| """Shared diff-to-patch-lines logic (spec §3.2's `patch_lines_vuln`), used by |
| every source loader so patch line numbers are always relative to the |
| extracted function body itself — never to the original file's absolute line |
| numbers. See DECISIONS.md "SecVulEval patch-line numbers were file-absolute, |
| not function-relative" for why this needed to be a shared, source-agnostic |
| computation rather than trusting any one dataset's own line-number field. |
| """ |
| from __future__ import annotations |
|
|
| import difflib |
|
|
|
|
| def diff_and_patch_lines(func_vuln: str, func_patch: str) -> tuple[str, list[int]]: |
| """Unified diff of the function pair + 1-based line numbers in func_vuln |
| that were removed/changed by the patch (spec §3.2 `patch_lines_vuln`). |
| |
| A pure insertion (e.g. a missing null-check added, CWE-476/CWE-20 style |
| fixes) has no line in func_vuln that was literally changed, but the |
| patch location is still meaningful for localization: we anchor to the |
| line immediately preceding the insertion point (1-based), or line 1 if |
| the insertion is at the very top. |
| """ |
| vuln_lines = func_vuln.splitlines(keepends=True) |
| patch_lines = func_patch.splitlines(keepends=True) |
| diff = "".join( |
| difflib.unified_diff(vuln_lines, patch_lines, fromfile="vuln", tofile="patch") |
| ) |
| changed: set[int] = set() |
| sm = difflib.SequenceMatcher(a=vuln_lines, b=patch_lines, autojunk=False) |
| for tag, i1, i2, _j1, _j2 in sm.get_opcodes(): |
| if tag in ("replace", "delete"): |
| changed.update(range(i1 + 1, i2 + 1)) |
| elif tag == "insert": |
| changed.add(max(1, i1)) |
| return diff, sorted(changed) |
|
|
