about summary refs log tree commit diff stats
diff options
context:
space:
mode:
authorAndrew Dolgov <fox@fakecake.org>2026-06-22 18:23:53 +0300
committerAndrew Dolgov <fox@fakecake.org>2026-06-22 18:23:53 +0300
commitf83b210bc4cec556b82b0a6bfa1900830455132b (patch)
tree45ee420686e32583a255925f93505e1e31838bea
parent7e515ef3a86cb63ef467bdfcbe3c52393d3a67e6 (diff)
downloadthe-epube-f83b210bc4cec556b82b0a6bfa1900830455132b.tar.gz
the-epube-f83b210bc4cec556b82b0a6bfa1900830455132b.tar.xz
more diff stuff
-rw-r--r--review_mr.py125
1 files changed, 104 insertions, 21 deletions
diff --git a/review_mr.py b/review_mr.py
index 901d6b8..acf57d6 100644
--- a/review_mr.py
+++ b/review_mr.py
@@ -61,21 +61,70 @@ def _api_post(url: str, headers: dict, payload: dict) -> dict:
 # Diff fetching
 # ---------------------------------------------------------------------------
 
+def _format_diff_with_line_numbers(diff_text: str) -> str:
+    """Format a unified diff string with absolute line numbers.
+
+    Parses @@ hunk headers to compute line numbers, then prepends them.
+    Only lines that exist in the new file (additions and context) get
+    line numbers; removal lines are left without a number.
+
+    Example output:
+        === src/App.php ===
+        @@ -10,7 +10,7 @@
+             -    return $user->name;
+           10  +    return $user->name ?? "";
+    """
+    import re as _re
+
+    lines = diff_text.split("\n")
+    hunk_re = _re.compile(r"^@@ -\d+(?:,\d+)? \+(\d+)(?:,\d+)? @@")
+
+    output = []
+    new_line = None  # current line number in the new file
+
+    for line in lines:
+        m = hunk_re.match(line)
+        if m:
+            new_line = int(m.group(1))
+            output.append(line)  # keep hunk header as-is
+            continue
+
+        if new_line is None:
+            output.append(line)
+            continue
+
+        if line.startswith("+"):
+            output.append(f"{new_line:>5}  {line}")
+            new_line += 1
+        elif line.startswith(" "):
+            output.append(f"{new_line:>5}{line}")
+            new_line += 1
+        elif line.startswith("-"):
+            # Removal — no line number in the new file
+            output.append(f"{'':>5}  {line}")
+        else:
+            # Other lines (e.g., index, new file, etc.)
+            output.append(line)
+
+    return "\n".join(output)
+
+
 def fetch_diff(project_id: str, mr_iid: str, token: str, server_url: str):
     """Fetch the MR diff and return (formatted_text, raw_diffs).
 
     raw_diffs is a list of dicts with keys: old_path, new_path, diff, position.
+    The formatted text includes file headers and line numbers for each diff hunk.
     """
     url = f"{server_url}/api/v4/projects/{project_id}/merge_requests/{mr_iid}/diffs"
     diffs = _api_get(url, {"PRIVATE-TOKEN": token})
 
-    # Build readable text with file headers for the LLM
+    # Build readable text with file headers and line numbers for the LLM
     parts = []
     for d in diffs:
         path = d.get("new_path") or d.get("old_path", "<unknown>")
         parts.append(f"=== {path} ===")
         if d.get("diff"):
-            parts.append(d["diff"])
+            parts.append(_format_diff_with_line_numbers(d["diff"]))
 
     diff_text = "\n".join(parts)[:10000]
     return diff_text, diffs
@@ -89,6 +138,9 @@ SYSTEM_PROMPT = (
     "You are a code review assistant. Review the diff below and return "
     "a JSON object with an \"annotations\" key containing an array of "
     "line annotations.\n\n"
+    "The diff is formatted with absolute line numbers in the new file "
+    "padded to 5 characters on the left. Only lines with a line number "
+    "are valid annotation targets.\n\n"
     "JSON Schema:\n"
     "```json\n"
     "{\n"
@@ -112,7 +164,8 @@ SYSTEM_PROMPT = (
     '}\n'
     "```\n\n"
     "Rules:\n"
-    "- Only annotate lines that actually exist in the diff.\n"
+    "- Use the line numbers shown in the diff output (the 5-char number on the left).\n"
+    "- Only annotate lines that have a line number prefix.\n"
     "- Be specific — reference the code you're commenting on.\n"
     "- Skip unchanged lines.\n"
     "- Return ONLY the JSON object, no markdown wrapping, no explanation."
@@ -228,17 +281,59 @@ def post_comment(
     print("Plain review comment posted.")
 
 
+def _build_positions_from_diff(diffs: list[dict], base_sha: str, head_sha: str) -> dict[tuple, dict]:
+    """Build position lookup from diff entries by parsing hunk headers.
+
+    Returns a mapping of (old_path, new_path, old_line, new_line) -> position dict.
+    """
+    import re as _re
+
+    hunk_re = _re.compile(r"^@@ -\d+(?:,\d+)? \+(\d+)(?:,\d+)? @@")
+    lookup: dict[tuple, dict] = {}
+
+    for d in diffs:
+        old_path = d.get("old_path", "")
+        new_path = d.get("new_path", old_path)
+        diff_text = d.get("diff", "")
+        if not diff_text:
+            continue
+
+        new_line = None
+        for line in diff_text.split("\n"):
+            m = hunk_re.match(line)
+            if m:
+                new_line = int(m.group(1))
+                continue
+            if new_line is None:
+                continue
+            if line.startswith("+"):
+                key = (old_path, new_path, None, new_line)
+                lookup[key] = {
+                    "base_sha": base_sha,
+                    "head_sha": head_sha,
+                    "position_type": "text",
+                }
+                new_line += 1
+            elif line.startswith(" "):
+                key = (old_path, new_path, new_line, new_line)
+                lookup[key] = {
+                    "base_sha": base_sha,
+                    "head_sha": head_sha,
+                    "position_type": "text",
+                }
+                new_line += 1
+            # removal lines (-) have no new_line, skip
+
+    return lookup
+
+
 def _fetch_position_metadata(
     project_id: str, mr_iid: str, token: str, server_url: str
 ) -> dict[tuple, dict]:
     """Fetch diff position metadata (base_sha, head_sha) for annotation positions.
 
     Returns a mapping of (old_path, new_path, old_line, new_line) -> position dict.
-    Falls back to MR-level SHA if the positions endpoint is unavailable.
     """
-    url = (
-        f"{server_url}/api/v4/projects/{project_id}/merge_requests/{mr_iid}/notes"
-    )
     lookup: dict[tuple, dict] = {}
 
     # Try the /positions endpoint first
@@ -264,29 +359,17 @@ def _fetch_position_metadata(
     except requests.HTTPError:
         pass  # Fall through to MR-level fallback
 
-    # Fallback: get MR info and use its SHA for both base and head
+    # Fallback: get MR info and parse diff text for line numbers
     mr_url = f"{server_url}/api/v4/projects/{project_id}/merge_requests/{mr_iid}"
     mr = _api_get(mr_url, {"PRIVATE-TOKEN": token})
     sha = mr.get("sha", "")
     print("  (Using MR-level SHA as fallback for position metadata)", file=sys.stderr)
 
-    # We still need old_path/new_path/line info from the diff
-    # Build minimal position entries from the diff itself
     diffs_url = (
         f"{server_url}/api/v4/projects/{project_id}/merge_requests/{mr_iid}/diffs"
     )
     diffs = _api_get(diffs_url, {"PRIVATE-TOKEN": token})
-    for d in diffs:
-        path = d.get("new_path") or d.get("old_path", "")
-        old_line = d.get("old_line")
-        new_line = d.get("new_line")
-        if path and new_line is not None:
-            key = (d.get("old_path"), path, old_line, new_line)
-            lookup[key] = {
-                "base_sha": sha,
-                "head_sha": sha,
-                "position_type": "text",
-            }
+    lookup = _build_positions_from_diff(diffs, sha, sha)
     return lookup