feat: add 4JLibs comparison tooling

Tools for comparing 4JLibs binary changes between git refs:
- compare-4jlibs.py: Extracts libs, parses symbol tables via dumpbin,
  demangles with undname, generates structured diff reports
- extract_lib.py: Extracts .obj members from COFF .lib archives
- ExportLibInfo.java: Ghidra headless script for non-LTCG object files
- list-lib-symbols.sh / compare-4jlibs.sh: Shell wrappers

tools/ghidra/extract_lib.py

@@ -0,0 +1,104 @@
+"""Extract .obj members from a COFF .lib (ar archive) into a directory.
+
+Usage:
+    python extract_lib.py <input.lib> <output_dir>
+
+Each member is written as <output_dir>/<name>.obj. The first/second linker
+members and the long-name string table are skipped.
+"""
+
+import os
+import sys
+
+
+def extract_lib(lib_path, out_dir):
+    os.makedirs(out_dir, exist_ok=True)
+
+    with open(lib_path, "rb") as f:
+        magic = f.read(8)
+        if magic != b"!<arch>\n":
+            print(f"ERROR: Not an ar archive: {lib_path}", file=sys.stderr)
+            sys.exit(1)
+
+        # Read the long-name string table if present
+        long_names = b""
+        members = []
+
+        while True:
+            pos = f.tell()
+            header = f.read(60)
+            if len(header) < 60:
+                break
+
+            raw_name = header[0:16]
+            size_str = header[48:58].decode("ascii").strip()
+            end_marker = header[58:60]
+
+            if end_marker != b"\x60\x0a":
+                print(f"WARNING: Bad end marker at offset {pos}, stopping.", file=sys.stderr)
+                break
+
+            size = int(size_str)
+            data = f.read(size)
+
+            # Pad to even boundary
+            if size % 2 != 0:
+                f.read(1)
+
+            name = raw_name.decode("ascii", errors="replace").rstrip()
+
+            # Skip first and second linker members (both named "/")
+            if name == "/":
+                continue
+
+            # Long-name string table
+            if name == "//":
+                long_names = data
+                continue
+
+            # Resolve long name references like "/26"
+            if name.startswith("/") and name[1:].isdigit():
+                offset = int(name[1:])
+                end = long_names.find(b"\x00", offset)
+                if end == -1:
+                    # Try newline-terminated (common in MSVC libs)
+                    end = long_names.find(b"\n", offset)
+                if end == -1:
+                    end = len(long_names)
+                resolved = long_names[offset:end].decode("ascii", errors="replace").rstrip("/")
+                name = resolved
+
+            # Clean the name for filesystem use
+            safe_name = name.replace("/", "_").replace("\\", "_").replace("..", "_")
+            if not safe_name.endswith(".obj"):
+                safe_name += ".obj"
+
+            members.append((safe_name, data))
+
+        # Write members
+        written = 0
+        for safe_name, data in members:
+            out_path = os.path.join(out_dir, safe_name)
+
+            # Handle duplicate names by appending a counter
+            if os.path.exists(out_path):
+                base, ext = os.path.splitext(safe_name)
+                counter = 2
+                while os.path.exists(os.path.join(out_dir, f"{base}_{counter}{ext}")):
+                    counter += 1
+                out_path = os.path.join(out_dir, f"{base}_{counter}{ext}")
+
+            with open(out_path, "wb") as out_f:
+                out_f.write(data)
+            written += 1
+
+        print(f"Extracted {written} object files from {os.path.basename(lib_path)} -> {out_dir}")
+        return written
+
+
+if __name__ == "__main__":
+    if len(sys.argv) != 3:
+        print(f"Usage: {sys.argv[0]} <input.lib> <output_dir>", file=sys.stderr)
+        sys.exit(1)
+
+    extract_lib(sys.argv[1], sys.argv[2])