Minor fixes to benchmarks and added new find_shadowed_symbols script

benchmarks/bin_symbol_size.py

@@ -24,7 +24,9 @@ def is_elf_file(filepath):
     return magic_number == b"\x7fELF"
 
 
-for root, dirs, files in itertools.chain(os.walk("/bin"), os.walk("/usr/bin")):
+for root, dirs, files in itertools.chain(
+    os.walk("/bin"), os.walk("/usr/bin"), os.walk("/lib/x86_64-linux-gnu")
+):
     for file in files:
         full_path_file = os.path.join(root, file)
         if not os.path.isfile(full_path_file) or not os.access(full_path_file, os.X_OK):

benchmarks/find_shadowed_symbols.py

@@ -0,0 +1,66 @@
+#! /usr/bin/env python3
+import os
+import pprint
+import signal
+import sys
+
+from sqlelf import elf, sql
+
+
+class TimeOutException(Exception):
+    pass
+
+
+def alarm_handler(signum, frame):
+    print("ALARM signal received")  # noqa: T201
+    raise TimeOutException()
+
+
+directory = sys.argv[1]
+
+for root, dirs, files in os.walk(directory):
+    for file in files:
+        try:
+            full_filename = os.path.join(root, file)
+            print(f"Investigating {full_filename}")  # noqa: T201
+
+            # setup the timeout
+            signal.signal(signal.SIGALRM, alarm_handler)
+            signal.alarm(10)
+
+            engine = sql.make_sql_engine(
+                [full_filename], recursive=True, cache_flags=elf.CacheFlag.SYMBOLS
+            )
+            # ignore .bss since it is symbols which are declared but have no value
+            # i need to understand them more but they tend to show up.
+            # let us also exclude the main binary and catch only interposition
+            # from shared objects
+            results = engine.execute(
+                """
+                SELECT name, elf_symbols.version, count(*) as symbol_count,
+                    GROUP_CONCAT(elf_headers.path, ':') as libraries
+                FROM elf_symbols, elf_headers
+                WHERE elf_symbols.path = elf_headers.path AND
+                      elf_headers.is_pie = 0 AND
+                      exported = TRUE AND section != '.bss'
+                GROUP BY name, elf_symbols.version
+                HAVING count(*) >= 2
+                """
+            )
+            rows = list(results)
+            rows = list(
+                filter(
+                    lambda r: "libc" not in r["libraries"]
+                    and "libm" not in r["libraries"],
+                    rows,
+                )
+            )
+            if len(rows) > 0:
+                print(f"Found {len(rows)} duplicate symbols")  # noqa: T201
+                pprint.pprint(rows)  # noqa: T203
+        except TimeOutException as ex:
+            print(ex)  # noqa: T201
+        except Exception as ex:
+            print(ex)  # noqa: T201
+        finally:
+            signal.alarm(0)

benchmarks/graph_bin_symbol_size.py

@@ -11,6 +11,7 @@
     ggplot,
     labs,
     save_as_pdf_pages,
+    scale_y_log10,
     theme_minimal,
 )
 
@@ -22,12 +23,14 @@
 
 # Convert to DataFrame
 df = pd.DataFrame(list(data_dict.items()), columns=["File", "Symbols"])
+df = df.query("Symbols > 0")
 
 # Create plot
 plot = (
     ggplot(df, aes(x="Symbols", y=after_stat("ncount")))
     + geom_histogram(bins=100, fill="skyblue", color="black", alpha=0.7)
     + labs(title="", x="Number of Symbols", y="Normalized Count")
+    + scale_y_log10(expand=(0, 0, 0.1, 0))
     + theme_minimal()
 )