unicorn/bindings/const_generator.py
2015-09-07 19:25:13 -07:00

143 lines
4.9 KiB
Python

# Unicorn Engine
# By Dang Hoang Vu, 2013
from __future__ import print_function
import sys, re
INCL_DIR = '../include/unicorn/'
include = [ 'arm.h', 'arm64.h', 'mips.h', 'x86.h', 'sparc.h', 'm68k.h', 'unicorn.h' ]
template = {
'python': {
'header': "# For Unicorn Engine. AUTO-GENERATED FILE, DO NOT EDIT [%s_const.py]\n",
'footer': "",
'line_format': 'UC_%s = %s\n',
'out_file': './python/unicorn/%s_const.py',
# prefixes for constant filenames of all archs - case sensitive
'arm.h': 'arm',
'arm64.h': 'arm64',
'mips.h': 'mips',
'x86.h': 'x86',
'sparc.h': 'sparc',
'm68k.h': 'm68k',
'unicorn.h': 'unicorn',
'comment_open': '#',
'comment_close': '',
},
'go': {
'header': "package unicorn\n// For Unicorn Engine. AUTO-GENERATED FILE, DO NOT EDIT [%s_const.go]\nconst (\n",
'footer': ")",
'line_format': '\t%s = %s\n',
'out_file': './go/unicorn/%s_const.go',
# prefixes for constant filenames of all archs - case sensitive
'arm.h': 'arm',
'arm64.h': 'arm64',
'mips.h': 'mips',
'x86.h': 'x86',
'sparc.h': 'sparc',
'm68k.h': 'm68k',
'unicorn.h': 'unicorn',
'comment_open': '//',
'comment_close': '',
},
'java': {
'header': "// For Unicorn Engine. AUTO-GENERATED FILE, DO NOT EDIT\n\npackage unicorn;\n\npublic interface %sConst {\n",
'footer': "\n}\n",
'line_format': ' public static final int UC_%s = %s;\n',
'out_file': './java/unicorn/%sConst.java',
# prefixes for constant filenames of all archs - case sensitive
'arm.h': 'Arm',
'arm64.h': 'Arm64',
'mips.h': 'Mips',
'x86.h': 'X86',
'sparc.h': 'Sparc',
'm68k.h': 'M68k',
'unicorn.h': 'Unicorn',
'comment_open': '//',
'comment_close': '',
},
}
# markup for comments to be added to autogen files
MARKUP = '//>'
def gen(lang):
global include, INCL_DIR
templ = template[lang]
for target in include:
prefix = templ[target]
outfile = open(templ['out_file'] %(prefix), 'w')
outfile.write(templ['header'] % (prefix))
if target == 'unicorn.h':
prefix = ''
lines = open(INCL_DIR + target).readlines()
previous = {}
count = 0
for line in lines:
line = line.strip()
if line.startswith(MARKUP): # markup for comments
outfile.write("\n%s%s%s\n" %(templ['comment_open'], \
line.replace(MARKUP, ''), templ['comment_close']))
continue
if line == '' or line.startswith('//'):
continue
tmp = line.strip().split(',')
for t in tmp:
t = t.strip()
if not t or t.startswith('//'): continue
f = re.split('\s+', t)
# parse #define UC_TARGET (num)
define = False
if f[0] == '#define' and len(f) >= 3 and f[2].isdigit():
define = True
f.pop(0)
f.insert(1, '=')
if f[0].startswith("UC_" + prefix.upper()):
if len(f) > 1 and f[1] not in ('//', '='):
print("Error: Unable to convert %s" % f)
continue
elif len(f) > 1 and f[1] == '=':
rhs = ''.join(f[2:])
else:
rhs = str(count)
count += 1
lhs = f[0].strip()
# evaluate bitshifts in constants e.g. "UC_X86 = 1 << 1"
match = re.match(r'(?P<rhs>\s*\d+\s*<<\s*\d+\s*)', rhs)
if match:
rhs = eval(match.group(1))
else:
# evaluate references to other constants e.g. "UC_ARM_REG_X = UC_ARM_REG_SP"
match = re.match(r'^([^\d]\w+)$', rhs)
if match:
rhs = previous[match.group(1)]
lhs_strip = re.sub(r'^UC_', '', lhs)
count = int(rhs) + 1
if (count == 1):
outfile.write("\n")
outfile.write(templ['line_format'] % (lhs_strip, rhs))
previous[lhs] = rhs
outfile.write(templ['footer'])
outfile.close()
def main():
lang = sys.argv[1]
if not lang in template:
raise RuntimeError("Unsupported binding %s" % lang)
gen(sys.argv[1])
if __name__ == "__main__":
if len(sys.argv) < 2:
print("Usage:", sys.argv[0], " <python>")
sys.exit(1)
main()