# This script converts a C file to use the PEP 384 type definition API # Usage: abitype.py < old_code > new_code import re, sys ############ Simplistic C scanner ################################## tokenizer = re.compile( r"(?P#.*\n)" r"|(?P/\*.*?\*/)" r"|(?P[a-zA-Z_][a-zA-Z0-9_]*)" r"|(?P[ \t\n]+)" r"|(?P.)", re.MULTILINE) tokens = [] source = sys.stdin.read() pos = 0 while pos != len(source): m = tokenizer.match(source, pos) tokens.append([m.lastgroup, m.group()]) pos += len(tokens[-1][1]) if tokens[-1][0] == 'preproc': # continuation lines are considered # only in preprocess statements while tokens[-1][1].endswith('\\\n'): nl = source.find('\n', pos) if nl == -1: line = source[pos:] else: line = source[pos:nl+1] tokens[-1][1] += line pos += len(line) ###### Replacement of PyTypeObject static instances ############## # classify each token, giving it a one-letter code: # S: static # T: PyTypeObject # I: ident # W: whitespace # =, {, }, ; : themselves def classify(): res = [] for t,v in tokens: if t == 'other' and v in "={};": res.append(v) elif t == 'ident': if v == 'PyTypeObject': res.append('T') elif v == 'static': res.append('S') else: res.append('I') elif t == 'ws': res.append('W') else: res.append('.') return ''.join(res) # Obtain a list of fields of a PyTypeObject, in declaration order, # skipping ob_base # All comments are dropped from the variable (which are typically # just the slot names, anyway), and information is discarded whether # the original type was static. def get_fields(start, real_end): pos = start # static? if tokens[pos][1] == 'static': pos += 2 # PyTypeObject pos += 2 # name name = tokens[pos][1] pos += 1 while tokens[pos][1] != '{': pos += 1 pos += 1 # PyVarObject_HEAD_INIT while tokens[pos][0] in ('ws', 'comment'): pos += 1 if tokens[pos][1] != 'PyVarObject_HEAD_INIT': raise Exception, '%s has no PyVarObject_HEAD_INIT' % name while tokens[pos][1] != ')': pos += 1 pos += 1 # field definitions: various tokens, comma-separated fields = [] while True: while tokens[pos][0] in ('ws', 'comment'): pos += 1 end = pos while tokens[end][1] not in ',}': if tokens[end][1] == '(': nesting = 1 while nesting: end += 1 if tokens[end][1] == '(': nesting+=1 if tokens[end][1] == ')': nesting-=1 end += 1 assert end < real_end # join field, excluding separator and trailing ws end1 = end-1 while tokens[end1][0] in ('ws', 'comment'): end1 -= 1 fields.append(''.join(t[1] for t in tokens[pos:end1+1])) if tokens[end][1] == '}': break pos = end+1 return name, fields # List of type slots as of Python 3.2, omitting ob_base typeslots = [ 'tp_name', 'tp_basicsize', 'tp_itemsize', 'tp_dealloc', 'tp_print', 'tp_getattr', 'tp_setattr', 'tp_reserved', 'tp_repr', 'tp_as_number', 'tp_as_sequence', 'tp_as_mapping', 'tp_hash', 'tp_call', 'tp_str', 'tp_getattro', 'tp_setattro', 'tp_as_buffer', 'tp_flags', 'tp_doc', 'tp_traverse', 'tp_clear', 'tp_richcompare', 'tp_weaklistoffset', 'tp_iter', 'iternextfunc', 'tp_methods', 'tp_members', 'tp_getset', 'tp_base', 'tp_dict', 'tp_descr_get', 'tp_descr_set', 'tp_dictoffset', 'tp_init', 'tp_alloc', 'tp_new', 'tp_free', 'tp_is_gc', 'tp_bases', 'tp_mro', 'tp_cache', 'tp_subclasses', 'tp_weaklist', 'tp_del' 'tp_version_tag' ] # Generate a PyType_Spec definition def make_slots(name, fields): res = [] res.append('static PyType_Slot %s_slots[] = {' % name) # defaults for spec spec = { 'tp_doc':'NULL', 'tp_itemsize':'0' } for i, val in enumerate(fields): if val.endswith('0'): continue if typeslots[i] in ('tp_name', 'tp_doc', 'tp_basicsize', 'tp_itemsize', 'tp_flags'): spec[typeslots[i]] = val continue res.append(' {Py_%s, %s},' % (typeslots[i], val)) res.append('};') res.append('static PyType_Spec %s_spec = {' % name) res.append(' %s,' % spec['tp_name']) res.append(' %s,' % spec['tp_doc']) res.append(' %s,' % spec['tp_basicsize']) res.append(' %s,' % spec['tp_itemsize']) res.append(' %s,' % spec['tp_flags']) res.append(' %s_slots,' % name) res.append('};\n') return '\n'.join(res) # Main loop: replace all static PyTypeObjects until # there are none left. while 1: c = classify() m = re.search('(SW)?TWIW?=W?{.*?};', c) if not m: break start = m.start() end = m.end() name, fields = get_fields(start, m) tokens[start:end] = [('',make_slots(name, fields))] # Output result to stdout for t, v in tokens: sys.stdout.write(v)