summaryrefslogtreecommitdiff
path: root/util/newconfig/yappsrt.py
diff options
context:
space:
mode:
Diffstat (limited to 'util/newconfig/yappsrt.py')
-rw-r--r--util/newconfig/yappsrt.py172
1 files changed, 0 insertions, 172 deletions
diff --git a/util/newconfig/yappsrt.py b/util/newconfig/yappsrt.py
deleted file mode 100644
index 2ce2480f08..0000000000
--- a/util/newconfig/yappsrt.py
+++ /dev/null
@@ -1,172 +0,0 @@
-# Yapps 2.0 Runtime
-#
-# This module is needed to run generated parsers.
-
-from string import *
-import exceptions
-import re
-
-class SyntaxError(Exception):
- """When we run into an unexpected token, this is the exception to use"""
- def __init__(self, pos=-1, msg="Bad Token"):
- self.pos = pos
- self.msg = msg
- def __repr__(self):
- if self.pos < 0: return "#<syntax-error>"
- else: return "SyntaxError[@ char " + `self.pos` + ": " + self.msg + "]"
-
-class NoMoreTokens(Exception):
- """Another exception object, for when we run out of tokens"""
- pass
-
-class Scanner:
- def __init__(self, patterns, ignore, input):
- """Patterns is [(terminal,regex)...]
- Ignore is [terminal,...];
- Input is a string"""
- self.tokens = []
- self.restrictions = []
- self.input = input
- self.pos = 0
- self.ignore = ignore
- # The stored patterns are a pair (compiled regex,source
- # regex). If the patterns variable passed in to the
- # constructor is None, we assume that the class already has a
- # proper .patterns list constructed
- if patterns is not None:
- self.patterns = []
- for k,r in patterns:
- self.patterns.append( (k, re.compile(r)) )
-
- def token(self, i, restrict=0):
- """Get the i'th token, and if i is one past the end, then scan
- for another token; restrict is a list of tokens that
- are allowed, or 0 for any token."""
- if i == len(self.tokens): self.scan(restrict)
- if i < len(self.tokens):
- # Make sure the restriction is more restricted
- if restrict and self.restrictions[i]:
- for r in restrict:
- if r not in self.restrictions[i]:
- raise "Unimplemented: restriction set changed"
- return self.tokens[i]
- raise NoMoreTokens()
-
- def __repr__(self):
- """Print the last 10 tokens that have been scanned in"""
- output = ''
- for t in self.tokens[-10:]:
- output = '%s\n (@%s) %s = %s' % (output,t[0],t[2],`t[3]`)
- return output
-
- def scan(self, restrict):
- """Should scan another token and add it to the list, self.tokens,
- and add the restriction to self.restrictions"""
- # Keep looking for a token, ignoring any in self.ignore
- while 1:
- # Search the patterns for the longest match, with earlier
- # tokens in the list having preference
- best_match = -1
- best_pat = '(error)'
- for p, regexp in self.patterns:
- # First check to see if we're ignoring this token
- if restrict and p not in restrict and p not in self.ignore:
- continue
- m = regexp.match(self.input, self.pos)
- if m and len(m.group(0)) > best_match:
- # We got a match that's better than the previous one
- best_pat = p
- best_match = len(m.group(0))
-
- # If we didn't find anything, raise an error
- if best_pat == '(error)' and best_match < 0:
- msg = "Bad Token"
- if restrict:
- msg = "Trying to find one of "+join(restrict,", ")
- raise SyntaxError(self.pos, msg)
-
- # If we found something that isn't to be ignored, return it
- if best_pat not in self.ignore:
- # Create a token with this data
- token = (self.pos, self.pos+best_match, best_pat,
- self.input[self.pos:self.pos+best_match])
- self.pos = self.pos + best_match
- # Only add this token if it's not in the list
- # (to prevent looping)
- if not self.tokens or token != self.tokens[-1]:
- self.tokens.append(token)
- self.restrictions.append(restrict)
- return
- else:
- # This token should be ignored ..
- self.pos = self.pos + best_match
-
-class Parser:
- def __init__(self, scanner):
- self._scanner = scanner
- self._pos = 0
-
- def _peek(self, *types):
- """Returns the token type for lookahead; if there are any args
- then the list of args is the set of token types to allow"""
- tok = self._scanner.token(self._pos, types)
- return tok[2]
-
- def _scan(self, type):
- """Returns the matched text, and moves to the next token"""
- tok = self._scanner.token(self._pos, [type])
- if tok[2] != type:
- raise SyntaxError(tok[0], 'Trying to find '+type)
- self._pos = 1+self._pos
- return tok[3]
-
-
-
-def print_error(input, err, scanner):
- """This is a really dumb long function to print error messages nicely."""
- p = err.pos
- # Figure out the line number
- line = count(input[:p], '\n')
- print err.msg+" on line "+`line+1`+":"
- # Now try printing part of the line
- text = input[max(p-80,0):p+80]
- p = p - max(p-80,0)
-
- # Strip to the left
- i = rfind(text[:p],'\n')
- j = rfind(text[:p],'\r')
- if i < 0 or (j < i and j >= 0): i = j
- if i >= 0 and i < p:
- p = p - i - 1
- text = text[i+1:]
-
- # Strip to the right
- i = find(text,'\n',p)
- j = find(text,'\r',p)
- if i < 0 or (j < i and j >= 0): i = j
- if i >= 0:
- text = text[:i]
-
- # Now shorten the text
- while len(text) > 70 and p > 60:
- # Cut off 10 chars
- text = "..." + text[10:]
- p = p - 7
-
- # Now print the string, along with an indicator
- print '> ',text
- print '> ',' '*p + '^'
- print 'List of nearby tokens:', scanner
-
-def wrap_error_reporter(parser, rule):
- try: return getattr(parser, rule)()
- except SyntaxError, s:
- input = parser._scanner.input
- try:
- print_error(input, s, parser._scanner)
- except ImportError:
- print 'Syntax Error',s.msg,'on line',1+count(input[:s.pos], '\n')
- except NoMoreTokens:
- print 'Could not complete parsing; stopped around here:'
- print parser._scanner
-