You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

481 lines
17 KiB

import re
import sys
import unittest
sys.path.insert(0, '..')
from pycparser.c_lexer import CLexer
def token_list(clex):
return list(iter(clex.token, None))
def token_types(clex):
return [i.type for i in token_list(clex)]
class TestCLexerNoErrors(unittest.TestCase):
""" Test lexing of strings that are not supposed to cause
errors. Therefore, the error_func passed to the lexer
raises an exception.
"""
def error_func(self, msg, line, column):
self.fail(msg)
def on_lbrace_func(self):
pass
def on_rbrace_func(self):
pass
def type_lookup_func(self, typ):
if typ.startswith('mytype'):
return True
else:
return False
def setUp(self):
self.clex = CLexer(self.error_func, lambda: None, lambda: None,
self.type_lookup_func)
self.clex.build(optimize=False)
def assertTokensTypes(self, str, types):
self.clex.input(str)
self.assertEqual(token_types(self.clex), types)
def test_trivial_tokens(self):
self.assertTokensTypes('1', ['INT_CONST_DEC'])
self.assertTokensTypes('-', ['MINUS'])
self.assertTokensTypes('volatile', ['VOLATILE'])
self.assertTokensTypes('...', ['ELLIPSIS'])
self.assertTokensTypes('++', ['PLUSPLUS'])
self.assertTokensTypes('case int', ['CASE', 'INT'])
self.assertTokensTypes('caseint', ['ID'])
self.assertTokensTypes('$dollar cent$', ['ID', 'ID'])
self.assertTokensTypes('i ^= 1;', ['ID', 'XOREQUAL', 'INT_CONST_DEC', 'SEMI'])
def test_id_typeid(self):
self.assertTokensTypes('myt', ['ID'])
self.assertTokensTypes('mytype', ['TYPEID'])
self.assertTokensTypes('mytype6 var', ['TYPEID', 'ID'])
def test_integer_constants(self):
self.assertTokensTypes('12', ['INT_CONST_DEC'])
self.assertTokensTypes('12u', ['INT_CONST_DEC'])
self.assertTokensTypes('12l', ['INT_CONST_DEC'])
self.assertTokensTypes('199872Ul', ['INT_CONST_DEC'])
self.assertTokensTypes('199872lU', ['INT_CONST_DEC'])
self.assertTokensTypes('199872LL', ['INT_CONST_DEC'])
self.assertTokensTypes('199872ull', ['INT_CONST_DEC'])
self.assertTokensTypes('199872llu', ['INT_CONST_DEC'])
self.assertTokensTypes('1009843200000uLL', ['INT_CONST_DEC'])
self.assertTokensTypes('1009843200000LLu', ['INT_CONST_DEC'])
self.assertTokensTypes('077', ['INT_CONST_OCT'])
self.assertTokensTypes('0123456L', ['INT_CONST_OCT'])
self.assertTokensTypes('0xf7', ['INT_CONST_HEX'])
self.assertTokensTypes('0b110', ['INT_CONST_BIN'])
self.assertTokensTypes('0x01202AAbbf7Ul', ['INT_CONST_HEX'])
self.assertTokensTypes("'12'", ['INT_CONST_CHAR'])
self.assertTokensTypes("'123'", ['INT_CONST_CHAR'])
self.assertTokensTypes("'1AB4'", ['INT_CONST_CHAR'])
self.assertTokensTypes(r"'1A\n4'", ['INT_CONST_CHAR'])
# no 0 before x, so ID catches it
self.assertTokensTypes('xf7', ['ID'])
# - is MINUS, the rest a constnant
self.assertTokensTypes('-1', ['MINUS', 'INT_CONST_DEC'])
def test_special_names(self):
self.assertTokensTypes('sizeof offsetof', ['SIZEOF', 'OFFSETOF'])
def test_floating_constants(self):
self.assertTokensTypes('1.5f', ['FLOAT_CONST'])
self.assertTokensTypes('01.5', ['FLOAT_CONST'])
self.assertTokensTypes('.15L', ['FLOAT_CONST'])
self.assertTokensTypes('0.', ['FLOAT_CONST'])
# but just a period is a period
self.assertTokensTypes('.', ['PERIOD'])
self.assertTokensTypes('3.3e-3', ['FLOAT_CONST'])
self.assertTokensTypes('.7e25L', ['FLOAT_CONST'])
self.assertTokensTypes('6.e+125f', ['FLOAT_CONST'])
self.assertTokensTypes('666e666', ['FLOAT_CONST'])
self.assertTokensTypes('00666e+3', ['FLOAT_CONST'])
# but this is a hex integer + 3
self.assertTokensTypes('0x0666e+3', ['INT_CONST_HEX', 'PLUS', 'INT_CONST_DEC'])
def test_hexadecimal_floating_constants(self):
self.assertTokensTypes('0xDE.488641p0', ['HEX_FLOAT_CONST'])
self.assertTokensTypes('0x.488641p0', ['HEX_FLOAT_CONST'])
self.assertTokensTypes('0X12.P0', ['HEX_FLOAT_CONST'])
def test_char_constants(self):
self.assertTokensTypes(r"""'x'""", ['CHAR_CONST'])
self.assertTokensTypes(r"""L'x'""", ['WCHAR_CONST'])
self.assertTokensTypes(r"""'\t'""", ['CHAR_CONST'])
self.assertTokensTypes(r"""'\''""", ['CHAR_CONST'])
self.assertTokensTypes(r"""'\?'""", ['CHAR_CONST'])
self.assertTokensTypes(r"""'\0'""", ['CHAR_CONST'])
self.assertTokensTypes(r"""'\012'""", ['CHAR_CONST'])
self.assertTokensTypes(r"""'\x2f'""", ['CHAR_CONST'])
self.assertTokensTypes(r"""'\x2f12'""", ['CHAR_CONST'])
self.assertTokensTypes(r"""L'\xaf'""", ['WCHAR_CONST'])
def test_on_rbrace_lbrace(self):
braces = []
def on_lbrace():
braces.append('{')
def on_rbrace():
braces.append('}')
clex = CLexer(self.error_func, on_lbrace, on_rbrace,
self.type_lookup_func)
clex.build(optimize=False)
clex.input('hello { there } } and again }}{')
token_list(clex)
self.assertEqual(braces, ['{', '}', '}', '}', '}', '{'])
def test_string_literal(self):
self.assertTokensTypes('"a string"', ['STRING_LITERAL'])
self.assertTokensTypes('L"ing"', ['WSTRING_LITERAL'])
self.assertTokensTypes(
'"i am a string too \t"',
['STRING_LITERAL'])
self.assertTokensTypes(
r'''"esc\ape \"\'\? \0234 chars \rule"''',
['STRING_LITERAL'])
self.assertTokensTypes(
r'''"hello 'joe' wanna give it a \"go\"?"''',
['STRING_LITERAL'])
self.assertTokensTypes(
'"\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123"',
['STRING_LITERAL'])
# Note: a-zA-Z and '.-~^_!=&;,' are allowed as escape chars to support #line
# directives with Windows paths as filenames (..\..\dir\file)
self.assertTokensTypes(
r'"\x"',
['STRING_LITERAL'])
self.assertTokensTypes(
r'"\a\b\c\d\e\f\g\h\i\j\k\l\m\n\o\p\q\r\s\t\u\v\w\x\y\z\A\B\C\D\E\F\G\H\I\J\K\L\M\N\O\P\Q\R\S\T\U\V\W\X\Y\Z"',
['STRING_LITERAL'])
self.assertTokensTypes(
r'"C:\x\fa\x1e\xited"',
['STRING_LITERAL'])
# The lexer is permissive and allows decimal escapes (not just octal)
self.assertTokensTypes(
'"jx\9"',
['STRING_LITERAL'])
self.assertTokensTypes(
'"fo\9999999"',
['STRING_LITERAL'])
def test_mess(self):
self.assertTokensTypes(
r'[{}]()',
['LBRACKET',
'LBRACE', 'RBRACE',
'RBRACKET',
'LPAREN', 'RPAREN'])
self.assertTokensTypes(
r'()||!C&~Z?J',
['LPAREN', 'RPAREN',
'LOR',
'LNOT', 'ID',
'AND',
'NOT', 'ID',
'CONDOP', 'ID'])
self.assertTokensTypes(
r'+-*/%|||&&&^><>=<===!=',
['PLUS', 'MINUS', 'TIMES', 'DIVIDE', 'MOD',
'LOR', 'OR',
'LAND', 'AND',
'XOR',
'GT', 'LT', 'GE', 'LE', 'EQ', 'NE'])
self.assertTokensTypes(
r'++--->?.,;:',
['PLUSPLUS', 'MINUSMINUS',
'ARROW', 'CONDOP',
'PERIOD', 'COMMA', 'SEMI', 'COLON'])
def test_exprs(self):
self.assertTokensTypes(
'bb-cc',
['ID', 'MINUS', 'ID'])
self.assertTokensTypes(
'foo & 0xFF',
['ID', 'AND', 'INT_CONST_HEX'])
self.assertTokensTypes(
'(2+k) * 62',
['LPAREN', 'INT_CONST_DEC', 'PLUS', 'ID',
'RPAREN', 'TIMES', 'INT_CONST_DEC'],)
self.assertTokensTypes(
'x | y >> z',
['ID', 'OR', 'ID', 'RSHIFT', 'ID'])
self.assertTokensTypes(
'x <<= z << 5',
['ID', 'LSHIFTEQUAL', 'ID', 'LSHIFT', 'INT_CONST_DEC'])
self.assertTokensTypes(
'x = y > 0 ? y : -6',
['ID', 'EQUALS',
'ID', 'GT', 'INT_CONST_OCT',
'CONDOP',
'ID',
'COLON',
'MINUS', 'INT_CONST_DEC'])
self.assertTokensTypes(
'a+++b',
['ID', 'PLUSPLUS', 'PLUS', 'ID'])
def test_statements(self):
self.assertTokensTypes(
'for (int i = 0; i < n; ++i)',
['FOR', 'LPAREN',
'INT', 'ID', 'EQUALS', 'INT_CONST_OCT', 'SEMI',
'ID', 'LT', 'ID', 'SEMI',
'PLUSPLUS', 'ID',
'RPAREN'])
self.assertTokensTypes(
'self: goto self;',
['ID', 'COLON', 'GOTO', 'ID', 'SEMI'])
self.assertTokensTypes(
""" switch (typ)
{
case TYPE_ID:
m = 5;
break;
default:
m = 8;
}""",
['SWITCH', 'LPAREN', 'ID', 'RPAREN',
'LBRACE',
'CASE', 'ID', 'COLON',
'ID', 'EQUALS', 'INT_CONST_DEC', 'SEMI',
'BREAK', 'SEMI',
'DEFAULT', 'COLON',
'ID', 'EQUALS', 'INT_CONST_DEC', 'SEMI',
'RBRACE'])
def test_preprocessor_line(self):
self.assertTokensTypes('#abracadabra', ['PPHASH', 'ID'])
str = r"""
546
#line 66 "kwas\df.h"
id 4
dsf
# 9
armo
#line 10 "..\~..\test.h"
tok1
#line 99999 "include/me.h"
tok2
"""
#~ self.clex.filename
self.clex.input(str)
self.clex.reset_lineno()
t1 = self.clex.token()
self.assertEqual(t1.type, 'INT_CONST_DEC')
self.assertEqual(t1.lineno, 2)
t2 = self.clex.token()
self.assertEqual(t2.type, 'ID')
self.assertEqual(t2.value, 'id')
self.assertEqual(t2.lineno, 66)
self.assertEqual(self.clex.filename, r'kwas\df.h')
for i in range(3):
t = self.clex.token()
self.assertEqual(t.type, 'ID')
self.assertEqual(t.value, 'armo')
self.assertEqual(t.lineno, 9)
self.assertEqual(self.clex.filename, r'kwas\df.h')
t4 = self.clex.token()
self.assertEqual(t4.type, 'ID')
self.assertEqual(t4.value, 'tok1')
self.assertEqual(t4.lineno, 10)
self.assertEqual(self.clex.filename, r'..\~..\test.h')
t5 = self.clex.token()
self.assertEqual(t5.type, 'ID')
self.assertEqual(t5.value, 'tok2')
self.assertEqual(t5.lineno, 99999)
self.assertEqual(self.clex.filename, r'include/me.h')
def test_preprocessor_line_funny(self):
str = r'''
#line 10 "..\6\joe.h"
10
'''
self.clex.input(str)
self.clex.reset_lineno()
t1 = self.clex.token()
self.assertEqual(t1.type, 'INT_CONST_DEC')
self.assertEqual(t1.lineno, 10)
self.assertEqual(self.clex.filename, r'..\6\joe.h')
def test_preprocessor_pragma(self):
str = '''
42
#pragma
#pragma helo me
#pragma once
# pragma omp parallel private(th_id)
#\tpragma {pack: 2, smack: 3}
#pragma <includeme.h> "nowit.h"
#pragma "string"
#pragma somestring="some_other_string"
#pragma id 124124 and numbers 0235495
59
'''
# Check that pragmas are tokenized, including trailing string
self.clex.input(str)
self.clex.reset_lineno()
t1 = self.clex.token()
self.assertEqual(t1.type, 'INT_CONST_DEC')
t2 = self.clex.token()
self.assertEqual(t2.type, 'PPPRAGMA')
t3 = self.clex.token()
self.assertEqual(t3.type, 'PPPRAGMA')
t4 = self.clex.token()
self.assertEqual(t4.type, 'PPPRAGMASTR')
self.assertEqual(t4.value, 'helo me')
for i in range(3):
t = self.clex.token()
t5 = self.clex.token()
self.assertEqual(t5.type, 'PPPRAGMASTR')
self.assertEqual(t5.value, 'omp parallel private(th_id)')
for i in range(5):
ta = self.clex.token()
self.assertEqual(ta.type, 'PPPRAGMA')
tb = self.clex.token()
self.assertEqual(tb.type, 'PPPRAGMASTR')
t6 = self.clex.token()
self.assertEqual(t6.type, 'INT_CONST_DEC')
self.assertEqual(t6.lineno, 12)
# Keeps all the errors the lexer spits in one place, to allow
# easier modification if the error syntax changes.
#
ERR_ILLEGAL_CHAR = 'Illegal character'
ERR_OCTAL = 'Invalid octal constant'
ERR_UNMATCHED_QUOTE = 'Unmatched \''
ERR_INVALID_CCONST = 'Invalid char constant'
ERR_STRING_ESCAPE = 'String contains invalid escape'
ERR_FILENAME_BEFORE_LINE = 'filename before line'
ERR_LINENUM_MISSING = 'line number missing'
ERR_INVALID_LINE_DIRECTIVE = 'invalid #line directive'
class TestCLexerErrors(unittest.TestCase):
""" Test lexing of erroneous strings.
Works by passing an error functions that saves the error
in an attribute for later perusal.
"""
def error_func(self, msg, line, column):
self.error = msg
def on_lbrace_func(self):
pass
def on_rbrace_func(self):
pass
def type_lookup_func(self, typ):
return False
def setUp(self):
self.clex = CLexer(self.error_func, self.on_lbrace_func,
self.on_rbrace_func, self.type_lookup_func)
self.clex.build(optimize=False)
self.error = ""
def assertLexerError(self, str, error_like):
# feed the string to the lexer
self.clex.input(str)
# Pulls all tokens from the string. Errors will
# be written into self.error by the error_func
# callback
#
token_types(self.clex)
# compare the error to the expected
self.assertTrue(re.search(error_like, self.error),
"\nExpected error matching: %s\nGot: %s" %
(error_like, self.error))
# clear last error, for the sake of subsequent invocations
self.error = ""
def test_trivial_tokens(self):
self.assertLexerError('@', ERR_ILLEGAL_CHAR)
self.assertLexerError('`', ERR_ILLEGAL_CHAR)
self.assertLexerError('\\', ERR_ILLEGAL_CHAR)
def test_integer_constants(self):
self.assertLexerError('029', ERR_OCTAL)
self.assertLexerError('012345678', ERR_OCTAL)
def test_char_constants(self):
self.assertLexerError("'", ERR_UNMATCHED_QUOTE)
self.assertLexerError("'b\n", ERR_UNMATCHED_QUOTE)
self.assertLexerError("'\\xaa\n'", ERR_UNMATCHED_QUOTE)
self.assertLexerError(r"'123\12a'", ERR_INVALID_CCONST)
self.assertLexerError(r"'123\xabg'", ERR_INVALID_CCONST)
self.assertLexerError("''", ERR_INVALID_CCONST)
self.assertLexerError("'abcjx'", ERR_INVALID_CCONST)
self.assertLexerError(r"'\*'", ERR_INVALID_CCONST)
def test_string_literals(self):
self.assertLexerError(r'"jx\`"', ERR_STRING_ESCAPE)
self.assertLexerError(r'"hekllo\* on ix"', ERR_STRING_ESCAPE)
self.assertLexerError(r'L"hekllo\* on ix"', ERR_STRING_ESCAPE)
# Should not suffer from slow backtracking
self.assertLexerError(r'"\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\`\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123"', ERR_STRING_ESCAPE)
self.assertLexerError(r'"\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23\x23\`\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23\xf1\x23"', ERR_STRING_ESCAPE)
# Should not suffer from slow backtracking when there's no end quote
self.assertLexerError(r'"\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\`\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\123\12\123456', ERR_ILLEGAL_CHAR)
self.assertLexerError(r'"\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\`\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x23\x2\x23456', ERR_ILLEGAL_CHAR)
def test_preprocessor(self):
self.assertLexerError('#line "ka"', ERR_FILENAME_BEFORE_LINE)
self.assertLexerError('#line df', ERR_INVALID_LINE_DIRECTIVE)
self.assertLexerError('#line \n', ERR_LINENUM_MISSING)
if __name__ == '__main__':
unittest.main()