| # Module 'parser' | |
| # | |
| # Parse S-expressions output by the Panel Editor | |
| # (which is written in Scheme so it can't help writing S-expressions). | |
| # | |
| # See notes at end of file. | |
| from warnings import warnpy3k | |
| warnpy3k("the panelparser module has been removed in Python 3.0", stacklevel=2) | |
| del warnpy3k | |
| whitespace = ' \t\n' | |
| operators = '()\'' | |
| separators = operators + whitespace + ';' + '"' | |
| # Tokenize a string. | |
| # Return a list of tokens (strings). | |
| # | |
| def tokenize_string(s): | |
| tokens = [] | |
| while s: | |
| c = s[:1] | |
| if c in whitespace: | |
| s = s[1:] | |
| elif c == ';': | |
| s = '' | |
| elif c == '"': | |
| n = len(s) | |
| i = 1 | |
| while i < n: | |
| c = s[i] | |
| i = i+1 | |
| if c == '"': break | |
| if c == '\\': i = i+1 | |
| tokens.append(s[:i]) | |
| s = s[i:] | |
| elif c in operators: | |
| tokens.append(c) | |
| s = s[1:] | |
| else: | |
| n = len(s) | |
| i = 1 | |
| while i < n: | |
| if s[i] in separators: break | |
| i = i+1 | |
| tokens.append(s[:i]) | |
| s = s[i:] | |
| return tokens | |
| # Tokenize a whole file (given as file object, not as file name). | |
| # Return a list of tokens (strings). | |
| # | |
| def tokenize_file(fp): | |
| tokens = [] | |
| while 1: | |
| line = fp.readline() | |
| if not line: break | |
| tokens = tokens + tokenize_string(line) | |
| return tokens | |
| # Exception raised by parse_exr. | |
| # | |
| syntax_error = 'syntax error' | |
| # Parse an S-expression. | |
| # Input is a list of tokens as returned by tokenize_*(). | |
| # Return a pair (expr, tokens) | |
| # where expr is a list representing the s-expression, | |
| # and tokens contains the remaining tokens. | |
| # May raise syntax_error. | |
| # | |
| def parse_expr(tokens): | |
| if (not tokens) or tokens[0] != '(': | |
| raise syntax_error, 'expected "("' | |
| tokens = tokens[1:] | |
| expr = [] | |
| while 1: | |
| if not tokens: | |
| raise syntax_error, 'missing ")"' | |
| if tokens[0] == ')': | |
| return expr, tokens[1:] | |
| elif tokens[0] == '(': | |
| subexpr, tokens = parse_expr(tokens) | |
| expr.append(subexpr) | |
| else: | |
| expr.append(tokens[0]) | |
| tokens = tokens[1:] | |
| # Parse a file (given as file object, not as file name). | |
| # Return a list of parsed S-expressions found at the top level. | |
| # | |
| def parse_file(fp): | |
| tokens = tokenize_file(fp) | |
| exprlist = [] | |
| while tokens: | |
| expr, tokens = parse_expr(tokens) | |
| exprlist.append(expr) | |
| return exprlist | |
| # EXAMPLE: | |
| # | |
| # The input | |
| # '(hip (hop hur-ray))' | |
| # | |
| # passed to tokenize_string() returns the token list | |
| # ['(', 'hip', '(', 'hop', 'hur-ray', ')', ')'] | |
| # | |
| # When this is passed to parse_expr() it returns the expression | |
| # ['hip', ['hop', 'hur-ray']] | |
| # plus an empty token list (because there are no tokens left. | |
| # | |
| # When a file containing the example is passed to parse_file() it returns | |
| # a list whose only element is the output of parse_expr() above: | |
| # [['hip', ['hop', 'hur-ray']]] | |
| # TOKENIZING: | |
| # | |
| # Comments start with semicolon (;) and continue till the end of the line. | |
| # | |
| # Tokens are separated by whitespace, except the following characters | |
| # always form a separate token (outside strings): | |
| # ( ) ' | |
| # Strings are enclosed in double quotes (") and backslash (\) is used | |
| # as escape character in strings. |