[benchmarks] Start measuring native code size with Bloaty McBloatFace.

Found a bunch of low-hanging fruit.

benchmarks/native-code.sh

@@ -0,0 +1,65 @@
+#!/bin/bash
+#
+# Usage:
+#   ./native-code.sh <function name>
+
+set -o nounset
+set -o pipefail
+set -o errexit
+
+# Size profiler for binaries.
+bloaty() {
+  ~/git/other/bloaty/bloaty "$@"
+}
+
+pylibc-symbols() {
+  symbols _devbuild/py-ext/x86_64/libc.so
+}
+
+fastlex-symbols() {
+  symbols _devbuild/py-ext/x86_64/fastlex.so
+}
+
+symbols() {
+  local obj=$1
+  nm $obj
+  echo
+
+  bloaty $obj
+  echo
+
+  # fastlex_MatchToken is 21.2 KiB.  That doesn't seem to large compared ot
+  # the 14K line output?
+  bloaty -d symbols $obj
+  echo
+
+  ls -l $obj
+  echo
+}
+
+# Big functions:
+# - PyEval_EvalFrameEx (38 KiB)
+# - fastlex_MatchOSHToken (22.5 KiB)
+# - convertitem() in args.py (9.04 KiB)
+# - PyString_Format() in args.py (6.84 KiB)
+#
+# Easy removals:
+# - marshal_dumps and marshal_dump!  We never use those.
+# - Remove all docstrings!!!  Like sys_doc.
+
+cpython-bloat() {
+  # ovm-opt.stripped doesn't show a report.
+  local file=_build/oil/ovm-opt
+
+  # Slightly different.
+  #local file=_build/oil/ovm-dbg
+
+  #bloaty -n 30 -d symbols $file
+
+  # Full output
+  # 3,588 lines!
+  bloaty --tsv -n 0 -d symbols $file
+}
+
+
+"$@"

build/codegen.sh

@@ -100,26 +100,6 @@ ast-id-lex() {
   osh-lex-gen-native
 }
 
-# Size profiler for binaries.  TODO: Fold this into benchmarks/
-bloaty() { ~/git/other/bloaty/bloaty "$@"; }
-
-stats() {
-  local obj=_devbuild/py-ext/x86_64/fastlex.so
-  nm $obj
-  echo
-
-  bloaty $obj
-  echo
-
-  # fastlex_MatchToken is 21.2 KiB.  That doesn't seem to large compared ot
-  # the 14K line output?
-  bloaty -d symbols $obj
-  echo
-
-  ls -l $obj
-  echo
-}
-
 # NOTES:
 # - core/id_kind_gen.py generates the mapping from Id to Kind.
 #   - It needs a mapping output by the Python superoptimizatio script.