Work diff canvas treats removed '-- ' and added '++ ' lines as file headers #274

Expected and actual behavior / 实际现象与期望

The Work attempt diff canvas (parse_unified_diff in server/unified_diff.py, called from WorkLedgerCoordinator for attempt diffs) should render every removed and added line of the attempt.

When a removed source line starts with -- (SQL, Lua and Haskell comments, Markdown -- signatures) or an added line starts with ++ , Git prefixes them to --- ... / +++ .... The parser checks for file-header markers before looking at hunk content, so these lines are treated as file headers inside a hunk:

• the file is renamed in the canvas to the line content (e.g. path: "counter", oldPath: "legacy index");
• the line is dropped from the hunk, and the file and total additions / deletions counts are understated;
• the context line numbers after it no longer correspond to the hunk.

Steps to reproduce / 复现步骤

from server.unified_diff import parse_unified_diff

parse_unified_diff("""diff --git a/schema.sql b/schema.sql
--- a/schema.sql
+++ b/schema.sql
@@ -1,3 +1,3 @@
 CREATE TABLE t (id INT);
--- legacy index
+++ counter
 SELECT 1;
""")

Actual: one file with path == "counter", oldPath == "legacy index", additions == 0, deletions == 0, and only the two context lines in the hunk.

Expected: path == "schema.sql", one remove line (-- legacy index) and one add line (++ counter).

This is the exact output of git diff after replacing a line -- legacy index with ++ counter in a tracked file.

Relevant environment / 使用环境

Platform-independent; current main (65cea5f). Affects any Work attempt whose diff removes a line starting with -- or adds one starting with ++ .

Sanitized evidence / 错误信息或截图（请先脱敏）

Hunk headers already declare how many old/new lines follow (@@ -1,3 +1,3 @@), and the parser reads those counts into oldCount / newCount. Inside a hunk, --- / +++ can only be file headers once those counts have been consumed.

Related issues or fixes / 相关问题或修复

None found.