Align Multi-SWE patch extraction with upstream (#1523)

* Align Multi-SWE patch extraction with upstream

* Group Multi-SWE taskset files

Author: rasdani

verifiers/envs/experimental/composable/tasksets/swe/create_fix_patch.sh

@@ -0,0 +1,3 @@
+from .taskset import MultiSWERubric, MultiSWETaskSet, restore_row
+
+__all__ = ["MultiSWERubric", "MultiSWETaskSet", "restore_row"]

verifiers/envs/experimental/composable/tasksets/swe/multi_swe/__init__.py

@@ -0,0 +1,55 @@
+#!/bin/bash
+
+# Generate /home/fix.patch from the current working-tree diff using the same
+# fix_patch/test_patch path split as upstream Multi-SWE dataset construction.
+
+set -euo pipefail
+
+REPO_DIR="${1:-$(pwd)}"
+BASE_COMMIT="${2:-HEAD}"
+OUTPUT_PATH="${3:-/home/fix.patch}"
+cd "$REPO_DIR"
+
+if [ -z "$BASE_COMMIT" ]; then
+    BASE_COMMIT="HEAD"
+fi
+
+FIX_FILES=()
+TEST_FILES=()
+UNTRACKED_FILES=()
+
+while IFS= read -r -d '' file; do
+    UNTRACKED_FILES+=("$file")
+done < <(git ls-files --others --exclude-standard -z)
+
+if [ "${#UNTRACKED_FILES[@]}" -gt 0 ]; then
+    git add --intent-to-add -- "${UNTRACKED_FILES[@]}"
+fi
+
+while IFS= read -r -d '' file; do
+    lowered="$(printf '%s' "$file" | tr '[:upper:]' '[:lower:]')"
+    if [[ "$lowered" == *"test"* ]] || [[ "$lowered" == *"tests"* ]] || [[ "$lowered" == *"e2e"* ]] || [[ "$lowered" == *"testing"* ]]; then
+        TEST_FILES+=("$file")
+    else
+        FIX_FILES+=("$file")
+    fi
+done < <(git diff --name-only -z "$BASE_COMMIT" --)
+
+if [ "${#FIX_FILES[@]}" -eq 0 ]; then
+    : > "$OUTPUT_PATH"
+else
+    git diff --binary "$BASE_COMMIT" -- "${FIX_FILES[@]}" > "$OUTPUT_PATH"
+fi
+
+if [ "${#TEST_FILES[@]}" -gt 0 ]; then
+    printf 'Dropped %d test-like file(s) from Multi-SWE fix patch.\n' "${#TEST_FILES[@]}" >&2
+fi
+
+if [ ! -s "$OUTPUT_PATH" ]; then
+    : > "$OUTPUT_PATH"
+fi
+
+git reset --hard "$BASE_COMMIT" >/dev/null
+if [ "${#UNTRACKED_FILES[@]}" -gt 0 ]; then
+    git clean -fd -- "${UNTRACKED_FILES[@]}" >/dev/null
+fi

verifiers/envs/experimental/composable/tasksets/swe/multi_swe/extract_fix_patch.sh

@@ -261,7 +261,7 @@ async def setup(self, state) -> None:
             # Acquire::Retries=3: harden against transient archive.ubuntu.com CDN
             # mirror-sync mismatches mid-rollout (launchpad bug #1876035).
             "command -v patch || (apt-get -o Acquire::Retries=3 update && apt-get -o Acquire::Retries=3 install -y patch)",
-            "rm -f /home/fix.patch /home/test_output.txt /home/create_fix_patch.sh",
+            "rm -f /home/fix.patch /home/test_output.txt /home/create_fix_patch.sh /home/extract_multiswe_fix_patch.sh",
         ]
         for command in commands:
             results = await sandbox_client.execute_command(
@@ -281,21 +281,26 @@ async def _run_tests(
         test_timeout: int,
     ) -> str:
         repo_path = self.get_workdir(state["info"])
-        script_path = SCRIPTS_DIR / "create_fix_patch.sh"
+        multiswe_row = restore_row(state["info"])
+        base = multiswe_row.get("base")
+        base_commit = base.get("sha") if isinstance(base, dict) else None
+        base_commit = base_commit or "HEAD"
+        script_path = SCRIPTS_DIR / "extract_fix_patch.sh"
         await sandbox_client.upload_file(
-            sandbox_id, "/home/create_fix_patch.sh", str(script_path)
+            sandbox_id, "/home/extract_multiswe_fix_patch.sh", str(script_path)
         )
 
         prep = await sandbox_client.execute_command(
             sandbox_id,
-            "chmod +x /home/create_fix_patch.sh && bash /home/create_fix_patch.sh",
+            "chmod +x /home/extract_multiswe_fix_patch.sh"
+            f" && bash /home/extract_multiswe_fix_patch.sh . {shlex.quote(base_commit)}",
             working_dir=repo_path,
             timeout=min(test_timeout, 300),
         )
         if prep.exit_code != 0:
             stderr = (prep.stderr or "")[:1000]
             raise RuntimeError(
-                f"Failed to create Multi-SWE fix patch: exit_code={prep.exit_code} stderr={stderr}"
+                f"Failed to extract Multi-SWE fix patch: exit_code={prep.exit_code} stderr={stderr}"
             )
 
         command = "bash -o pipefail -c 'bash /home/fix-run.sh 2>&1 | tee /home/test_output.txt'"