genv8constants.py

#!/usr/bin/env python

#
# genv8constants.py output_file libv8_base.a
#
# Emits v8dbg constants stored in libv8_base.a in a format suitable for the V8
# ustack helper.
#

import re
import subprocess
import sys
import errno

if len(sys.argv) != 3:
  print "usage: objsym.py outfile libv8_base.a"
  sys.exit(2);

outfile = file(sys.argv[1], 'w');
try:
  pipe = subprocess.Popen([ 'objdump', '-z', '-D', sys.argv[2] ],
      bufsize=-1, stdout=subprocess.PIPE).stdout;
except OSError, e:
  if e.errno == errno.ENOENT:
    print '''
      Node.js compile error: could not find objdump

      Check that GNU binutils are installed and included in PATH
      '''
  else:
    print 'problem running objdump: ', e.strerror

  sys.exit()

pattern = re.compile('([0-9a-fA-F]{8}|[0-9a-fA-F]{16}) <(.*)>:');
v8dbg = re.compile('^v8dbg.*$')
numpattern = re.compile('^[0-9a-fA-F]{2} $');
octets = 4

outfile.write("""
/*
 * File automatically generated by genv8constants. Do not edit.
 *
 * The following offsets are dynamically from libv8_base.a.  See src/v8ustack.d
 * for details on how these values are used.
 */

#ifndef V8_CONSTANTS_H
#define V8_CONSTANTS_H

""");

curr_sym = None;
curr_val = 0;
curr_octet = 0;

def out_reset():
  global curr_sym, curr_val, curr_octet
  curr_sym = None;
  curr_val = 0;
  curr_octet = 0;

def out_define():
  global curr_sym, curr_val, curr_octet, outfile, octets
  if curr_sym != None:
    wrapped_val = curr_val & 0xffffffff;
    if curr_val & 0x80000000 != 0:
      wrapped_val = 0x100000000 - wrapped_val;
      outfile.write("#define %s -0x%x\n" % (curr_sym.upper(), wrapped_val));
    else:
      outfile.write("#define %s 0x%x\n" % (curr_sym.upper(), wrapped_val));
  out_reset();

for line in pipe:
  if curr_sym != None:
    #
    # This bit of code has nasty knowledge of the objdump text output
    # format, but this is the most obvious robust approach.  We could almost
    # rely on looking at numbered fields, but some instructions look very
    # much like hex numbers (e.g., "adc"), and we don't want to risk picking
    # those up by mistake, so we look at character-based columns intead.
    #
    for i in range (0, 3):
      # 6-character margin, 2-characters + 1 space for each field
      idx = 6 + i * 3;
      octetstr = line[idx:idx+3]
      if not numpattern.match(octetstr):
        break;

      if curr_octet > octets:
        break;

      curr_val += int('0x%s' % octetstr, 16) << (curr_octet * 8);
      curr_octet += 1;

  match = pattern.match(line)
  if match == None:
    continue;

  # Print previous symbol
  out_define();

  v8match = v8dbg.match(match.group(2));
  if v8match != None:
    out_reset();
    curr_sym = match.group(2);

# Print last symbol
out_define();

outfile.write("""

#endif /* V8_CONSTANTS_H */
""");