
+# Higher values reduce wall-clock time; lower values reduce peak API load.
+# Set to 1 to restore the original sequential behaviour.
+concurrency = 3

{diffrag-0.2.3 → diffrag-0.3.0}/src/diffrag/review/reviewer.py

@@ -1,5 +1,6 @@
 """CodeReviewer: orchestrates diff extraction, RAG retrieval, and LLM review."""
 
+import asyncio
 import logging
 from collections.abc import Callable
 from dataclasses import dataclass, field
@@ -79,6 +80,7 @@ class CodeReviewer:
         indexer: Vector index for context retrieval.
         similarity_top_k: Number of context chunks retrieved per diff chunk.
         max_prompt_tokens: Approximate token budget per diff before hunk-splitting.
+        concurrency: Maximum number of file reviews sent to the AI simultaneously.
     """
 
     def __init__(
@@ -88,12 +90,14 @@ class CodeReviewer:
         similarity_top_k: int = 5,
         max_prompt_tokens: int = 6000,
         verbosity: str = "standard",
+        concurrency: int = 3,
     ) -> None:
         self._ai = ai_client
         self._indexer = indexer
         self._top_k = similarity_top_k
         self._max_prompt_tokens = max_prompt_tokens
         self._verbosity = verbosity
+        self._concurrency = concurrency
 
     # ------------------------------------------------------------------
     # Public API
@@ -157,13 +161,28 @@ class CodeReviewer:
                 repo.path, on_progress=on_progress, commit_hash=commit_hash
             )
 
+        completed = 0
+        sem = asyncio.Semaphore(self._concurrency)
+
+        async def _bounded(fd: FileDiff) -> list[FileReview]:
+            nonlocal completed
+            async with sem:
+                _p(f"Reviewing {fd.path}...")
+                log.info("Reviewing %s (%s)", fd.path, fd.status)
+                result = await self._review_file(fd)
+                completed += 1
+                _p(f"Reviewed {fd.path} ({completed}/{n})")
+                return result
+
+        gathered = await asyncio.gather(
+            *[_bounded(fd) for fd in diff_result.file_diffs],
+            return_exceptions=True,
+        )
         file_reviews: list[FileReview] = []
-        for i, file_diff in enumerate(diff_result.file_diffs, 1):
-            _p(f"Reviewing {file_diff.path} ({i}/{n})...")
-            log.info("Reviewing %s (%s)", file_diff.path, file_diff.status)
-            reviews = await self._review_file(file_diff)
-            file_reviews.extend(reviews)
-            _p(f"Reviewed {file_diff.path} ({i}/{n})")
+        for r in gathered:
+            if isinstance(r, BaseException):
+                raise r
+            file_reviews.extend(r)
 
         if n > 1:
             _p("Consolidating review...")