2009-04-23 20:58:16 +00:00
|
|
|
import os
|
|
|
|
import re
|
|
|
|
|
|
|
|
#==[regexps]=================================================
|
|
|
|
|
|
|
|
# Clean desc
|
|
|
|
REdefine = re.compile(r''
|
|
|
|
r'(?P<desc>)' # /** *desc */
|
|
|
|
r'#\s*define\s(?P<name>[^(\n]+?)\s(?P<code>.+)$' # #define name value
|
|
|
|
, re.MULTILINE)
|
|
|
|
|
|
|
|
# Get structs
|
|
|
|
REstructs = re.compile(r''
|
|
|
|
#r'//\s?[\-]*\s(?P<desc>.*?)\*/\s' # /** *desc */
|
|
|
|
#r'//\s?[\-]*(?P<desc>.*?)\*/(?:.*?)' # garbage
|
|
|
|
r'//\s?[\-]*\s(?P<desc>.*?)\*/\W*?' # /** *desc */
|
|
|
|
r'struct\s(?:ASSIMP_API\s)?(?P<name>[a-z][a-z0-9_]\w+\b)' # struct name
|
|
|
|
r'[^{]*?\{' # {
|
|
|
|
r'(?P<code>.*?)' # code
|
|
|
|
r'\}\s*(PACK_STRUCT)?;' # };
|
|
|
|
, re.IGNORECASE + re.DOTALL + re.MULTILINE)
|
|
|
|
|
|
|
|
# Clean desc
|
|
|
|
REdesc = re.compile(r''
|
|
|
|
r'^\s*?([*]|/\*\*)(?P<line>.*?)' # * line
|
|
|
|
, re.IGNORECASE + re.DOTALL + re.MULTILINE)
|
|
|
|
|
|
|
|
# Remove #ifdef __cplusplus
|
|
|
|
RErmifdef = re.compile(r''
|
|
|
|
r'#ifdef __cplusplus' # #ifdef __cplusplus
|
|
|
|
r'(?P<code>.*)' # code
|
|
|
|
r'#endif(\s*//\s*!?\s*__cplusplus)*' # #endif
|
|
|
|
, re.IGNORECASE + re.DOTALL)
|
|
|
|
|
|
|
|
# Replace comments
|
|
|
|
RErpcom = re.compile(r''
|
|
|
|
r'[ ]*(/\*\*\s|\*/|\B\*\s|//!)' # /**
|
|
|
|
r'(?P<line>.*?)' # * line
|
|
|
|
, re.IGNORECASE + re.DOTALL)
|
|
|
|
|
|
|
|
# Restructure
|
|
|
|
def GetType(type, prefix='c_'):
|
|
|
|
t = type
|
|
|
|
while t.endswith('*'):
|
|
|
|
t = t[:-1]
|
|
|
|
types = {'unsigned int':'uint', 'unsigned char':'ubyte', 'size_t':'uint',}
|
|
|
|
if t in types:
|
|
|
|
t = types[t]
|
|
|
|
t = prefix + t
|
|
|
|
while type.endswith('*'):
|
|
|
|
t = "POINTER(" + t + ")"
|
|
|
|
type = type[:-1]
|
|
|
|
return t
|
|
|
|
|
|
|
|
def restructure( match ):
|
|
|
|
type = match.group("type")
|
|
|
|
if match.group("struct") == "":
|
|
|
|
type = GetType(type)
|
|
|
|
elif match.group("struct") == "C_ENUM ":
|
|
|
|
type = "c_uint"
|
|
|
|
else:
|
2009-04-25 21:12:53 +00:00
|
|
|
type = GetType(type[2:], '')
|
2009-04-23 20:58:16 +00:00
|
|
|
if match.group("index"):
|
|
|
|
type = type + "*" + match.group("index")
|
|
|
|
|
|
|
|
result = ""
|
|
|
|
for name in match.group("name").split(','):
|
|
|
|
result += "(\"" + name.strip() + "\", "+ type + "),"
|
|
|
|
|
|
|
|
return result
|
|
|
|
|
|
|
|
RErestruc = re.compile(r''
|
|
|
|
r'(?P<struct>C_STRUCT\s|C_ENUM\s|)' # [C_STRUCT]
|
|
|
|
r'(?P<type>\w+\s?\w+?[*]*)\s' # type
|
|
|
|
#r'(?P<name>\w+)' # name
|
|
|
|
r'(?P<name>\w+|[a-z0-9_, ]+)' # name
|
|
|
|
r'(:?\[(?P<index>\w+)\])?;' # []; (optional)
|
|
|
|
, re.DOTALL)
|
|
|
|
#==[template]================================================
|
|
|
|
template = """
|
|
|
|
class $NAME$(Structure):
|
|
|
|
\"\"\"
|
|
|
|
$DESCRIPTION$
|
|
|
|
\"\"\"
|
|
|
|
$DEFINES$
|
|
|
|
_fields_ = [
|
|
|
|
$FIELDS$
|
|
|
|
]
|
|
|
|
"""
|
|
|
|
|
|
|
|
templateSR = """
|
|
|
|
class $NAME$(Structure):
|
|
|
|
\"\"\"
|
|
|
|
$DESCRIPTION$
|
|
|
|
\"\"\"
|
|
|
|
$DEFINES$
|
|
|
|
|
|
|
|
$NAME$._fields_ = [
|
|
|
|
$FIELDS$
|
|
|
|
]
|
|
|
|
"""
|
|
|
|
#============================================================
|
|
|
|
def Structify(fileName):
|
|
|
|
file = open(fileName, 'r')
|
|
|
|
text = file.read()
|
|
|
|
result = []
|
|
|
|
|
|
|
|
# Get defines.
|
|
|
|
defs = REdefine.findall(text)
|
|
|
|
# Create defines
|
|
|
|
defines = "\n"
|
|
|
|
for define in defs:
|
|
|
|
# Clean desc
|
|
|
|
desc = REdesc.sub('', define[0])
|
|
|
|
# Replace comments
|
|
|
|
desc = RErpcom.sub('#\g<line>', desc)
|
|
|
|
defines += desc
|
|
|
|
defines += " "*4 + define[1] + " = " + define[2] + "\n"
|
|
|
|
|
|
|
|
# Get structs
|
|
|
|
rs = REstructs.finditer(text)
|
|
|
|
|
|
|
|
fileName = os.path.basename(fileName)
|
|
|
|
print fileName
|
|
|
|
for r in rs:
|
2009-04-25 21:12:53 +00:00
|
|
|
name = r.group('name')[2:]
|
2009-04-23 20:58:16 +00:00
|
|
|
desc = r.group('desc')
|
|
|
|
|
|
|
|
# Skip some structs
|
2009-04-25 21:12:53 +00:00
|
|
|
if name == "FileIO" or name == "File" or name == "locateFromAssimpHeap":
|
2009-04-23 20:58:16 +00:00
|
|
|
continue
|
|
|
|
|
|
|
|
text = r.group('code')
|
|
|
|
|
|
|
|
# Clean desc
|
|
|
|
desc = REdesc.sub('', desc)
|
|
|
|
|
|
|
|
desc = "See '"+ fileName +"' for details." #TODO
|
|
|
|
|
|
|
|
# Remove #ifdef __cplusplus
|
|
|
|
text = RErmifdef.sub('', text)
|
|
|
|
|
|
|
|
# Whether the struct contains more than just POD
|
|
|
|
primitive = text.find('C_STRUCT') == -1
|
|
|
|
|
|
|
|
# Restructure
|
|
|
|
text = RErestruc.sub(restructure, text)
|
|
|
|
|
|
|
|
# Replace comments
|
|
|
|
text = RErpcom.sub('#\g<line>', text)
|
|
|
|
|
|
|
|
# Whether it's selfreferencing: ex. struct Node { Node* parent; };
|
|
|
|
selfreferencing = text.find('POINTER('+name+')') != -1
|
|
|
|
|
2009-04-25 21:12:53 +00:00
|
|
|
complex = name == "Scene"
|
2009-04-23 20:58:16 +00:00
|
|
|
|
|
|
|
# Create description
|
|
|
|
description = ""
|
|
|
|
for line in desc.split('\n'):
|
|
|
|
description += " "*4 + line.strip() + "\n"
|
|
|
|
description = description.rstrip()
|
|
|
|
|
|
|
|
# Create fields
|
|
|
|
fields = ""
|
|
|
|
for line in text.split('\n'):
|
|
|
|
fields += " "*12 + line.strip() + "\n"
|
|
|
|
fields = fields.strip()
|
|
|
|
|
|
|
|
if selfreferencing:
|
|
|
|
templ = templateSR
|
|
|
|
else:
|
|
|
|
templ = template
|
|
|
|
|
|
|
|
# Put it all together
|
|
|
|
text = templ.replace('$NAME$', name)
|
|
|
|
text = text.replace('$DESCRIPTION$', description)
|
|
|
|
text = text.replace('$FIELDS$', fields)
|
|
|
|
|
2009-04-25 21:12:53 +00:00
|
|
|
if ((name.lower() == fileName.split('.')[0][2:].lower()) and (name != 'Material')) or name == "String":
|
2009-04-23 20:58:16 +00:00
|
|
|
text = text.replace('$DEFINES$', defines)
|
|
|
|
else:
|
|
|
|
text = text.replace('$DEFINES$', '')
|
|
|
|
|
|
|
|
result.append((primitive, selfreferencing, complex, text))
|
|
|
|
|
|
|
|
return result
|
|
|
|
|
|
|
|
text = "#-*- coding: UTF-8 -*-\n\n"
|
|
|
|
text += "from ctypes import POINTER, c_int, c_uint, c_char, c_float, Structure, c_char_p, c_double, c_ubyte\n\n"
|
|
|
|
|
|
|
|
structs1 = ""
|
|
|
|
structs2 = ""
|
|
|
|
structs3 = ""
|
|
|
|
structs4 = ""
|
|
|
|
|
|
|
|
path = '../include/'
|
|
|
|
files = os.listdir (path)
|
|
|
|
#files = ["aiScene.h", "aiTypes.h"]
|
|
|
|
for fileName in files:
|
|
|
|
if fileName.endswith('.h'):
|
|
|
|
for struct in Structify(os.path.join(path, fileName)):
|
|
|
|
primitive, sr, complex, struct = struct
|
|
|
|
if primitive:
|
|
|
|
structs1 += struct
|
|
|
|
elif sr:
|
|
|
|
structs2 += struct
|
|
|
|
elif complex:
|
|
|
|
structs4 += struct
|
|
|
|
else:
|
|
|
|
structs3 += struct
|
|
|
|
|
|
|
|
text += structs1 + structs2 + structs3 + structs4
|
|
|
|
|
|
|
|
file = open('structs.txt', 'w')
|
|
|
|
file.write(text)
|
|
|
|
file.close()
|