[SuperSize] Hide llvm-nm stderr messages during archive.

When llvm-nm finds no symbols in an object file (.o file or embedded in
in .a file) it prints 'no symbols' to stderr. This produces a lot of
noise when running 'supersize archive' from command line.

This CL makes SuperSize capture stderr for calls to llvm-nm (and nm,
although it seems to be more quiet). The number of stderr lines (all
'no symbols' as far as we see), if non-0, is returned by
nm.RunNmOnIntermediates(), and printed by obj_analyzer._RunNm() as a
warning.

Change-Id: I2090f8762614d5fc3102f0ec94ef7e83fc305bfd
Reviewed-on: https://chromium-review.googlesource.com/c/1289989Reviewed-by: Samuel Huang <huangs@chromium.org>
Reviewed-by: agrieve <agrieve@chromium.org>
Commit-Queue: Samuel Huang <huangs@chromium.org>
Cr-Commit-Position: refs/heads/master@{#601031}

tools/binary_size/libsupersize/nm.py

@@ -80,8 +80,11 @@ def CollectAliasesByAddress(elf_path, tool_prefix):
   # directly takes 3s.
   args = [path_util.GetNmPath(tool_prefix), '--no-sort', '--defined-only',
           elf_path]
-  output = subprocess.check_output(args)
-  for line in output.splitlines():
+  proc = subprocess.Popen(args, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+  # llvm-nm may write to stderr. Discard to denoise.
+  stdout, _ = proc.communicate()
+  assert proc.returncode == 0
+  for line in stdout.splitlines():
     space_idx = line.find(' ')
     address_str = line[:space_idx]
     section = line[space_idx + 1]
@@ -159,8 +162,14 @@ def RunNmOnIntermediates(target, tool_prefix, output_directory):
     args.append(target)
   else:
     args.extend(target)
-  output = subprocess.check_output(args, cwd=output_directory)
-  lines = output.splitlines()
+  proc = subprocess.Popen(args, cwd=output_directory, stdout=subprocess.PIPE,
+                          stderr=subprocess.PIPE)
+  # llvm-nm can print 'no symbols' to stderr. Capture and count the number of
+  # lines, to be returned to the caller.
+  stdout, stderr = proc.communicate()
+  assert proc.returncode == 0
+  num_no_symbols = len(stderr.splitlines())
+  lines = stdout.splitlines()
   # Empty .a file has no output.
   if not lines:
     return concurrent.EMPTY_ENCODED_DICT, concurrent.EMPTY_ENCODED_DICT
@@ -191,4 +200,5 @@ def RunNmOnIntermediates(target, tool_prefix, output_directory):
   # TODO(agrieve): We could use path indices as keys rather than paths to cut
   #     down on marshalling overhead.
   return (concurrent.EncodeDictOfLists(symbol_names_by_path),
-          concurrent.EncodeDictOfLists(string_addresses_by_path))
+          concurrent.EncodeDictOfLists(string_addresses_by_path),
+          num_no_symbols)

tools/binary_size/libsupersize/obj_analyzer.py

@@ -146,7 +146,9 @@ class _BulkObjectFileAnalyzerWorker(object):
 
     # Names are still mangled.
     all_paths_by_name = self._paths_by_name
-    for encoded_syms, encoded_strs in results:
+    total_no_symbols = 0
+    for encoded_syms, encoded_strs, num_no_symbols in results:
+      total_no_symbols += num_no_symbols
       symbol_names_by_path = concurrent.DecodeDictOfLists(encoded_syms)
       for path, names in symbol_names_by_path.iteritems():
         for name in names:
@@ -154,6 +156,8 @@ class _BulkObjectFileAnalyzerWorker(object):
 
       if encoded_strs != concurrent.EMPTY_ENCODED_DICT:
         self._encoded_string_addresses_by_path_chunks.append(encoded_strs)
+    if total_no_symbols:
+      logging.warn('nm found no symbols in %d objects.', total_no_symbols)
 
   def _RunLlvmBcAnalyzer(self, paths_by_type):
     """Calls llvm-bcanalyzer to extract string data (for LLD-LTO)."""