Multiple improvements to tab completion.

I refactored the API quite a bit, to retain readline compatibility but
make it more independent of readline.  There's still more to do in
cleaning up our init_readline() method, but now the completer objects
have separate rlcomplete() and complete() methods.  The former uses
the quirky readline API with a state flag, while the latter is
stateless, takes only text information, and is more suitable for GUIs
and other frontends to call programatically.

Made other minor fixes to ensure the test suite passes in full.

While all this code is a bit messy, we're getting in the direction of
the APIs we need in the long run.

IPython/core/completer.py

@@ -1,658 +1,722 b''
 """Word completion for IPython.
 
 This module is a fork of the rlcompleter module in the Python standard
 library.  The original enhancements made to rlcompleter have been sent
 upstream and were accepted as of Python 2.3, but we need a lot more
 functionality specific to IPython, so this module will continue to live as an
 IPython-specific utility.
 
 Original rlcompleter documentation:
 
 This requires the latest extension to the readline module (the
 completes keywords, built-ins and globals in __main__; when completing
 NAME.NAME..., it evaluates (!) the expression up to the last dot and
 completes its attributes.
 
 It's very cool to do "import string" type "string.", hit the
 completion key (twice), and see the list of names defined by the
 string module!
 
 Tip: to use the tab key as the completion key, call
 
     readline.parse_and_bind("tab: complete")
 
 Notes:
 
 - Exceptions raised by the completer function are *ignored* (and
 generally cause the completion to fail).  This is a feature -- since
 readline sets the tty device in raw (or cbreak) mode, printing a
 traceback wouldn't work well without some complicated hoopla to save,
 reset and restore the tty state.
 
 - The evaluation of the NAME.NAME... form may cause arbitrary
 application defined code to be executed if an object with a
 __getattr__ hook is found.  Since it is the responsibility of the
 application (or the user) to enable this feature, I consider this an
 acceptable risk.  More complicated expressions (e.g. function calls or
 indexing operations) are *not* evaluated.
 
 - GNU readline is also used by the built-in functions input() and
 raw_input(), and thus these also benefit/suffer from the completer
 features.  Clearly an interactive application can benefit by
 specifying its own completer function and using raw_input() for all
 its input.
 
 - When the original stdin is not a tty device, GNU readline is never
 used, and this module (and the readline module) are silently inactive.
 """
 
 #*****************************************************************************
 #
 # Since this file is essentially a minimally modified copy of the rlcompleter
 # module which is part of the standard Python distribution, I assume that the
 # proper procedure is to maintain its copyright as belonging to the Python
 # Software Foundation (in addition to my own, for all new code).
 #
 #       Copyright (C) 2008-2010 IPython Development Team
 #       Copyright (C) 2001-2007 Fernando Perez. <[e-mail]>
 #       Copyright (C) 2001 Python Software Foundation, www.python.org
 #
 #  Distributed under the terms of the BSD License.  The full license is in
 #  the file COPYING, distributed as part of this software.
 #
 #*****************************************************************************
 
 #-----------------------------------------------------------------------------
 # Imports
 #-----------------------------------------------------------------------------
 
 import __builtin__
 import __main__
 import glob
 import inspect
 import itertools
 import keyword
 import os
 import re
 import shlex
 import sys
 
 from IPython.core.error import TryNext
 from IPython.core.prefilter import ESC_MAGIC
 from IPython.utils import generics
 from IPython.utils.frame import debugx
 from IPython.utils.dir2 import dir2
-import IPython.utils.rlineimpl as readline
 
 #-----------------------------------------------------------------------------
 # Globals
 #-----------------------------------------------------------------------------
 
 # Public API
 __all__ = ['Completer','IPCompleter']
 
 if sys.platform == 'win32':
     PROTECTABLES = ' '
 else:
     PROTECTABLES = ' ()'
 
 #-----------------------------------------------------------------------------
 # Main functions and classes
 #-----------------------------------------------------------------------------
 
 def protect_filename(s):
     """Escape a string to protect certain characters."""
     
     return "".join([(ch in PROTECTABLES and '\\' + ch or ch)
                     for ch in s])
 
 
+def mark_dirs(matches):
+    """Mark directories in input list by appending '/' to their names."""
+    out = []
+    isdir = os.path.isdir
+    for x in matches:
+        if isdir(x):
+            out.append(x+'/')
+        else:
+            out.append(x)
+    return out
+
+
 def single_dir_expand(matches):
     "Recursively expand match lists containing a single dir."
 
     if len(matches) == 1 and os.path.isdir(matches[0]):
         # Takes care of links to directories also.  Use '/'
         # explicitly, even under Windows, so that name completions
         # don't end up escaped.
         d = matches[0]
         if d[-1] in ['/','\\']:
             d = d[:-1]
 
         subdirs = os.listdir(d)
         if subdirs:
             matches = [ (d + '/' + p) for p in subdirs]
             return single_dir_expand(matches)
         else:
             return matches
     else:
         return matches
 
 class Bunch: pass
 
 class Completer:
     def __init__(self,namespace=None,global_namespace=None):
         """Create a new completer for the command line.
 
         Completer([namespace,global_namespace]) -> completer instance.
 
         If unspecified, the default namespace where completions are performed
         is __main__ (technically, __main__.__dict__). Namespaces should be
         given as dictionaries.
 
         An optional second namespace can be given.  This allows the completer
         to handle cases where both the local and global scopes need to be
         distinguished.
 
         Completer instances should be used as the completion mechanism of
         readline via the set_completer() call:
 
         readline.set_completer(Completer(my_namespace).complete)
         """
 
         # Don't bind to namespace quite yet, but flag whether the user wants a
         # specific namespace or to use __main__.__dict__. This will allow us
         # to bind to __main__.__dict__ at completion time, not now.
         if namespace is None:
             self.use_main_ns = 1
         else:
             self.use_main_ns = 0
             self.namespace = namespace
 
         # The global namespace, if given, can be bound directly
         if global_namespace is None:
             self.global_namespace = {}
         else:
             self.global_namespace = global_namespace
 
     def complete(self, text, state):
         """Return the next possible completion for 'text'.
 
         This is called successively with state == 0, 1, 2, ... until it
         returns None.  The completion should begin with 'text'.
 
         """
         if self.use_main_ns:
             self.namespace = __main__.__dict__
             
         if state == 0:
             if "." in text:
                 self.matches = self.attr_matches(text)
             else:
                 self.matches = self.global_matches(text)
         try:
             return self.matches[state]
         except IndexError:
             return None
 
     def global_matches(self, text):
         """Compute matches when text is a simple name.
 
         Return a list of all keywords, built-in functions and names currently
         defined in self.namespace or self.global_namespace that match.
 
         """
         #print 'Completer->global_matches, txt=%r' % text # dbg
         matches = []
         match_append = matches.append
         n = len(text)
         for lst in [keyword.kwlist,
                     __builtin__.__dict__.keys(),
                     self.namespace.keys(),
                     self.global_namespace.keys()]:
             for word in lst:
                 if word[:n] == text and word != "__builtins__":
                     match_append(word)
         return matches
 
     def attr_matches(self, text):
         """Compute matches when text contains a dot.
 
         Assuming the text is of the form NAME.NAME....[NAME], and is
         evaluatable in self.namespace or self.global_namespace, it will be
         evaluated and its attributes (as revealed by dir()) are used as
         possible completions.  (For class instances, class members are are
         also considered.)
 
         WARNING: this can still invoke arbitrary C code, if an object
         with a __getattr__ hook is evaluated.
 
         """
 
         #print 'Completer->attr_matches, txt=%r' % text # dbg
         # Another option, seems to work great. Catches things like ''.<tab>
         m = re.match(r"(\S+(\.\w+)*)\.(\w*)$", text)
 
         if not m:
             return []
         
         expr, attr = m.group(1, 3)
         try:
             obj = eval(expr, self.namespace)
         except:
             try:
                 obj = eval(expr, self.global_namespace)
             except:
                 return []
 
         words = dir2(obj)
         
         try:
             words = generics.complete_object(obj, words)
         except TryNext:
             pass
         # Build match list to return
         n = len(attr)
         res = ["%s.%s" % (expr, w) for w in words if w[:n] == attr ]
         return res
 
 
 class IPCompleter(Completer):
     """Extension of the completer class with IPython-specific features"""
 
     def __init__(self,shell,namespace=None,global_namespace=None,
-                 omit__names=0,alias_table=None):
+                 omit__names=0, alias_table=None, use_readline=True):
         """IPCompleter() -> completer
 
         Return a completer object suitable for use by the readline library
         via readline.set_completer().
 
         Inputs:
 
         - shell: a pointer to the ipython shell itself.  This is needed
         because this completer knows about magic functions, and those can
         only be accessed via the ipython instance.
 
         - namespace: an optional dict where completions are performed.
 
         - global_namespace: secondary optional dict for completions, to
         handle cases (such as IPython embedded inside functions) where
         both Python scopes are visible.
 
         - The optional omit__names parameter sets the completer to omit the
         'magic' names (__magicname__) for python objects unless the text
         to be completed explicitly starts with one or more underscores.
 
         - If alias_table is supplied, it should be a dictionary of aliases
-        to complete. """
+        to complete.
+
+        use_readline : bool, optional
+          If true, use the readline library.  This completer can still function
+          without readline, though in that case callers must provide some extra
+          information on each call about the current line."""
 
         Completer.__init__(self,namespace,global_namespace)
 
         self.magic_escape = ESC_MAGIC
+
+        # Readline-dependent code
+        self.use_readline = use_readline
+        if use_readline:
+            import IPython.utils.rlineimpl as readline
         self.readline = readline
         delims = self.readline.get_completer_delims()
         delims = delims.replace(self.magic_escape,'')
         self.readline.set_completer_delims(delims)
         self.get_line_buffer = self.readline.get_line_buffer
         self.get_endidx = self.readline.get_endidx
+        # /end readline-dependent code
+
+        # List where completion matches will be stored
+        self.matches = []
         self.omit__names = omit__names
         self.merge_completions = shell.readline_merge_completions
         self.shell = shell.shell
         if alias_table is None:
             alias_table = {}
         self.alias_table = alias_table
         # Regexp to split filenames with spaces in them
         self.space_name_re = re.compile(r'([^\\] )')
         # Hold a local ref. to glob.glob for speed
         self.glob = glob.glob
 
         # Determine if we are running on 'dumb' terminals, like (X)Emacs
         # buffers, to avoid completion problems.
         term = os.environ.get('TERM','xterm')
         self.dumb_terminal = term in ['dumb','emacs']
         
         # Special handling of backslashes needed in win32 platforms
         if sys.platform == "win32":
             self.clean_glob = self._clean_glob_win32
         else:
             self.clean_glob = self._clean_glob
 
         # All active matcher routines for completion
         self.matchers = [self.python_matches,
                          self.file_matches,
                          self.magic_matches,
                          self.alias_matches,
-                         self.python_func_kw_matches]
+                         self.python_func_kw_matches,
+                         ]
     
     # Code contributed by Alex Schmolck, for ipython/emacs integration
     def all_completions(self, text):
         """Return all possible completions for the benefit of emacs."""
 
         completions = []
         comp_append = completions.append
         try:
             for i in xrange(sys.maxint):
                 res = self.complete(text, i, text)
                 if not res:
                     break
                 comp_append(res)
         #XXX workaround for ``notDefined.<tab>``
         except NameError:
             pass
         return completions
     # /end Alex Schmolck code.
 
     def _clean_glob(self,text):
         return self.glob("%s*" % text)
 
     def _clean_glob_win32(self,text):
         return [f.replace("\\","/")
                 for f in self.glob("%s*" % text)]            
 
     def file_matches(self, text):
         """Match filenames, expanding ~USER type strings.
 
         Most of the seemingly convoluted logic in this completer is an
         attempt to handle filenames with spaces in them.  And yet it's not
         quite perfect, because Python's readline doesn't expose all of the
         GNU readline details needed for this to be done correctly.
 
         For a filename with a space in it, the printed completions will be
         only the parts after what's already been typed (instead of the
         full completions, as is normally done).  I don't think with the
         current (as of Python 2.3) Python readline it's possible to do
         better."""
 
         #print 'Completer->file_matches: <%s>' % text # dbg
 
         # chars that require escaping with backslash - i.e. chars
         # that readline treats incorrectly as delimiters, but we
         # don't want to treat as delimiters in filename matching
         # when escaped with backslash
 
         if text.startswith('!'):
             text = text[1:]
             text_prefix = '!'
         else:
             text_prefix = ''
             
         lbuf = self.lbuf
         open_quotes = 0  # track strings with open quotes
         try:
             lsplit = shlex.split(lbuf)[-1]
         except ValueError:
             # typically an unmatched ", or backslash without escaped char.
             if lbuf.count('"')==1:
                 open_quotes = 1
                 lsplit = lbuf.split('"')[-1]
             elif lbuf.count("'")==1:
                 open_quotes = 1
                 lsplit = lbuf.split("'")[-1]
             else:
                 return []
         except IndexError:
             # tab pressed on empty line
             lsplit = ""
 
         if lsplit != protect_filename(lsplit):
             # if protectables are found, do matching on the whole escaped
             # name
             has_protectables = 1
             text0,text = text,lsplit
         else:
             has_protectables = 0
             text = os.path.expanduser(text)
 
         if text == "":
             return [text_prefix + protect_filename(f) for f in self.glob("*")]
 
         m0 = self.clean_glob(text.replace('\\',''))
         if has_protectables:
             # If we had protectables, we need to revert our changes to the
             # beginning of filename so that we don't double-write the part
             # of the filename we have so far
             len_lsplit = len(lsplit)
             matches = [text_prefix + text0 + 
                        protect_filename(f[len_lsplit:]) for f in m0]
         else:
             if open_quotes:
                 # if we have a string with an open quote, we don't need to
                 # protect the names at all (and we _shouldn't_, as it
                 # would cause bugs when the filesystem call is made).
                 matches = m0
             else:
                 matches = [text_prefix + 
                            protect_filename(f) for f in m0]
 
         #print 'mm',matches  # dbg
-        return single_dir_expand(matches)
+        #return single_dir_expand(matches)
+        return mark_dirs(matches)
 
     def magic_matches(self, text):
         """Match magics"""
         #print 'Completer->magic_matches:',text,'lb',self.lbuf # dbg
         # Get all shell magics now rather than statically, so magics loaded at
         # runtime show up too
         magics = self.shell.lsmagic()
         pre = self.magic_escape
         baretext = text.lstrip(pre)
         return [ pre+m for m in magics if m.startswith(baretext)]
 
     def alias_matches(self, text):
         """Match internal system aliases"""        
         #print 'Completer->alias_matches:',text,'lb',self.lbuf # dbg
         
         # if we are not in the first 'item', alias matching 
         # doesn't make sense - unless we are starting with 'sudo' command.
         if ' ' in self.lbuf.lstrip() and \
                not self.lbuf.lstrip().startswith('sudo'):
             return []
         text = os.path.expanduser(text)
         aliases =  self.alias_table.keys()
         if text == "":
             return aliases
         else:
             return [alias for alias in aliases if alias.startswith(text)]
 
     def python_matches(self,text):
         """Match attributes or global python names"""
 
         #print 'Completer->python_matches, txt=%r' % text # dbg
         if "." in text:
             try:
                 matches = self.attr_matches(text)
                 if text.endswith('.') and self.omit__names:
                     if self.omit__names == 1:
                         # true if txt is _not_ a __ name, false otherwise:
                         no__name = (lambda txt:
                                     re.match(r'.*\.__.*?__',txt) is None)
                     else:
                         # true if txt is _not_ a _ name, false otherwise:
                         no__name = (lambda txt:
                                     re.match(r'.*\._.*?',txt) is None)
                     matches = filter(no__name, matches)
             except NameError:
                 # catches <undefined attributes>.<tab>
                 matches = []
         else:
             matches = self.global_matches(text)
 
         return matches
 
     def _default_arguments(self, obj):
         """Return the list of default arguments of obj if it is callable,
         or empty list otherwise."""
 
         if not (inspect.isfunction(obj) or inspect.ismethod(obj)):
             # for classes, check for __init__,__new__
             if inspect.isclass(obj):
                 obj = (getattr(obj,'__init__',None) or
                        getattr(obj,'__new__',None))
             # for all others, check if they are __call__able
             elif hasattr(obj, '__call__'):
                 obj = obj.__call__
             # XXX: is there a way to handle the builtins ?
         try:
             args,_,_1,defaults = inspect.getargspec(obj)
             if defaults:
                 return args[-len(defaults):]
         except TypeError: pass
         return []
 
     def python_func_kw_matches(self,text):
         """Match named parameters (kwargs) of the last open function"""
 
         if "." in text: # a parameter cannot be dotted
             return []
         try: regexp = self.__funcParamsRegex
         except AttributeError:
             regexp = self.__funcParamsRegex = re.compile(r'''
                 '.*?' |    # single quoted strings or
                 ".*?" |    # double quoted strings or
                 \w+   |    # identifier
                 \S         # other characters
                 ''', re.VERBOSE | re.DOTALL)
         # 1. find the nearest identifier that comes before an unclosed
         # parenthesis e.g. for "foo (1+bar(x), pa", the candidate is "foo"
         tokens = regexp.findall(self.get_line_buffer())
         tokens.reverse()
         iterTokens = iter(tokens); openPar = 0
         for token in iterTokens:
             if token == ')':
                 openPar -= 1
             elif token == '(':
                 openPar += 1
                 if openPar > 0:
                     # found the last unclosed parenthesis
                     break
         else:
             return []
         # 2. Concatenate dotted names ("foo.bar" for "foo.bar(x, pa" )
         ids = []
         isId = re.compile(r'\w+$').match
         while True:
             try:
                 ids.append(iterTokens.next())
                 if not isId(ids[-1]):
                     ids.pop(); break
                 if not iterTokens.next() == '.':
                     break
             except StopIteration:
                 break
         # lookup the candidate callable matches either using global_matches
         # or attr_matches for dotted names
         if len(ids) == 1:
             callableMatches = self.global_matches(ids[0])
         else:
             callableMatches = self.attr_matches('.'.join(ids[::-1]))
         argMatches = []
         for callableMatch in callableMatches:
             try:
                 namedArgs = self._default_arguments(eval(callableMatch,
                                                          self.namespace))
             except:
                 continue
             for namedArg in namedArgs:
                 if namedArg.startswith(text):
                     argMatches.append("%s=" %namedArg)
         return argMatches
 
     def dispatch_custom_completer(self,text):
         #print "Custom! '%s' %s" % (text, self.custom_completers) # dbg
         line = self.full_lbuf        
         if not line.strip():
             return None
 
         event = Bunch()
         event.line = line
         event.symbol = text
         cmd = line.split(None,1)[0]
         event.command = cmd
         #print "\ncustom:{%s]\n" % event # dbg
         
         # for foo etc, try also to find completer for %foo
         if not cmd.startswith(self.magic_escape):
             try_magic = self.custom_completers.s_matches(
               self.magic_escape + cmd)            
         else:
             try_magic = []        
         
         for c in itertools.chain(self.custom_completers.s_matches(cmd),
                                  try_magic,
                                  self.custom_completers.flat_matches(self.lbuf)):
             #print "try",c # dbg
             try:
                 res = c(event)
                 # first, try case sensitive match
                 withcase = [r for r in res if r.startswith(text)]
                 if withcase:
                     return withcase
                 # if none, then case insensitive ones are ok too
                 text_low = text.lower()
                 return [r for r in res if r.lower().startswith(text_low)]
             except TryNext:
                 pass
             
         return None
                
-    def complete(self, text, state, line_buffer=None):
-        """Return the next possible completion for 'text'.
+    def complete(self, text, line_buffer, cursor_pos=None):
+        """Return the state-th possible completion for 'text'.
 
         This is called successively with state == 0, 1, 2, ... until it
         returns None.  The completion should begin with 'text'.
 
-        :Keywords:
-        - line_buffer: string
-        If not given, the completer attempts to obtain the current line buffer
-        via readline.  This keyword allows clients which are requesting for
-        text completions in non-readline contexts to inform the completer of
-        the entire text.
-        """
-
-        #print '\n*** COMPLETE: <%s> (%s)' % (text,state)  # dbg
+        Parameters
+        ----------
+          text : string
+            Text to perform the completion on.
 
-        # if there is only a tab on a line with only whitespace, instead
-        # of the mostly useless 'do you want to see all million
-        # completions' message, just do the right thing and give the user
-        # his tab!  Incidentally, this enables pasting of tabbed text from
-        # an editor (as long as autoindent is off).
+          line_buffer : string, optional
+            If not given, the completer attempts to obtain the current line
+            buffer via readline.  This keyword allows clients which are
+            requesting for text completions in non-readline contexts to inform
+            the completer of the entire text.
 
-        # It should be noted that at least pyreadline still shows
-        # file completions - is there a way around it?
-        
-        # don't apply this on 'dumb' terminals, such as emacs buffers, so we
-        # don't interfere with their own tab-completion mechanism.
-        if line_buffer is None:
-            self.full_lbuf = self.get_line_buffer()
-        else:
-            self.full_lbuf = line_buffer
-            
-        if not (self.dumb_terminal or self.full_lbuf.strip()):
-            self.readline.insert_text('\t')
-            return None
+          cursor_pos : int, optional
+            Index of the cursor in the full line buffer.  Should be provided by
+            remote frontends where kernel has no access to frontend state.
+        """
         
         magic_escape = self.magic_escape
+        self.full_lbuf = line_buffer
+        self.lbuf = self.full_lbuf[:cursor_pos]
 
-        self.lbuf = self.full_lbuf[:self.get_endidx()]
-
-        try:
             if text.startswith('~'):
                 text = os.path.expanduser(text)
-            if state == 0:
+
+        # Start with a clean slate of completions
+        self.matches[:] = []
                 custom_res = self.dispatch_custom_completer(text)
                 if custom_res is not None:
                     # did custom completers produce something?
                     self.matches = custom_res
                 else:
                     # Extend the list of completions with the results of each
                     # matcher, so we return results to the user from all
                     # namespaces.
                     if self.merge_completions:
                         self.matches = []
                         for matcher in self.matchers:
                             self.matches.extend(matcher(text))
                     else:
                         for matcher in self.matchers:
                             self.matches = matcher(text)
                             if self.matches:
                                 break
-                    self.matches = list(set(self.matches))
+        # FIXME: we should extend our api to return a dict with completions for
+        # different types of objects.  The rlcomplete() method could then
+        # simply collapse the dict into a list for readline, but we'd have
+        # richer completion semantics in other evironments.
+        self.matches = sorted(set(self.matches))
+        #from IPython.utils.io import rprint; rprint(self.matches) # dbg
+        return self.matches
+
+    def rlcomplete(self, text, state):
+        """Return the state-th possible completion for 'text'.
+
+        This is called successively with state == 0, 1, 2, ... until it
+        returns None.  The completion should begin with 'text'.
+
+        Parameters
+        ----------
+          text : string
+            Text to perform the completion on.
+
+          state : int
+            Counter used by readline.
+
+        """
+
+        #print "rlcomplete! '%s' %s" % (text, state) # dbg
+
+        if state==0:
+            self.full_lbuf = line_buffer = self.get_line_buffer()
+            cursor_pos = self.get_endidx()
+
+            # if there is only a tab on a line with only whitespace, instead of
+            # the mostly useless 'do you want to see all million completions'
+            # message, just do the right thing and give the user his tab!
+            # Incidentally, this enables pasting of tabbed text from an editor
+            # (as long as autoindent is off).
+
+            # It should be noted that at least pyreadline still shows file
+            # completions - is there a way around it?
+
+            # don't apply this on 'dumb' terminals, such as emacs buffers, so
+            # we don't interfere with their own tab-completion mechanism.
+            if not (self.dumb_terminal or self.full_lbuf.strip()):
+                self.readline.insert_text('\t')
+                sys.stdout.flush()
+                return None
+
+            # This method computes the self.matches array
+            self.complete(text, line_buffer, cursor_pos)
+
+            # Debug version, since readline silences all exceptions making it
+            # impossible to debug any problem in the above code
+
+            ## try:
+            ##     self.complete(text, line_buffer, cursor_pos)
+            ## except:
+            ##     import traceback; traceback.print_exc()
+
             try:
-                #print "MATCH: %r" % self.matches[state] # dbg
                 return self.matches[state]
             except IndexError:
                 return None
-        except:
-            #from IPython.core.ultratb import AutoFormattedTB; # dbg
-            #tb=AutoFormattedTB('Verbose');tb() #dbg
             
-            # If completion fails, don't annoy the user.
-            return None

IPython/core/interactiveshell.py

@@ -1,2150 +1,2147 b''
 # -*- coding: utf-8 -*-
 """Main IPython class."""
 
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2001 Janko Hauser <[e-mail]>
 #  Copyright (C) 2001-2007 Fernando Perez. <[e-mail]>
 #  Copyright (C) 2008-2010  The IPython Development Team
 #
 #  Distributed under the terms of the BSD License.  The full license is in
 #  the file COPYING, distributed as part of this software.
 #-----------------------------------------------------------------------------
 
 #-----------------------------------------------------------------------------
 # Imports
 #-----------------------------------------------------------------------------
 
 from __future__ import with_statement
 from __future__ import absolute_import
 
 import __builtin__
 import abc
 import codeop
 import exceptions
 import new
 import os
 import re
 import string
 import sys
 import tempfile
 from contextlib import nested
 
 from IPython.config.configurable import Configurable
 from IPython.core import debugger, oinspect
 from IPython.core import history as ipcorehist
 from IPython.core import prefilter
 from IPython.core import shadowns
 from IPython.core import ultratb
 from IPython.core.alias import AliasManager
 from IPython.core.builtin_trap import BuiltinTrap
 from IPython.core.display_trap import DisplayTrap
 from IPython.core.displayhook import DisplayHook
 from IPython.core.error import UsageError
 from IPython.core.extensions import ExtensionManager
 from IPython.core.fakemodule import FakeModule, init_fakemod_dict
 from IPython.core.inputlist import InputList
 from IPython.core.logger import Logger
 from IPython.core.magic import Magic
 from IPython.core.payload import PayloadManager
 from IPython.core.plugin import PluginManager
 from IPython.core.prefilter import PrefilterManager
 from IPython.external.Itpl import ItplNS
 from IPython.utils import PyColorize
 from IPython.utils import io
 from IPython.utils import pickleshare
 from IPython.utils.doctestreload import doctest_reload
 from IPython.utils.io import ask_yes_no, rprint
 from IPython.utils.ipstruct import Struct
 from IPython.utils.path import get_home_dir, get_ipython_dir, HomeDirError
 from IPython.utils.process import getoutput, getoutputerror
 from IPython.utils.strdispatch import StrDispatch
 from IPython.utils.syspathcontext import prepended_to_syspath
 from IPython.utils.text import num_ini_spaces
 from IPython.utils.traitlets import (Int, Str, CBool, CaselessStrEnum, Enum,
                                      List, Unicode, Instance, Type)
 from IPython.utils.warn import warn, error, fatal
 import IPython.core.hooks
 
 # from IPython.utils import growl
 # growl.start("IPython")
 
 #-----------------------------------------------------------------------------
 # Globals
 #-----------------------------------------------------------------------------
 
 # compiled regexps for autoindent management
 dedent_re = re.compile(r'^\s+raise|^\s+return|^\s+pass')
 
 #-----------------------------------------------------------------------------
 # Utilities
 #-----------------------------------------------------------------------------
 
 # store the builtin raw_input globally, and use this always, in case user code
 # overwrites it (like wx.py.PyShell does)
 raw_input_original = raw_input
 
 def softspace(file, newvalue):
     """Copied from code.py, to remove the dependency"""
 
     oldvalue = 0
     try:
         oldvalue = file.softspace
     except AttributeError:
         pass
     try:
         file.softspace = newvalue
     except (AttributeError, TypeError):
         # "attribute-less object" or "read-only attributes"
         pass
     return oldvalue
 
 
 def no_op(*a, **kw): pass
 
 class SpaceInInput(exceptions.Exception): pass
 
 class Bunch: pass
 
 
 def get_default_colors():
     if sys.platform=='darwin':
         return "LightBG"
     elif os.name=='nt':
         return 'Linux'
     else:
         return 'Linux'
 
 
 class SeparateStr(Str):
     """A Str subclass to validate separate_in, separate_out, etc.
 
     This is a Str based trait that converts '0'->'' and '\\n'->'\n'.
     """
 
     def validate(self, obj, value):
         if value == '0': value = ''
         value = value.replace('\\n','\n')
         return super(SeparateStr, self).validate(obj, value)
 
 class MultipleInstanceError(Exception):
     pass
 
 
 #-----------------------------------------------------------------------------
 # Main IPython class
 #-----------------------------------------------------------------------------
 
 
 class InteractiveShell(Configurable, Magic):
     """An enhanced, interactive shell for Python."""
 
     _instance = None
     autocall = Enum((0,1,2), default_value=1, config=True)
     # TODO: remove all autoindent logic and put into frontends.
     # We can't do this yet because even runlines uses the autoindent.
     autoindent = CBool(True, config=True)
     automagic = CBool(True, config=True)
     cache_size = Int(1000, config=True)
     color_info = CBool(True, config=True)
     colors = CaselessStrEnum(('NoColor','LightBG','Linux'), 
                              default_value=get_default_colors(), config=True)
     debug = CBool(False, config=True)
     deep_reload = CBool(False, config=True)
     displayhook_class = Type(DisplayHook)
     filename = Str("<ipython console>")
     ipython_dir= Unicode('', config=True) # Set to get_ipython_dir() in __init__
     logstart = CBool(False, config=True)
     logfile = Str('', config=True)
     logappend = Str('', config=True)
     object_info_string_level = Enum((0,1,2), default_value=0,
                                     config=True)
     pdb = CBool(False, config=True)
     pprint = CBool(True, config=True)
     profile = Str('', config=True)
     prompt_in1 = Str('In [\\#]: ', config=True)
     prompt_in2 = Str('   .\\D.: ', config=True)
     prompt_out = Str('Out[\\#]: ', config=True)
     prompts_pad_left = CBool(True, config=True)
     quiet = CBool(False, config=True)
 
     # The readline stuff will eventually be moved to the terminal subclass
     # but for now, we can't do that as readline is welded in everywhere.
     readline_use = CBool(True, config=True)
     readline_merge_completions = CBool(True, config=True)
     readline_omit__names = Enum((0,1,2), default_value=0, config=True)
     readline_remove_delims = Str('-/~', config=True)
     readline_parse_and_bind = List([
             'tab: complete',
             '"\C-l": clear-screen',
             'set show-all-if-ambiguous on',
             '"\C-o": tab-insert',
             '"\M-i": "    "',
             '"\M-o": "\d\d\d\d"',
             '"\M-I": "\d\d\d\d"',
             '"\C-r": reverse-search-history',
             '"\C-s": forward-search-history',
             '"\C-p": history-search-backward',
             '"\C-n": history-search-forward',
             '"\e[A": history-search-backward',
             '"\e[B": history-search-forward',
             '"\C-k": kill-line',
             '"\C-u": unix-line-discard',
         ], allow_none=False, config=True)
 
     # TODO: this part of prompt management should be moved to the frontends.
     # Use custom TraitTypes that convert '0'->'' and '\\n'->'\n'
     separate_in = SeparateStr('\n', config=True)
     separate_out = SeparateStr('\n', config=True)
     separate_out2 = SeparateStr('\n', config=True)
     system_header = Str('IPython system call: ', config=True)
     system_verbose = CBool(False, config=True)
     wildcards_case_sensitive = CBool(True, config=True)
     xmode = CaselessStrEnum(('Context','Plain', 'Verbose'), 
                             default_value='Context', config=True)
 
     # Subcomponents of InteractiveShell
     alias_manager = Instance('IPython.core.alias.AliasManager')
     prefilter_manager = Instance('IPython.core.prefilter.PrefilterManager')
     builtin_trap = Instance('IPython.core.builtin_trap.BuiltinTrap')
     display_trap = Instance('IPython.core.display_trap.DisplayTrap')
     extension_manager = Instance('IPython.core.extensions.ExtensionManager')
     plugin_manager = Instance('IPython.core.plugin.PluginManager')
     payload_manager = Instance('IPython.core.payload.PayloadManager')
 
     def __init__(self, config=None, ipython_dir=None,
                  user_ns=None, user_global_ns=None,
                  custom_exceptions=((),None)):
 
         # This is where traits with a config_key argument are updated
         # from the values on config.
         super(InteractiveShell, self).__init__(config=config)
 
         # These are relatively independent and stateless
         self.init_ipython_dir(ipython_dir)
         self.init_instance_attrs()
 
         # Create namespaces (user_ns, user_global_ns, etc.)
         self.init_create_namespaces(user_ns, user_global_ns)
         # This has to be done after init_create_namespaces because it uses
         # something in self.user_ns, but before init_sys_modules, which
         # is the first thing to modify sys.
         # TODO: When we override sys.stdout and sys.stderr before this class
         # is created, we are saving the overridden ones here. Not sure if this
         # is what we want to do.
         self.save_sys_module_state()
         self.init_sys_modules()
 
         self.init_history()
         self.init_encoding()
         self.init_prefilter()
 
         Magic.__init__(self, self)
 
         self.init_syntax_highlighting()
         self.init_hooks()
         self.init_pushd_popd_magic()
         # self.init_traceback_handlers use to be here, but we moved it below
         # because it and init_io have to come after init_readline.
         self.init_user_ns()
         self.init_logger()
         self.init_alias()
         self.init_builtins()
 
         # pre_config_initialization
         self.init_shadow_hist()
 
         # The next section should contain averything that was in ipmaker.
         self.init_logstart()
 
         # The following was in post_config_initialization
         self.init_inspector()
         # init_readline() must come before init_io(), because init_io uses
         # readline related things.
         self.init_readline()
         # TODO: init_io() needs to happen before init_traceback handlers
         # because the traceback handlers hardcode the stdout/stderr streams.
         # This logic in in debugger.Pdb and should eventually be changed.
         self.init_io()
         self.init_traceback_handlers(custom_exceptions)
         self.init_prompts()
         self.init_displayhook()
         self.init_reload_doctest()
         self.init_magics()
         self.init_pdb()
         self.init_extension_manager()
         self.init_plugin_manager()
         self.init_payload()
         self.hooks.late_startup_hook()
 
     @classmethod
     def instance(cls, *args, **kwargs):
         """Returns a global InteractiveShell instance."""
         if cls._instance is None:
             inst = cls(*args, **kwargs)
             # Now make sure that the instance will also be returned by
             # the subclasses instance attribute.
             for subclass in cls.mro():
                 if issubclass(cls, subclass) and issubclass(subclass, InteractiveShell):
                     subclass._instance = inst
                 else:
                     break
         if isinstance(cls._instance, cls):
             return cls._instance
         else:
             raise MultipleInstanceError(
                 'Multiple incompatible subclass instances of '
                 'InteractiveShell are being created.'
             )
 
     @classmethod
     def initialized(cls):
         return hasattr(cls, "_instance")
 
     def get_ipython(self):
         """Return the currently running IPython instance."""
         return self
 
     #-------------------------------------------------------------------------
     # Trait changed handlers
     #-------------------------------------------------------------------------
 
     def _ipython_dir_changed(self, name, new):
         if not os.path.isdir(new):
             os.makedirs(new, mode = 0777)
 
     def set_autoindent(self,value=None):
         """Set the autoindent flag, checking for readline support.
 
         If called with no arguments, it acts as a toggle."""
 
         if not self.has_readline:
             if os.name == 'posix':
                 warn("The auto-indent feature requires the readline library")
             self.autoindent = 0
             return
         if value is None:
             self.autoindent = not self.autoindent
         else:
             self.autoindent = value
 
     #-------------------------------------------------------------------------
     # init_* methods called by __init__
     #-------------------------------------------------------------------------
 
     def init_ipython_dir(self, ipython_dir):
         if ipython_dir is not None:
             self.ipython_dir = ipython_dir
             self.config.Global.ipython_dir = self.ipython_dir
             return
 
         if hasattr(self.config.Global, 'ipython_dir'):
             self.ipython_dir = self.config.Global.ipython_dir
         else:
             self.ipython_dir = get_ipython_dir()
 
         # All children can just read this
         self.config.Global.ipython_dir = self.ipython_dir
 
     def init_instance_attrs(self):
         self.more = False
 
         # command compiler
         self.compile = codeop.CommandCompiler()
 
         # User input buffer
         self.buffer = []
 
         # Make an empty namespace, which extension writers can rely on both
         # existing and NEVER being used by ipython itself.  This gives them a
         # convenient location for storing additional information and state
         # their extensions may require, without fear of collisions with other
         # ipython names that may develop later.
         self.meta = Struct()
 
         # Object variable to store code object waiting execution.  This is
         # used mainly by the multithreaded shells, but it can come in handy in
         # other situations.  No need to use a Queue here, since it's a single
         # item which gets cleared once run.
         self.code_to_run = None
 
         # Temporary files used for various purposes.  Deleted at exit.
         self.tempfiles = []
 
         # Keep track of readline usage (later set by init_readline)
         self.has_readline = False
 
         # keep track of where we started running (mainly for crash post-mortem)
         # This is not being used anywhere currently.
         self.starting_dir = os.getcwd()
 
         # Indentation management
         self.indent_current_nsp = 0
 
     def init_encoding(self):
         # Get system encoding at startup time.  Certain terminals (like Emacs
         # under Win32 have it set to None, and we need to have a known valid
         # encoding to use in the raw_input() method
         try:
             self.stdin_encoding = sys.stdin.encoding or 'ascii'
         except AttributeError:
             self.stdin_encoding = 'ascii'
 
     def init_syntax_highlighting(self):
         # Python source parser/formatter for syntax highlighting
         pyformat = PyColorize.Parser().format
         self.pycolorize = lambda src: pyformat(src,'str',self.colors)
 
     def init_pushd_popd_magic(self):
         # for pushd/popd management
         try:
             self.home_dir = get_home_dir()
         except HomeDirError, msg:
             fatal(msg)
 
         self.dir_stack = []
 
     def init_logger(self):
         self.logger = Logger(self, logfname='ipython_log.py', logmode='rotate')
         # local shortcut, this is used a LOT
         self.log = self.logger.log
 
     def init_logstart(self):
         if self.logappend:
             self.magic_logstart(self.logappend + ' append')
         elif self.logfile:
             self.magic_logstart(self.logfile)
         elif self.logstart:
             self.magic_logstart()
 
     def init_builtins(self):
         self.builtin_trap = BuiltinTrap(shell=self)
 
     def init_inspector(self):
         # Object inspector
         self.inspector = oinspect.Inspector(oinspect.InspectColors,
                                             PyColorize.ANSICodeColors,
                                             'NoColor',
                                             self.object_info_string_level)
 
     def init_io(self):
         import IPython.utils.io
         if sys.platform == 'win32' and self.has_readline:
             Term = io.IOTerm(
                 cout=self.readline._outputfile,cerr=self.readline._outputfile
             )
         else:
             Term = io.IOTerm()
         io.Term = Term
 
     def init_prompts(self):
         # TODO: This is a pass for now because the prompts are managed inside
         # the DisplayHook. Once there is a separate prompt manager, this 
         # will initialize that object and all prompt related information.
         pass
 
     def init_displayhook(self):
         # Initialize displayhook, set in/out prompts and printing system
         self.displayhook = self.displayhook_class(
             shell=self,
             cache_size=self.cache_size,
             input_sep = self.separate_in,
             output_sep = self.separate_out,
             output_sep2 = self.separate_out2,
             ps1 = self.prompt_in1,
             ps2 = self.prompt_in2,
             ps_out = self.prompt_out,
             pad_left = self.prompts_pad_left
         )
         # This is a context manager that installs/revmoes the displayhook at
         # the appropriate time.
         self.display_trap = DisplayTrap(hook=self.displayhook)
 
     def init_reload_doctest(self):
         # Do a proper resetting of doctest, including the necessary displayhook
         # monkeypatching
         try:
             doctest_reload()
         except ImportError:
             warn("doctest module does not exist.")
 
     #-------------------------------------------------------------------------
     # Things related to injections into the sys module
     #-------------------------------------------------------------------------
 
     def save_sys_module_state(self):
         """Save the state of hooks in the sys module.
 
         This has to be called after self.user_ns is created.
         """
         self._orig_sys_module_state = {}
         self._orig_sys_module_state['stdin'] = sys.stdin
         self._orig_sys_module_state['stdout'] = sys.stdout
         self._orig_sys_module_state['stderr'] = sys.stderr
         self._orig_sys_module_state['excepthook'] = sys.excepthook
         try:
             self._orig_sys_modules_main_name = self.user_ns['__name__']
         except KeyError:
             pass
 
     def restore_sys_module_state(self):
         """Restore the state of the sys module."""
         try:
             for k, v in self._orig_sys_module_state.items():
                 setattr(sys, k, v)
         except AttributeError:
             pass
         try:
             delattr(sys, 'ipcompleter')
         except AttributeError:
             pass
         # Reset what what done in self.init_sys_modules
         try:
             sys.modules[self.user_ns['__name__']] = self._orig_sys_modules_main_name
         except (AttributeError, KeyError):
             pass
 
     #-------------------------------------------------------------------------
     # Things related to hooks
     #-------------------------------------------------------------------------
 
     def init_hooks(self):
         # hooks holds pointers used for user-side customizations
         self.hooks = Struct()
 
         self.strdispatchers = {}
 
         # Set all default hooks, defined in the IPython.hooks module.
         hooks = IPython.core.hooks
         for hook_name in hooks.__all__:
             # default hooks have priority 100, i.e. low; user hooks should have
             # 0-100 priority
             self.set_hook(hook_name,getattr(hooks,hook_name), 100)
 
     def set_hook(self,name,hook, priority = 50, str_key = None, re_key = None):
         """set_hook(name,hook) -> sets an internal IPython hook.
 
         IPython exposes some of its internal API as user-modifiable hooks.  By
         adding your function to one of these hooks, you can modify IPython's 
         behavior to call at runtime your own routines."""
 
         # At some point in the future, this should validate the hook before it
         # accepts it.  Probably at least check that the hook takes the number
         # of args it's supposed to.
         
         f = new.instancemethod(hook,self,self.__class__)
 
         # check if the hook is for strdispatcher first
         if str_key is not None:
             sdp = self.strdispatchers.get(name, StrDispatch())
             sdp.add_s(str_key, f, priority )
             self.strdispatchers[name] = sdp
             return
         if re_key is not None:
             sdp = self.strdispatchers.get(name, StrDispatch())
             sdp.add_re(re.compile(re_key), f, priority )
             self.strdispatchers[name] = sdp
             return
             
         dp = getattr(self.hooks, name, None)
         if name not in IPython.core.hooks.__all__:
             print "Warning! Hook '%s' is not one of %s" % (name, IPython.core.hooks.__all__ )
         if not dp:
             dp = IPython.core.hooks.CommandChainDispatcher()
         
         try:
             dp.add(f,priority)
         except AttributeError:
             # it was not commandchain, plain old func - replace
             dp = f
 
         setattr(self.hooks,name, dp)
 
     #-------------------------------------------------------------------------
     # Things related to the "main" module
     #-------------------------------------------------------------------------
 
     def new_main_mod(self,ns=None):
         """Return a new 'main' module object for user code execution.
         """
         main_mod = self._user_main_module
         init_fakemod_dict(main_mod,ns)
         return main_mod
 
     def cache_main_mod(self,ns,fname):
         """Cache a main module's namespace.
 
         When scripts are executed via %run, we must keep a reference to the
         namespace of their __main__ module (a FakeModule instance) around so
         that Python doesn't clear it, rendering objects defined therein
         useless.
 
         This method keeps said reference in a private dict, keyed by the
         absolute path of the module object (which corresponds to the script
         path).  This way, for multiple executions of the same script we only
         keep one copy of the namespace (the last one), thus preventing memory
         leaks from old references while allowing the objects from the last
         execution to be accessible.
 
         Note: we can not allow the actual FakeModule instances to be deleted,
         because of how Python tears down modules (it hard-sets all their
         references to None without regard for reference counts).  This method
         must therefore make a *copy* of the given namespace, to allow the
         original module's __dict__ to be cleared and reused.
 
         
         Parameters
         ----------
           ns : a namespace (a dict, typically)
 
           fname : str
             Filename associated with the namespace.
 
         Examples
         --------
 
         In [10]: import IPython
 
         In [11]: _ip.cache_main_mod(IPython.__dict__,IPython.__file__)
 
         In [12]: IPython.__file__ in _ip._main_ns_cache
         Out[12]: True
         """
         self._main_ns_cache[os.path.abspath(fname)] = ns.copy()
 
     def clear_main_mod_cache(self):
         """Clear the cache of main modules.
 
         Mainly for use by utilities like %reset.
 
         Examples
         --------
 
         In [15]: import IPython
 
         In [16]: _ip.cache_main_mod(IPython.__dict__,IPython.__file__)
 
         In [17]: len(_ip._main_ns_cache) > 0
         Out[17]: True
 
         In [18]: _ip.clear_main_mod_cache()
 
         In [19]: len(_ip._main_ns_cache) == 0
         Out[19]: True
         """
         self._main_ns_cache.clear()
 
     #-------------------------------------------------------------------------
     # Things related to debugging
     #-------------------------------------------------------------------------
 
     def init_pdb(self):
         # Set calling of pdb on exceptions
         # self.call_pdb is a property
         self.call_pdb = self.pdb
 
     def _get_call_pdb(self):
         return self._call_pdb
 
     def _set_call_pdb(self,val):
 
         if val not in (0,1,False,True):
             raise ValueError,'new call_pdb value must be boolean'
 
         # store value in instance
         self._call_pdb = val
 
         # notify the actual exception handlers
         self.InteractiveTB.call_pdb = val
 
     call_pdb = property(_get_call_pdb,_set_call_pdb,None,
                         'Control auto-activation of pdb at exceptions')
 
     def debugger(self,force=False):
         """Call the pydb/pdb debugger.
 
         Keywords:
 
           - force(False): by default, this routine checks the instance call_pdb
           flag and does not actually invoke the debugger if the flag is false.
           The 'force' option forces the debugger to activate even if the flag
           is false.
         """
 
         if not (force or self.call_pdb):
             return
 
         if not hasattr(sys,'last_traceback'):
             error('No traceback has been produced, nothing to debug.')
             return
 
         # use pydb if available
         if debugger.has_pydb:
             from pydb import pm
         else:
             # fallback to our internal debugger
             pm = lambda : self.InteractiveTB.debugger(force=True)
         self.history_saving_wrapper(pm)()
 
     #-------------------------------------------------------------------------
     # Things related to IPython's various namespaces
     #-------------------------------------------------------------------------
 
     def init_create_namespaces(self, user_ns=None, user_global_ns=None):
         # Create the namespace where the user will operate.  user_ns is
         # normally the only one used, and it is passed to the exec calls as
         # the locals argument.  But we do carry a user_global_ns namespace
         # given as the exec 'globals' argument,  This is useful in embedding
         # situations where the ipython shell opens in a context where the
         # distinction between locals and globals is meaningful.  For
         # non-embedded contexts, it is just the same object as the user_ns dict.
 
         # FIXME. For some strange reason, __builtins__ is showing up at user
         # level as a dict instead of a module. This is a manual fix, but I
         # should really track down where the problem is coming from. Alex
         # Schmolck reported this problem first.
 
         # A useful post by Alex Martelli on this topic:
         # Re: inconsistent value from __builtins__
         # Von: Alex Martelli <[e-mail]>
         # Datum: Freitag 01 Oktober 2004 04:45:34 nachmittags/abends
         # Gruppen: comp.lang.python
 
         # Michael Hohn <[e-mail]> wrote:
         # > >>> print type(builtin_check.get_global_binding('__builtins__'))
         # > <type 'dict'>
         # > >>> print type(__builtins__)
         # > <type 'module'>
         # > Is this difference in return value intentional?
 
         # Well, it's documented that '__builtins__' can be either a dictionary
         # or a module, and it's been that way for a long time. Whether it's
         # intentional (or sensible), I don't know. In any case, the idea is
         # that if you need to access the built-in namespace directly, you
         # should start with "import __builtin__" (note, no 's') which will
         # definitely give you a module. Yeah, it's somewhat confusing:-(.
 
         # These routines return properly built dicts as needed by the rest of
         # the code, and can also be used by extension writers to generate
         # properly initialized namespaces.
         user_ns, user_global_ns = self.make_user_namespaces(user_ns, user_global_ns)
 
         # Assign namespaces
         # This is the namespace where all normal user variables live
         self.user_ns = user_ns
         self.user_global_ns = user_global_ns
 
         # An auxiliary namespace that checks what parts of the user_ns were
         # loaded at startup, so we can list later only variables defined in
         # actual interactive use.  Since it is always a subset of user_ns, it
         # doesn't need to be separately tracked in the ns_table.
         self.user_ns_hidden = {}
 
         # A namespace to keep track of internal data structures to prevent
         # them from cluttering user-visible stuff.  Will be updated later
         self.internal_ns = {}
 
         # Now that FakeModule produces a real module, we've run into a nasty
         # problem: after script execution (via %run), the module where the user
         # code ran is deleted.  Now that this object is a true module (needed
         # so docetst and other tools work correctly), the Python module
         # teardown mechanism runs over it, and sets to None every variable
         # present in that module.  Top-level references to objects from the
         # script survive, because the user_ns is updated with them.  However,
         # calling functions defined in the script that use other things from
         # the script will fail, because the function's closure had references
         # to the original objects, which are now all None.  So we must protect
         # these modules from deletion by keeping a cache.
         # 
         # To avoid keeping stale modules around (we only need the one from the
         # last run), we use a dict keyed with the full path to the script, so
         # only the last version of the module is held in the cache.  Note,
         # however, that we must cache the module *namespace contents* (their
         # __dict__).  Because if we try to cache the actual modules, old ones
         # (uncached) could be destroyed while still holding references (such as
         # those held by GUI objects that tend to be long-lived)>
         # 
         # The %reset command will flush this cache.  See the cache_main_mod()
         # and clear_main_mod_cache() methods for details on use.
 
         # This is the cache used for 'main' namespaces
         self._main_ns_cache = {}
         # And this is the single instance of FakeModule whose __dict__ we keep
         # copying and clearing for reuse on each %run
         self._user_main_module = FakeModule()
 
         # A table holding all the namespaces IPython deals with, so that
         # introspection facilities can search easily.
         self.ns_table = {'user':user_ns,
                          'user_global':user_global_ns,
                          'internal':self.internal_ns,
                          'builtin':__builtin__.__dict__
                          }
 
         # Similarly, track all namespaces where references can be held and that
         # we can safely clear (so it can NOT include builtin).  This one can be
         # a simple list.
         self.ns_refs_table = [ user_ns, user_global_ns, self.user_ns_hidden,
                                self.internal_ns, self._main_ns_cache ]
 
     def make_user_namespaces(self, user_ns=None, user_global_ns=None):
         """Return a valid local and global user interactive namespaces.
 
         This builds a dict with the minimal information needed to operate as a
         valid IPython user namespace, which you can pass to the various
         embedding classes in ipython. The default implementation returns the
         same dict for both the locals and the globals to allow functions to
         refer to variables in the namespace. Customized implementations can
         return different dicts. The locals dictionary can actually be anything
         following the basic mapping protocol of a dict, but the globals dict
         must be a true dict, not even a subclass. It is recommended that any
         custom object for the locals namespace synchronize with the globals
         dict somehow.
 
         Raises TypeError if the provided globals namespace is not a true dict.
 
         Parameters
         ----------
         user_ns : dict-like, optional
             The current user namespace. The items in this namespace should
             be included in the output. If None, an appropriate blank
             namespace should be created.
         user_global_ns : dict, optional
             The current user global namespace. The items in this namespace
             should be included in the output. If None, an appropriate
             blank namespace should be created.
 
         Returns
         -------
             A pair of dictionary-like object to be used as the local namespace
             of the interpreter and a dict to be used as the global namespace.
         """
 
 
         # We must ensure that __builtin__ (without the final 's') is always
         # available and pointing to the __builtin__ *module*.  For more details:
         # http://mail.python.org/pipermail/python-dev/2001-April/014068.html
 
         if user_ns is None:
             # Set __name__ to __main__ to better match the behavior of the
             # normal interpreter.
             user_ns = {'__name__'     :'__main__',
                        '__builtin__' : __builtin__,
                        '__builtins__' : __builtin__,
                       }
         else:
             user_ns.setdefault('__name__','__main__')
             user_ns.setdefault('__builtin__',__builtin__)
             user_ns.setdefault('__builtins__',__builtin__)
 
         if user_global_ns is None:
             user_global_ns = user_ns
         if type(user_global_ns) is not dict:
             raise TypeError("user_global_ns must be a true dict; got %r"
                 % type(user_global_ns))
 
         return user_ns, user_global_ns
 
     def init_sys_modules(self):
         # We need to insert into sys.modules something that looks like a
         # module but which accesses the IPython namespace, for shelve and
         # pickle to work interactively. Normally they rely on getting
         # everything out of __main__, but for embedding purposes each IPython
         # instance has its own private namespace, so we can't go shoving
         # everything into __main__.
 
         # note, however, that we should only do this for non-embedded
         # ipythons, which really mimic the __main__.__dict__ with their own
         # namespace.  Embedded instances, on the other hand, should not do
         # this because they need to manage the user local/global namespaces
         # only, but they live within a 'normal' __main__ (meaning, they
         # shouldn't overtake the execution environment of the script they're
         # embedded in).
 
         # This is overridden in the InteractiveShellEmbed subclass to a no-op.
 
         try:
             main_name = self.user_ns['__name__']
         except KeyError:
             raise KeyError('user_ns dictionary MUST have a "__name__" key')
         else:
             sys.modules[main_name] = FakeModule(self.user_ns)
 
     def init_user_ns(self):
         """Initialize all user-visible namespaces to their minimum defaults.
 
         Certain history lists are also initialized here, as they effectively
         act as user namespaces.
 
         Notes
         -----
         All data structures here are only filled in, they are NOT reset by this
         method.  If they were not empty before, data will simply be added to
         therm.
         """
         # This function works in two parts: first we put a few things in
         # user_ns, and we sync that contents into user_ns_hidden so that these
         # initial variables aren't shown by %who.  After the sync, we add the
         # rest of what we *do* want the user to see with %who even on a new
         # session (probably nothing, so theye really only see their own stuff)
 
         # The user dict must *always* have a __builtin__ reference to the
         # Python standard __builtin__ namespace,  which must be imported.
         # This is so that certain operations in prompt evaluation can be
         # reliably executed with builtins.  Note that we can NOT use
         # __builtins__ (note the 's'),  because that can either be a dict or a
         # module, and can even mutate at runtime, depending on the context
         # (Python makes no guarantees on it).  In contrast, __builtin__ is
         # always a module object, though it must be explicitly imported.
         
         # For more details:
         # http://mail.python.org/pipermail/python-dev/2001-April/014068.html
         ns = dict(__builtin__ = __builtin__)
         
         # Put 'help' in the user namespace
         try:
             from site import _Helper
             ns['help'] = _Helper()
         except ImportError:
             warn('help() not available - check site.py')
 
         # make global variables for user access to the histories
         ns['_ih'] = self.input_hist
         ns['_oh'] = self.output_hist
         ns['_dh'] = self.dir_hist
 
         ns['_sh'] = shadowns
 
         # user aliases to input and output histories.  These shouldn't show up
         # in %who, as they can have very large reprs.
         ns['In']  = self.input_hist
         ns['Out'] = self.output_hist
 
         # Store myself as the public api!!!
         ns['get_ipython'] = self.get_ipython
 
         # Sync what we've added so far to user_ns_hidden so these aren't seen
         # by %who
         self.user_ns_hidden.update(ns)
 
         # Anything put into ns now would show up in %who.  Think twice before
         # putting anything here, as we really want %who to show the user their
         # stuff, not our variables.
         
         # Finally, update the real user's namespace
         self.user_ns.update(ns)
 
 
     def reset(self):
         """Clear all internal namespaces.
 
         Note that this is much more aggressive than %reset, since it clears
         fully all namespaces, as well as all input/output lists.
         """
         for ns in self.ns_refs_table:
             ns.clear()
 
         self.alias_manager.clear_aliases()
 
         # Clear input and output histories
         self.input_hist[:] = []
         self.input_hist_raw[:] = []
         self.output_hist.clear()
 
         # Restore the user namespaces to minimal usability
         self.init_user_ns()
 
         # Restore the default and user aliases
         self.alias_manager.init_aliases()
 
     def reset_selective(self, regex=None):
         """Clear selective variables from internal namespaces based on a specified regular expression.
 
         Parameters
         ----------
         regex : string or compiled pattern, optional
             A regular expression pattern that will be used in searching variable names in the users 
             namespaces.
         """
         if regex is not None:
             try:
                 m = re.compile(regex)
             except TypeError:
                 raise TypeError('regex must be a string or compiled pattern')
             # Search for keys in each namespace that match the given regex
             # If a match is found, delete the key/value pair.
             for ns in self.ns_refs_table:
                 for var in ns:
                     if m.search(var):
                         del ns[var]        
         
     def push(self, variables, interactive=True):
         """Inject a group of variables into the IPython user namespace.
 
         Parameters
         ----------
         variables : dict, str or list/tuple of str
             The variables to inject into the user's namespace.  If a dict,
             a simple update is done.  If a str, the string is assumed to 
             have variable names separated by spaces.  A list/tuple of str
             can also be used to give the variable names.  If just the variable
             names are give (list/tuple/str) then the variable values looked
             up in the callers frame.
         interactive : bool
             If True (default), the variables will be listed with the ``who``
             magic.
         """
         vdict = None
 
         # We need a dict of name/value pairs to do namespace updates.
         if isinstance(variables, dict):
             vdict = variables
         elif isinstance(variables, (basestring, list, tuple)):
             if isinstance(variables, basestring):
                 vlist = variables.split()
             else:
                 vlist = variables
             vdict = {}
             cf = sys._getframe(1)
             for name in vlist:
                 try:
                     vdict[name] = eval(name, cf.f_globals, cf.f_locals)
                 except:
                     print ('Could not get variable %s from %s' %
                            (name,cf.f_code.co_name))
         else:
             raise ValueError('variables must be a dict/str/list/tuple')
             
         # Propagate variables to user namespace
         self.user_ns.update(vdict)
 
         # And configure interactive visibility
         config_ns = self.user_ns_hidden
         if interactive:
             for name, val in vdict.iteritems():
                 config_ns.pop(name, None)
         else:
             for name,val in vdict.iteritems():
                 config_ns[name] = val
 
     #-------------------------------------------------------------------------
     # Things related to history management
     #-------------------------------------------------------------------------
 
     def init_history(self):
         # List of input with multi-line handling.
         self.input_hist = InputList()
         # This one will hold the 'raw' input history, without any
         # pre-processing.  This will allow users to retrieve the input just as
         # it was exactly typed in by the user, with %hist -r.
         self.input_hist_raw = InputList()
 
         # list of visited directories
         try:
             self.dir_hist = [os.getcwd()]
         except OSError:
             self.dir_hist = []
 
         # dict of output history
         self.output_hist = {}
 
         # Now the history file
         if self.profile:
             histfname = 'history-%s' % self.profile
         else:
             histfname = 'history'
         self.histfile = os.path.join(self.ipython_dir, histfname)
 
         # Fill the history zero entry, user counter starts at 1
         self.input_hist.append('\n')
         self.input_hist_raw.append('\n')
 
     def init_shadow_hist(self):
         try:
             self.db = pickleshare.PickleShareDB(self.ipython_dir + "/db")
         except exceptions.UnicodeDecodeError:
             print "Your ipython_dir can't be decoded to unicode!"
             print "Please set HOME environment variable to something that"
             print r"only has ASCII characters, e.g. c:\home"
             print "Now it is", self.ipython_dir
             sys.exit()
         self.shadowhist = ipcorehist.ShadowHist(self.db)
 
     def savehist(self):
         """Save input history to a file (via readline library)."""
 
         try:
             self.readline.write_history_file(self.histfile)
         except:
             print 'Unable to save IPython command history to file: ' + \
                   `self.histfile`
 
     def reloadhist(self):
         """Reload the input history from disk file."""
 
         try:
             self.readline.clear_history()
             self.readline.read_history_file(self.shell.histfile)
         except AttributeError:
             pass
 
     def history_saving_wrapper(self, func):
         """ Wrap func for readline history saving
 
         Convert func into callable that saves & restores
         history around the call """
 
         if self.has_readline:
             from IPython.utils import rlineimpl as readline
         else:
             return func
 
         def wrapper():
             self.savehist()
             try:
                 func()
             finally:
                 readline.read_history_file(self.histfile)
         return wrapper
 
     def get_history(self, index=None, raw=False, output=True):
         """Get the history list.
 
         Get the input and output history.
 
         Parameters
         ----------
         index : n or (n1, n2) or None
             If n, then the last entries. If a tuple, then all in
             range(n1, n2). If None, then all entries. Raises IndexError if
             the format of index is incorrect.
         raw : bool
             If True, return the raw input.
         output : bool
             If True, then return the output as well.
 
         Returns
         -------
         If output is True, then return a dict of tuples, keyed by the prompt
         numbers and with values of (input, output). If output is False, then
         a dict, keyed by the prompt number with the values of input. Raises
         IndexError if no history is found.
         """
         if raw:
             input_hist = self.input_hist_raw
         else:
             input_hist = self.input_hist
         if output:
             output_hist = self.user_ns['Out']
         n = len(input_hist)
         if index is None:
             start=0; stop=n
         elif isinstance(index, int):
             start=n-index; stop=n
         elif isinstance(index, tuple) and len(index) == 2:
             start=index[0]; stop=index[1]
         else:
             raise IndexError('Not a valid index for the input history: %r' % index)
         hist = {}
         for i in range(start, stop):
             if output:
                 hist[i] = (input_hist[i], output_hist.get(i))
             else:
                 hist[i] = input_hist[i]
         if len(hist)==0:
             raise IndexError('No history for range of indices: %r' % index)
         return hist
 
     #-------------------------------------------------------------------------
     # Things related to exception handling and tracebacks (not debugging)
     #-------------------------------------------------------------------------
 
     def init_traceback_handlers(self, custom_exceptions):
         # Syntax error handler.
         self.SyntaxTB = ultratb.SyntaxTB(color_scheme='NoColor')
         
         # The interactive one is initialized with an offset, meaning we always
         # want to remove the topmost item in the traceback, which is our own
         # internal code. Valid modes: ['Plain','Context','Verbose']
         self.InteractiveTB = ultratb.AutoFormattedTB(mode = 'Plain',
                                                      color_scheme='NoColor',
                                                      tb_offset = 1)
 
         # The instance will store a pointer to the system-wide exception hook,
         # so that runtime code (such as magics) can access it.  This is because
         # during the read-eval loop, it may get temporarily overwritten.
         self.sys_excepthook = sys.excepthook
 
         # and add any custom exception handlers the user may have specified
         self.set_custom_exc(*custom_exceptions)
 
         # Set the exception mode
         self.InteractiveTB.set_mode(mode=self.xmode)
 
     def set_custom_exc(self, exc_tuple, handler):
         """set_custom_exc(exc_tuple,handler)
 
         Set a custom exception handler, which will be called if any of the
         exceptions in exc_tuple occur in the mainloop (specifically, in the
         runcode() method.
 
         Inputs:
 
           - exc_tuple: a *tuple* of valid exceptions to call the defined
           handler for.  It is very important that you use a tuple, and NOT A
           LIST here, because of the way Python's except statement works.  If
           you only want to trap a single exception, use a singleton tuple:
 
             exc_tuple == (MyCustomException,)
 
           - handler: this must be defined as a function with the following
           basic interface::
 
             def my_handler(self, etype, value, tb, tb_offset=None)
                 ...
                 # The return value must be
                 return structured_traceback
 
           This will be made into an instance method (via new.instancemethod)
           of IPython itself, and it will be called if any of the exceptions
           listed in the exc_tuple are caught.  If the handler is None, an
           internal basic one is used, which just prints basic info.
 
         WARNING: by putting in your own exception handler into IPython's main
         execution loop, you run a very good chance of nasty crashes.  This
         facility should only be used if you really know what you are doing."""
 
         assert type(exc_tuple)==type(()) , \
                "The custom exceptions must be given AS A TUPLE."
 
         def dummy_handler(self,etype,value,tb):
             print '*** Simple custom exception handler ***'
             print 'Exception type :',etype
             print 'Exception value:',value
             print 'Traceback      :',tb
             print 'Source code    :','\n'.join(self.buffer)
 
         if handler is None: handler = dummy_handler
 
         self.CustomTB = new.instancemethod(handler,self,self.__class__)
         self.custom_exceptions = exc_tuple
 
     def excepthook(self, etype, value, tb):
       """One more defense for GUI apps that call sys.excepthook.
 
       GUI frameworks like wxPython trap exceptions and call
       sys.excepthook themselves.  I guess this is a feature that
       enables them to keep running after exceptions that would
       otherwise kill their mainloop. This is a bother for IPython
       which excepts to catch all of the program exceptions with a try:
       except: statement.
 
       Normally, IPython sets sys.excepthook to a CrashHandler instance, so if
       any app directly invokes sys.excepthook, it will look to the user like
       IPython crashed.  In order to work around this, we can disable the
       CrashHandler and replace it with this excepthook instead, which prints a
       regular traceback using our InteractiveTB.  In this fashion, apps which
       call sys.excepthook will generate a regular-looking exception from
       IPython, and the CrashHandler will only be triggered by real IPython
       crashes.
 
       This hook should be used sparingly, only in places which are not likely
       to be true IPython errors.
       """
       self.showtraceback((etype,value,tb),tb_offset=0)
 
     def showtraceback(self,exc_tuple = None,filename=None,tb_offset=None,
                       exception_only=False):
         """Display the exception that just occurred.
 
         If nothing is known about the exception, this is the method which
         should be used throughout the code for presenting user tracebacks,
         rather than directly invoking the InteractiveTB object.
 
         A specific showsyntaxerror() also exists, but this method can take
         care of calling it if needed, so unless you are explicitly catching a
         SyntaxError exception, don't try to analyze the stack manually and
         simply call this method."""
         
         try:
             if exc_tuple is None:
                 etype, value, tb = sys.exc_info()
             else:
                 etype, value, tb = exc_tuple
 
             if etype is None:
                 if hasattr(sys, 'last_type'):
                     etype, value, tb = sys.last_type, sys.last_value, \
                                        sys.last_traceback
                 else:
                     self.write_err('No traceback available to show.\n')
                     return
     
             if etype is SyntaxError:
                 # Though this won't be called by syntax errors in the input
                 # line, there may be SyntaxError cases whith imported code.
                 self.showsyntaxerror(filename)
             elif etype is UsageError:
                 print "UsageError:", value
             else:
                 # WARNING: these variables are somewhat deprecated and not
                 # necessarily safe to use in a threaded environment, but tools
                 # like pdb depend on their existence, so let's set them.  If we
                 # find problems in the field, we'll need to revisit their use.
                 sys.last_type = etype
                 sys.last_value = value
                 sys.last_traceback = tb
     
                 if etype in self.custom_exceptions:
                     # FIXME: Old custom traceback objects may just return a
                     # string, in that case we just put it into a list
                     stb = self.CustomTB(etype, value, tb, tb_offset)
                     if isinstance(ctb, basestring):
                         stb = [stb]
                 else:
                     if exception_only:
                         stb = ['An exception has occurred, use %tb to see '
-                               'the full traceback.']
+                               'the full traceback.\n']
                         stb.extend(self.InteractiveTB.get_exception_only(etype,
                                                                          value))
                     else:
                         stb = self.InteractiveTB.structured_traceback(etype,
                                                 value, tb, tb_offset=tb_offset)
                         # FIXME: the pdb calling should be done by us, not by
                         # the code computing the traceback.
                         if self.InteractiveTB.call_pdb:
                             # pdb mucks up readline, fix it back
                             self.set_completer()
 
                 # Actually show the traceback
                 self._showtraceback(etype, value, stb)
                 
         except KeyboardInterrupt:
             self.write_err("\nKeyboardInterrupt\n")
 
     def _showtraceback(self, etype, evalue, stb):
         """Actually show a traceback.
 
         Subclasses may override this method to put the traceback on a different
         place, like a side channel.
         """
-        self.write_err('\n'.join(stb))
+        # FIXME: this should use the proper write channels, but our test suite
+        # relies on it coming out of stdout...
+        print >> sys.stdout, self.InteractiveTB.stb2text(stb)
 
     def showsyntaxerror(self, filename=None):
         """Display the syntax error that just occurred.
 
         This doesn't display a stack trace because there isn't one.
 
         If a filename is given, it is stuffed in the exception instead
         of what was there before (because Python's parser always uses
         "<string>" when reading from a string).
         """
         etype, value, last_traceback = sys.exc_info()
 
         # See note about these variables in showtraceback() above
         sys.last_type = etype
         sys.last_value = value
         sys.last_traceback = last_traceback
         
         if filename and etype is SyntaxError:
             # Work hard to stuff the correct filename in the exception
             try:
                 msg, (dummy_filename, lineno, offset, line) = value
             except:
                 # Not the format we expect; leave it alone
                 pass
             else:
                 # Stuff in the right filename
                 try:
                     # Assume SyntaxError is a class exception
                     value = SyntaxError(msg, (filename, lineno, offset, line))
                 except:
                     # If that failed, assume SyntaxError is a string
                     value = msg, (filename, lineno, offset, line)
         stb = self.SyntaxTB.structured_traceback(etype, value, [])
         self._showtraceback(etype, value, stb)
 
     #-------------------------------------------------------------------------
     # Things related to tab completion
     #-------------------------------------------------------------------------
 
-    def complete(self, text):
+    def complete(self, text, line=None, cursor_pos=None):
         """Return a sorted list of all possible completions on text.
 
-        Inputs:
+        Parameters
+        ----------
+
+           text : string
+             A string of text to be completed on.
 
-          - text: a string of text to be completed on.
+           line : string, optional
+             The complete line that text is part of.
+
+           cursor_pos : int, optional
+             The position of the cursor on the input line.
+
+        The optional arguments allow the completion to take more context into
+        account, and are part of the low-level completion API.
 
         This is a wrapper around the completion mechanism, similar to what
         readline does at the command line when the TAB key is hit.  By
         exposing it as a method, it can be used by other non-readline
         environments (such as GUIs) for text completion.
 
         Simple usage example:
 
         In [7]: x = 'hello'
 
         In [8]: x
         Out[8]: 'hello'
 
         In [9]: print x
         hello
 
         In [10]: _ip.complete('x.l')
         Out[10]: ['x.ljust', 'x.lower', 'x.lstrip']
         """
 
         # Inject names into __builtin__ so we can complete on the added names.
         with self.builtin_trap:
-            complete = self.Completer.complete
-            state = 0
-            # use a dict so we get unique keys, since ipyhton's multiple
-            # completers can return duplicates.  When we make 2.4 a requirement,
-            # start using sets instead, which are faster.
-            comps = {}
-            while True:
-                newcomp = complete(text,state,line_buffer=text)
-                if newcomp is None:
-                    break
-                comps[newcomp] = 1
-                state += 1
-            outcomps = comps.keys()
-            outcomps.sort()
-            #print "T:",text,"OC:",outcomps  # dbg
-            #print "vars:",self.user_ns.keys()
-            return outcomps
+            return self.Completer.complete(text,line_buffer=text)
 
     def set_custom_completer(self,completer,pos=0):
         """Adds a new custom completer function.
 
         The position argument (defaults to 0) is the index in the completers
         list where you want the completer to be inserted."""
 
         newcomp = new.instancemethod(completer,self.Completer,
                                      self.Completer.__class__)
         self.Completer.matchers.insert(pos,newcomp)
 
     def set_completer(self):
         """Reset readline's completer to be our own."""
-        self.readline.set_completer(self.Completer.complete)
+        self.readline.set_completer(self.Completer.rlcomplete)
 
     def set_completer_frame(self, frame=None):
         """Set the frame of the completer."""
         if frame:
             self.Completer.namespace = frame.f_locals
             self.Completer.global_namespace = frame.f_globals
         else:
             self.Completer.namespace = self.user_ns
             self.Completer.global_namespace = self.user_global_ns
 
     #-------------------------------------------------------------------------
     # Things related to readline
     #-------------------------------------------------------------------------
 
     def init_readline(self):
         """Command history completion/saving/reloading."""
 
         if self.readline_use:
             import IPython.utils.rlineimpl as readline
 
         self.rl_next_input = None
         self.rl_do_indent = False
 
         if not self.readline_use or not readline.have_readline:
             self.has_readline = False
             self.readline = None
             # Set a number of methods that depend on readline to be no-op
             self.savehist = no_op
             self.reloadhist = no_op
             self.set_completer = no_op
             self.set_custom_completer = no_op
             self.set_completer_frame = no_op
             warn('Readline services not available or not loaded.')
         else:
             self.has_readline = True
             self.readline = readline
             sys.modules['readline'] = readline
             import atexit
             from IPython.core.completer import IPCompleter
             self.Completer = IPCompleter(self,
                                          self.user_ns,
                                          self.user_global_ns,
                                          self.readline_omit__names,
                                          self.alias_manager.alias_table)
             sdisp = self.strdispatchers.get('complete_command', StrDispatch())
             self.strdispatchers['complete_command'] = sdisp
             self.Completer.custom_completers = sdisp
             # Platform-specific configuration
             if os.name == 'nt':
                 self.readline_startup_hook = readline.set_pre_input_hook
             else:
                 self.readline_startup_hook = readline.set_startup_hook
 
             # Load user's initrc file (readline config)
             # Or if libedit is used, load editrc.
             inputrc_name = os.environ.get('INPUTRC')
             if inputrc_name is None:
                 home_dir = get_home_dir()
                 if home_dir is not None:
                     inputrc_name = '.inputrc'
                     if readline.uses_libedit:
                         inputrc_name = '.editrc'
                     inputrc_name = os.path.join(home_dir, inputrc_name)
             if os.path.isfile(inputrc_name):
                 try:
                     readline.read_init_file(inputrc_name)
                 except:
                     warn('Problems reading readline initialization file <%s>'
                          % inputrc_name)
             
             # save this in sys so embedded copies can restore it properly
-            sys.ipcompleter = self.Completer.complete
+            sys.ipcompleter = self.Completer.rlcomplete
             self.set_completer()
 
             # Configure readline according to user's prefs
             # This is only done if GNU readline is being used.  If libedit
             # is being used (as on Leopard) the readline config is
             # not run as the syntax for libedit is different.
             if not readline.uses_libedit:
                 for rlcommand in self.readline_parse_and_bind:
                     #print "loading rl:",rlcommand  # dbg
                     readline.parse_and_bind(rlcommand)
 
             # Remove some chars from the delimiters list.  If we encounter
             # unicode chars, discard them.
             delims = readline.get_completer_delims().encode("ascii", "ignore")
             delims = delims.translate(string._idmap,
                                       self.readline_remove_delims)
             readline.set_completer_delims(delims)
             # otherwise we end up with a monster history after a while:
             readline.set_history_length(1000)
             try:
                 #print '*** Reading readline history'  # dbg
                 readline.read_history_file(self.histfile)
             except IOError:
                 pass  # It doesn't exist yet.
 
             atexit.register(self.atexit_operations)
             del atexit
 
         # Configure auto-indent for all platforms
         self.set_autoindent(self.autoindent)
 
     def set_next_input(self, s):
         """ Sets the 'default' input string for the next command line.
         
         Requires readline.
         
         Example:
         
         [D:\ipython]|1> _ip.set_next_input("Hello Word")
         [D:\ipython]|2> Hello Word_  # cursor is here        
         """
 
         self.rl_next_input = s
 
     # Maybe move this to the terminal subclass?
     def pre_readline(self):
         """readline hook to be used at the start of each line.
 
         Currently it handles auto-indent only."""
 
         if self.rl_do_indent:
             self.readline.insert_text(self._indent_current_str())
         if self.rl_next_input is not None:
             self.readline.insert_text(self.rl_next_input)
             self.rl_next_input = None
 
     def _indent_current_str(self):
         """return the current level of indentation as a string"""
         return self.indent_current_nsp * ' '
 
     #-------------------------------------------------------------------------
     # Things related to magics
     #-------------------------------------------------------------------------
 
     def init_magics(self):
         # FIXME: Move the color initialization to the DisplayHook, which
         # should be split into a prompt manager and displayhook. We probably
         # even need a centralize colors management object.
         self.magic_colors(self.colors)
         # History was moved to a separate module
         from . import history
         history.init_ipython(self)
 
     def magic(self,arg_s):
         """Call a magic function by name.
 
         Input: a string containing the name of the magic function to call and any
         additional arguments to be passed to the magic.
 
         magic('name -opt foo bar') is equivalent to typing at the ipython
         prompt:
 
         In[1]: %name -opt foo bar
 
         To call a magic without arguments, simply use magic('name').
 
         This provides a proper Python function to call IPython's magics in any
         valid Python code you can type at the interpreter, including loops and
         compound statements.
         """
         args = arg_s.split(' ',1)
         magic_name = args[0]
         magic_name = magic_name.lstrip(prefilter.ESC_MAGIC)
 
         try:
             magic_args = args[1]
         except IndexError:
             magic_args = ''
         fn = getattr(self,'magic_'+magic_name,None)
         if fn is None:
             error("Magic function `%s` not found." % magic_name)
         else:
             magic_args = self.var_expand(magic_args,1)
             with nested(self.builtin_trap,):
                 result = fn(magic_args)
                 return result
 
     def define_magic(self, magicname, func):
         """Expose own function as magic function for ipython 
     
         def foo_impl(self,parameter_s=''):
             'My very own magic!. (Use docstrings, IPython reads them).'
             print 'Magic function. Passed parameter is between < >:'
             print '<%s>' % parameter_s
             print 'The self object is:',self
     
         self.define_magic('foo',foo_impl)
         """
         
         import new
         im = new.instancemethod(func,self, self.__class__)
         old = getattr(self, "magic_" + magicname, None)
         setattr(self, "magic_" + magicname, im)
         return old
 
     #-------------------------------------------------------------------------
     # Things related to macros
     #-------------------------------------------------------------------------
 
     def define_macro(self, name, themacro):
         """Define a new macro
 
         Parameters
         ----------
         name : str
             The name of the macro.
         themacro : str or Macro
             The action to do upon invoking the macro.  If a string, a new 
             Macro object is created by passing the string to it.
         """
         
         from IPython.core import macro
 
         if isinstance(themacro, basestring):
             themacro = macro.Macro(themacro)
         if not isinstance(themacro, macro.Macro):
             raise ValueError('A macro must be a string or a Macro instance.')
         self.user_ns[name] = themacro
 
     #-------------------------------------------------------------------------
     # Things related to the running of system commands
     #-------------------------------------------------------------------------
 
     def system(self, cmd):
         """Make a system call, using IPython."""
         return self.hooks.shell_hook(self.var_expand(cmd, depth=2))
 
     #-------------------------------------------------------------------------
     # Things related to aliases
     #-------------------------------------------------------------------------
 
     def init_alias(self):
         self.alias_manager = AliasManager(shell=self, config=self.config)
         self.ns_table['alias'] = self.alias_manager.alias_table,
 
     #-------------------------------------------------------------------------
     # Things related to extensions and plugins
     #-------------------------------------------------------------------------
 
     def init_extension_manager(self):
         self.extension_manager = ExtensionManager(shell=self, config=self.config)
 
     def init_plugin_manager(self):
         self.plugin_manager = PluginManager(config=self.config)
 
     #-------------------------------------------------------------------------
     # Things related to payloads
     #-------------------------------------------------------------------------
 
     def init_payload(self):
         self.payload_manager = PayloadManager(config=self.config)
 
     #-------------------------------------------------------------------------
     # Things related to the prefilter
     #-------------------------------------------------------------------------
 
     def init_prefilter(self):
         self.prefilter_manager = PrefilterManager(shell=self, config=self.config)
         # Ultimately this will be refactored in the new interpreter code, but
         # for now, we should expose the main prefilter method (there's legacy
         # code out there that may rely on this).
         self.prefilter = self.prefilter_manager.prefilter_lines
 
     #-------------------------------------------------------------------------
     # Things related to the running of code
     #-------------------------------------------------------------------------
 
     def ex(self, cmd):
         """Execute a normal python statement in user namespace."""
         with nested(self.builtin_trap,):
             exec cmd in self.user_global_ns, self.user_ns
 
     def ev(self, expr):
         """Evaluate python expression expr in user namespace.
 
         Returns the result of evaluation
         """
         with nested(self.builtin_trap,):
             return eval(expr, self.user_global_ns, self.user_ns)
 
     def safe_execfile(self, fname, *where, **kw):
         """A safe version of the builtin execfile().
 
         This version will never throw an exception, but instead print
         helpful error messages to the screen.  This only works on pure
         Python files with the .py extension.
 
         Parameters
         ----------
         fname : string
             The name of the file to be executed.
         where : tuple
             One or two namespaces, passed to execfile() as (globals,locals).
             If only one is given, it is passed as both.
         exit_ignore : bool (False)
             If True, then silence SystemExit for non-zero status (it is always
             silenced for zero status, as it is so common).
         """
         kw.setdefault('exit_ignore', False)
 
         fname = os.path.abspath(os.path.expanduser(fname))
 
         # Make sure we have a .py file
         if not fname.endswith('.py'):
             warn('File must end with .py to be run using execfile: <%s>' % fname)
 
         # Make sure we can open the file
         try:
             with open(fname) as thefile:
                 pass
         except:
             warn('Could not open file <%s> for safe execution.' % fname)
             return
 
         # Find things also in current directory.  This is needed to mimic the
         # behavior of running a script from the system command line, where
         # Python inserts the script's directory into sys.path
         dname = os.path.dirname(fname)
 
         with prepended_to_syspath(dname):
             try:
                 execfile(fname,*where)
             except SystemExit, status:
                 # If the call was made with 0 or None exit status (sys.exit(0)
                 # or sys.exit() ), don't bother showing a traceback, as both of
                 # these are considered normal by the OS:
                 # > python -c'import sys;sys.exit(0)'; echo $?
                 # 0
                 # > python -c'import sys;sys.exit()'; echo $?
                 # 0
                 # For other exit status, we show the exception unless
                 # explicitly silenced, but only in short form.
                 if status.code not in (0, None) and not kw['exit_ignore']:
                     self.showtraceback(exception_only=True)
             except:
                 self.showtraceback()
 
     def safe_execfile_ipy(self, fname):
         """Like safe_execfile, but for .ipy files with IPython syntax.
 
         Parameters
         ----------
         fname : str
             The name of the file to execute.  The filename must have a
             .ipy extension.
         """
         fname = os.path.abspath(os.path.expanduser(fname))
 
         # Make sure we have a .py file
         if not fname.endswith('.ipy'):
             warn('File must end with .py to be run using execfile: <%s>' % fname)
 
         # Make sure we can open the file
         try:
             with open(fname) as thefile:
                 pass
         except:
             warn('Could not open file <%s> for safe execution.' % fname)
             return
 
         # Find things also in current directory.  This is needed to mimic the
         # behavior of running a script from the system command line, where
         # Python inserts the script's directory into sys.path
         dname = os.path.dirname(fname)
 
         with prepended_to_syspath(dname):
             try:
                 with open(fname) as thefile:
                     script = thefile.read()
                     # self.runlines currently captures all exceptions
                     # raise in user code.  It would be nice if there were
                     # versions of runlines, execfile that did raise, so
                     # we could catch the errors.
                     self.runlines(script, clean=True)
             except:
                 self.showtraceback()
                 warn('Unknown failure executing file: <%s>' % fname)
 
     def runlines(self, lines, clean=False):
         """Run a string of one or more lines of source.
 
         This method is capable of running a string containing multiple source
         lines, as if they had been entered at the IPython prompt.  Since it
         exposes IPython's processing machinery, the given strings can contain
         magic calls (%magic), special shell access (!cmd), etc.
         """
         
         if isinstance(lines, (list, tuple)):
             lines = '\n'.join(lines)
 
         if clean:
             lines = self._cleanup_ipy_script(lines)
 
         # We must start with a clean buffer, in case this is run from an
         # interactive IPython session (via a magic, for example).
         self.resetbuffer()
         lines = lines.splitlines()
         more = 0
 
         with nested(self.builtin_trap, self.display_trap):
             for line in lines:
                 # skip blank lines so we don't mess up the prompt counter, but do
                 # NOT skip even a blank line if we are in a code block (more is
                 # true)
             
                 if line or more:
                     # push to raw history, so hist line numbers stay in sync
                     self.input_hist_raw.append("# " + line + "\n")
                     prefiltered = self.prefilter_manager.prefilter_lines(line,more)
                     more = self.push_line(prefiltered)
                     # IPython's runsource returns None if there was an error
                     # compiling the code.  This allows us to stop processing right
                     # away, so the user gets the error message at the right place.
                     if more is None:
                         break
                 else:
                     self.input_hist_raw.append("\n")
             # final newline in case the input didn't have it, so that the code
             # actually does get executed
             if more:
                 self.push_line('\n')
 
     def runsource(self, source, filename='<input>', symbol='single'):
         """Compile and run some source in the interpreter.
 
         Arguments are as for compile_command().
 
         One several things can happen:
 
         1) The input is incorrect; compile_command() raised an
         exception (SyntaxError or OverflowError).  A syntax traceback
         will be printed by calling the showsyntaxerror() method.
 
         2) The input is incomplete, and more input is required;
         compile_command() returned None.  Nothing happens.
 
         3) The input is complete; compile_command() returned a code
         object.  The code is executed by calling self.runcode() (which
         also handles run-time exceptions, except for SystemExit).
 
         The return value is:
 
           - True in case 2
 
           - False in the other cases, unless an exception is raised, where
           None is returned instead.  This can be used by external callers to
           know whether to continue feeding input or not.
 
         The return value can be used to decide whether to use sys.ps1 or
         sys.ps2 to prompt the next line."""
 
         # if the source code has leading blanks, add 'if 1:\n' to it
         # this allows execution of indented pasted code. It is tempting
         # to add '\n' at the end of source to run commands like ' a=1'
         # directly, but this fails for more complicated scenarios
         source=source.encode(self.stdin_encoding)
         if source[:1] in [' ', '\t']:
             source = 'if 1:\n%s' % source
 
         try:
             code = self.compile(source,filename,symbol)
         except (OverflowError, SyntaxError, ValueError, TypeError, MemoryError):
             # Case 1
             self.showsyntaxerror(filename)
             return None
 
         if code is None:
             # Case 2
             return True
 
         # Case 3
         # We store the code object so that threaded shells and
         # custom exception handlers can access all this info if needed.
         # The source corresponding to this can be obtained from the
         # buffer attribute as '\n'.join(self.buffer).
         self.code_to_run = code
         # now actually execute the code object
         if self.runcode(code) == 0:
             return False
         else:
             return None
 
     def runcode(self,code_obj):
         """Execute a code object.
 
         When an exception occurs, self.showtraceback() is called to display a
         traceback.
 
         Return value: a flag indicating whether the code to be run completed
         successfully:
 
           - 0: successful execution.
           - 1: an error occurred.
         """
 
         # Set our own excepthook in case the user code tries to call it
         # directly, so that the IPython crash handler doesn't get triggered
         old_excepthook,sys.excepthook = sys.excepthook, self.excepthook
 
         # we save the original sys.excepthook in the instance, in case config
         # code (such as magics) needs access to it.
         self.sys_excepthook = old_excepthook
         outflag = 1  # happens in more places, so it's easier as default
         try:
             try:
                 self.hooks.pre_runcode_hook()
                 #rprint('Running code') # dbg
                 exec code_obj in self.user_global_ns, self.user_ns
             finally:
                 # Reset our crash handler in place
                 sys.excepthook = old_excepthook
         except SystemExit:
             self.resetbuffer()
             self.showtraceback(exception_only=True)
             warn("To exit: use any of 'exit', 'quit', %Exit or Ctrl-D.", level=1)
         except self.custom_exceptions:
             etype,value,tb = sys.exc_info()
             self.CustomTB(etype,value,tb)
         except:
             self.showtraceback()
         else:
             outflag = 0
             if softspace(sys.stdout, 0):
                 print
         # Flush out code object which has been run (and source)
         self.code_to_run = None
         return outflag
         
     def push_line(self, line):
         """Push a line to the interpreter.
 
         The line should not have a trailing newline; it may have
         internal newlines.  The line is appended to a buffer and the
         interpreter's runsource() method is called with the
         concatenated contents of the buffer as source.  If this
         indicates that the command was executed or invalid, the buffer
         is reset; otherwise, the command is incomplete, and the buffer
         is left as it was after the line was appended.  The return
         value is 1 if more input is required, 0 if the line was dealt
         with in some way (this is the same as runsource()).
         """
 
         # autoindent management should be done here, and not in the
         # interactive loop, since that one is only seen by keyboard input.  We
         # need this done correctly even for code run via runlines (which uses
         # push).
 
         #print 'push line: <%s>' % line  # dbg
         for subline in line.splitlines():
             self._autoindent_update(subline)
         self.buffer.append(line)
         more = self.runsource('\n'.join(self.buffer), self.filename)
         if not more:
             self.resetbuffer()
         return more
 
     def resetbuffer(self):
         """Reset the input buffer."""
         self.buffer[:] = []
 
     def _is_secondary_block_start(self, s):
         if not s.endswith(':'):
             return False
         if (s.startswith('elif') or 
             s.startswith('else') or 
             s.startswith('except') or
             s.startswith('finally')):
             return True
 
     def _cleanup_ipy_script(self, script):
         """Make a script safe for self.runlines()
 
         Currently, IPython is lines based, with blocks being detected by
         empty lines.  This is a problem for block based scripts that may
         not have empty lines after blocks.  This script adds those empty
         lines to make scripts safe for running in the current line based
         IPython.
         """
         res = []
         lines = script.splitlines()
         level = 0
 
         for l in lines:
             lstripped = l.lstrip()
             stripped = l.strip()                
             if not stripped:
                 continue
             newlevel = len(l) - len(lstripped)                    
             if level > 0 and newlevel == 0 and \
                    not self._is_secondary_block_start(stripped): 
                 # add empty line
                 res.append('')
             res.append(l)
             level = newlevel
 
         return '\n'.join(res) + '\n'
 
     def _autoindent_update(self,line):
         """Keep track of the indent level."""
 
         #debugx('line')
         #debugx('self.indent_current_nsp')
         if self.autoindent:
             if line:
                 inisp = num_ini_spaces(line)
                 if inisp < self.indent_current_nsp:
                     self.indent_current_nsp = inisp
 
                 if line[-1] == ':':
                     self.indent_current_nsp += 4
                 elif dedent_re.match(line):
                     self.indent_current_nsp -= 4
             else:
                 self.indent_current_nsp = 0
 
     #-------------------------------------------------------------------------
     # Things related to GUI support and pylab
     #-------------------------------------------------------------------------
 
     def enable_pylab(self, gui=None):
         raise NotImplementedError('Implement enable_pylab in a subclass')
 
     #-------------------------------------------------------------------------
     # Utilities
     #-------------------------------------------------------------------------
 
     def getoutput(self, cmd):
         return getoutput(self.var_expand(cmd,depth=2),
                          header=self.system_header,
                          verbose=self.system_verbose)
 
     def getoutputerror(self, cmd):
         return getoutputerror(self.var_expand(cmd,depth=2),
                               header=self.system_header,
                               verbose=self.system_verbose)
 
     def var_expand(self,cmd,depth=0):
         """Expand python variables in a string.
 
         The depth argument indicates how many frames above the caller should
         be walked to look for the local namespace where to expand variables.
 
         The global namespace for expansion is always the user's interactive
         namespace.
         """
 
         return str(ItplNS(cmd,
                           self.user_ns,  # globals
                           # Skip our own frame in searching for locals:
                           sys._getframe(depth+1).f_locals # locals
                           ))
 
     def mktempfile(self,data=None):
         """Make a new tempfile and return its filename.
 
         This makes a call to tempfile.mktemp, but it registers the created
         filename internally so ipython cleans it up at exit time.
 
         Optional inputs:
 
           - data(None): if data is given, it gets written out to the temp file
           immediately, and the file is closed again."""
 
         filename = tempfile.mktemp('.py','ipython_edit_')
         self.tempfiles.append(filename)
         
         if data:
             tmp_file = open(filename,'w')
             tmp_file.write(data)
             tmp_file.close()
         return filename
 
     # TODO:  This should be removed when Term is refactored.
     def write(self,data):
         """Write a string to the default output"""
         io.Term.cout.write(data)
 
     # TODO:  This should be removed when Term is refactored.
     def write_err(self,data):
         """Write a string to the default error output"""
         io.Term.cerr.write(data)
 
     def ask_yes_no(self,prompt,default=True):
         if self.quiet:
             return True
         return ask_yes_no(prompt,default)
 
     #-------------------------------------------------------------------------
     # Things related to IPython exiting
     #-------------------------------------------------------------------------
 
     def atexit_operations(self):
         """This will be executed at the time of exit.
 
         Saving of persistent data should be performed here.
         """
         self.savehist()
 
         # Cleanup all tempfiles left around
         for tfile in self.tempfiles:
             try:
                 os.unlink(tfile)
             except OSError:
                 pass
 
         # Clear all user namespaces to release all references cleanly.
         self.reset()
 
         # Run user hooks
         self.hooks.shutdown_hook()
 
     def cleanup(self):
         self.restore_sys_module_state()
 
 
 class InteractiveShellABC(object):
     """An abstract base class for InteractiveShell."""
     __metaclass__ = abc.ABCMeta
 
 InteractiveShellABC.register(InteractiveShell)

IPython/core/ultratb.py

@@ -1,1202 +1,1201 b''
 # -*- coding: utf-8 -*-
 """
 ultratb.py -- Spice up your tracebacks!
 
 * ColorTB
 I've always found it a bit hard to visually parse tracebacks in Python.  The
 ColorTB class is a solution to that problem.  It colors the different parts of a
 traceback in a manner similar to what you would expect from a syntax-highlighting
 text editor.
 
 Installation instructions for ColorTB:
     import sys,ultratb
     sys.excepthook = ultratb.ColorTB()
 
 * VerboseTB  
 I've also included a port of Ka-Ping Yee's "cgitb.py" that produces all kinds
 of useful info when a traceback occurs.  Ping originally had it spit out HTML
 and intended it for CGI programmers, but why should they have all the fun?  I
 altered it to spit out colored text to the terminal.  It's a bit overwhelming,
 but kind of neat, and maybe useful for long-running programs that you believe
 are bug-free.  If a crash *does* occur in that type of program you want details.
 Give it a shot--you'll love it or you'll hate it.
 
 Note:
 
   The Verbose mode prints the variables currently visible where the exception
   happened (shortening their strings if too long). This can potentially be
   very slow, if you happen to have a huge data structure whose string
   representation is complex to compute. Your computer may appear to freeze for
   a while with cpu usage at 100%. If this occurs, you can cancel the traceback
   with Ctrl-C (maybe hitting it more than once).
 
   If you encounter this kind of situation often, you may want to use the
   Verbose_novars mode instead of the regular Verbose, which avoids formatting
   variables (but otherwise includes the information and context given by
   Verbose).
   
 
 Installation instructions for ColorTB:
     import sys,ultratb
     sys.excepthook = ultratb.VerboseTB()
 
 Note:  Much of the code in this module was lifted verbatim from the standard
 library module 'traceback.py' and Ka-Ping Yee's 'cgitb.py'.
 
 * Color schemes
 The colors are defined in the class TBTools through the use of the
 ColorSchemeTable class. Currently the following exist:
 
   - NoColor: allows all of this module to be used in any terminal (the color
   escapes are just dummy blank strings).
 
   - Linux: is meant to look good in a terminal like the Linux console (black
   or very dark background).
 
   - LightBG: similar to Linux but swaps dark/light colors to be more readable
   in light background terminals.
 
 You can implement other color schemes easily, the syntax is fairly
 self-explanatory. Please send back new schemes you develop to the author for
 possible inclusion in future releases.
 """
 
 #*****************************************************************************
 #       Copyright (C) 2001 Nathaniel Gray <[e-mail]>
 #       Copyright (C) 2001-2004 Fernando Perez <[e-mail]>
 #
 #  Distributed under the terms of the BSD License.  The full license is in
 #  the file COPYING, distributed as part of this software.
 #*****************************************************************************
 
 from __future__ import with_statement
 
 import inspect
 import keyword
 import linecache
 import os
 import pydoc
 import re
 import string
 import sys
 import time
 import tokenize
 import traceback
 import types
 
 # For purposes of monkeypatching inspect to fix a bug in it.
 from inspect import getsourcefile, getfile, getmodule,\
      ismodule, isclass, ismethod, isfunction, istraceback, isframe, iscode
 
 # IPython's own modules
 # Modified pdb which doesn't damage IPython's readline handling
 from IPython.core import debugger, ipapi
 from IPython.core.display_trap import DisplayTrap
 from IPython.core.excolors import exception_colors
 from IPython.utils import PyColorize
 from IPython.utils import io
 from IPython.utils.data import uniq_stable
 from IPython.utils.warn import info, error
 
 # Globals
 # amount of space to put line numbers before verbose tracebacks
 INDENT_SIZE = 8
 
 # Default color scheme.  This is used, for example, by the traceback
 # formatter.  When running in an actual IPython instance, the user's rc.colors
 # value is used, but havinga module global makes this functionality available
 # to users of ultratb who are NOT running inside ipython.
 DEFAULT_SCHEME = 'NoColor'
 
 #---------------------------------------------------------------------------
 # Code begins
 
 # Utility functions
 def inspect_error():
     """Print a message about internal inspect errors.
 
     These are unfortunately quite common."""
     
     error('Internal Python error in the inspect module.\n'
           'Below is the traceback from this internal error.\n')
 
 
 def findsource(object):
     """Return the entire source file and starting line number for an object.
 
     The argument may be a module, class, method, function, traceback, frame,
     or code object.  The source code is returned as a list of all the lines
     in the file and the line number indexes a line in that list.  An IOError
     is raised if the source code cannot be retrieved.
 
     FIXED version with which we monkeypatch the stdlib to work around a bug."""
 
     file = getsourcefile(object) or getfile(object)
     # If the object is a frame, then trying to get the globals dict from its
     # module won't work. Instead, the frame object itself has the globals
     # dictionary.
     globals_dict = None
     if inspect.isframe(object):
         # XXX: can this ever be false?
         globals_dict = object.f_globals
     else:
         module = getmodule(object, file)
         if module:
             globals_dict = module.__dict__
     lines = linecache.getlines(file, globals_dict)
     if not lines:
         raise IOError('could not get source code')
 
     if ismodule(object):
         return lines, 0
 
     if isclass(object):
         name = object.__name__
         pat = re.compile(r'^(\s*)class\s*' + name + r'\b')
         # make some effort to find the best matching class definition:
         # use the one with the least indentation, which is the one
         # that's most probably not inside a function definition.
         candidates = []
         for i in range(len(lines)):
             match = pat.match(lines[i])
             if match:
                 # if it's at toplevel, it's already the best one
                 if lines[i][0] == 'c':
                     return lines, i
                 # else add whitespace to candidate list
                 candidates.append((match.group(1), i))
         if candidates:
             # this will sort by whitespace, and by line number,
             # less whitespace first
             candidates.sort()
             return lines, candidates[0][1]
         else:
             raise IOError('could not find class definition')
 
     if ismethod(object):
         object = object.im_func
     if isfunction(object):
         object = object.func_code
     if istraceback(object):
         object = object.tb_frame
     if isframe(object):
         object = object.f_code
     if iscode(object):
         if not hasattr(object, 'co_firstlineno'):
             raise IOError('could not find function definition')
         pat = re.compile(r'^(\s*def\s)|(.*(?<!\w)lambda(:|\s))|^(\s*@)')
         pmatch = pat.match
         # fperez - fix: sometimes, co_firstlineno can give a number larger than
         # the length of lines, which causes an error.  Safeguard against that.
         lnum = min(object.co_firstlineno,len(lines))-1
         while lnum > 0:
             if pmatch(lines[lnum]): break
             lnum -= 1
  
         return lines, lnum
     raise IOError('could not find code object')
 
 # Monkeypatch inspect to apply our bugfix.  This code only works with py25
 if sys.version_info[:2] >= (2,5):
     inspect.findsource = findsource
 
 def fix_frame_records_filenames(records):
     """Try to fix the filenames in each record from inspect.getinnerframes().
 
     Particularly, modules loaded from within zip files have useless filenames
     attached to their code object, and inspect.getinnerframes() just uses it.
     """
     fixed_records = []
     for frame, filename, line_no, func_name, lines, index in records:
         # Look inside the frame's globals dictionary for __file__, which should
         # be better.
         better_fn = frame.f_globals.get('__file__', None)
         if isinstance(better_fn, str):
             # Check the type just in case someone did something weird with
             # __file__. It might also be None if the error occurred during
             # import.
             filename = better_fn
         fixed_records.append((frame, filename, line_no, func_name, lines, index))
     return fixed_records
 
 
 def _fixed_getinnerframes(etb, context=1,tb_offset=0):
     import linecache
     LNUM_POS, LINES_POS, INDEX_POS =  2, 4, 5
 
     records  = fix_frame_records_filenames(inspect.getinnerframes(etb, context))
 
     # If the error is at the console, don't build any context, since it would
     # otherwise produce 5 blank lines printed out (there is no file at the
     # console)
     rec_check = records[tb_offset:]
     try:
         rname = rec_check[0][1]
         if rname == '<ipython console>' or rname.endswith('<string>'):
             return rec_check
     except IndexError:
         pass
 
     aux = traceback.extract_tb(etb)
     assert len(records) == len(aux)
     for i, (file, lnum, _, _) in zip(range(len(records)), aux):
         maybeStart = lnum-1 - context//2
         start =  max(maybeStart, 0)
         end   = start + context
         lines = linecache.getlines(file)[start:end]
         # pad with empty lines if necessary
         if maybeStart < 0:
             lines = (['\n'] * -maybeStart) + lines
         if len(lines) < context:
             lines += ['\n'] * (context - len(lines))
         buf = list(records[i])
         buf[LNUM_POS] = lnum
         buf[INDEX_POS] = lnum - 1 - start
         buf[LINES_POS] = lines
         records[i] = tuple(buf)
     return records[tb_offset:]
 
 # Helper function -- largely belongs to VerboseTB, but we need the same
 # functionality to produce a pseudo verbose TB for SyntaxErrors, so that they
 # can be recognized properly by ipython.el's py-traceback-line-re
 # (SyntaxErrors have to be treated specially because they have no traceback)
 
 _parser = PyColorize.Parser()
     
 def _format_traceback_lines(lnum, index, lines, Colors, lvals=None,scheme=None):
     numbers_width = INDENT_SIZE - 1
     res = []
     i = lnum - index
 
     # This lets us get fully syntax-highlighted tracebacks.
     if scheme is None:
         ipinst = ipapi.get()
         if ipinst is not None:
             scheme = ipinst.colors
         else:
             scheme = DEFAULT_SCHEME
 
     _line_format = _parser.format2
 
     for line in lines:
         new_line, err = _line_format(line,'str',scheme)
         if not err: line = new_line
         
         if i == lnum:
             # This is the line with the error
             pad = numbers_width - len(str(i))
             if pad >= 3:
                 marker = '-'*(pad-3) + '-> '
             elif pad == 2:
                 marker = '> '
             elif pad == 1:
                 marker = '>'
             else:
                 marker = ''
             num = marker + str(i)
             line = '%s%s%s %s%s' %(Colors.linenoEm, num, 
                                    Colors.line, line, Colors.Normal)
         else:
             num = '%*s' % (numbers_width,i)
             line = '%s%s%s %s' %(Colors.lineno, num, 
                                  Colors.Normal, line)
 
         res.append(line)
         if lvals and i == lnum:
             res.append(lvals + '\n')
         i = i + 1
     return res
 
 
 #---------------------------------------------------------------------------
 # Module classes
 class TBTools(object):
     """Basic tools used by all traceback printer classes."""
 
     # This attribute us used in globalipapp.py to have stdout used for
     # writting exceptions. This is needed so nose can trap them. This attribute
     # should be None (the default, which will use IPython.utils.io.Term) or
     # the string 'stdout' which will cause the override to sys.stdout.
     out_stream = None
 
     # Number of frames to skip when reporting tracebacks
     tb_offset = 0
 
     def __init__(self,color_scheme = 'NoColor',call_pdb=False):
         # Whether to call the interactive pdb debugger after printing
         # tracebacks or not
         self.call_pdb = call_pdb
 
         # Create color table
         self.color_scheme_table = exception_colors()
 
         self.set_colors(color_scheme)
         self.old_scheme = color_scheme  # save initial value for toggles
 
         if call_pdb:
             self.pdb = debugger.Pdb(self.color_scheme_table.active_scheme_name)
         else:
             self.pdb = None
 
     def set_colors(self,*args,**kw):
         """Shorthand access to the color table scheme selector method."""
 
         # Set own color table
         self.color_scheme_table.set_active_scheme(*args,**kw)
         # for convenience, set Colors to the active scheme
         self.Colors = self.color_scheme_table.active_colors
         # Also set colors of debugger
         if hasattr(self,'pdb') and self.pdb is not None:
             self.pdb.set_colors(*args,**kw)
 
     def color_toggle(self):
         """Toggle between the currently active color scheme and NoColor."""
         
         if self.color_scheme_table.active_scheme_name == 'NoColor':
             self.color_scheme_table.set_active_scheme(self.old_scheme)
             self.Colors = self.color_scheme_table.active_colors
         else:
             self.old_scheme = self.color_scheme_table.active_scheme_name
             self.color_scheme_table.set_active_scheme('NoColor')
             self.Colors = self.color_scheme_table.active_colors
 
+    def stb2text(self, stb):
+        """Convert a structured traceback (a list) to a string."""
+        return '\n'.join(stb)
+
     def text(self, etype, value, tb, tb_offset=None, context=5):
         """Return formatted traceback.
 
         Subclasses may override this if they add extra arguments.
         """
         tb_list = self.structured_traceback(etype, value, tb,
                                             tb_offset, context)
-        return '\n'.join(tb_list)
+        return self.stb2text(tb_list)
 
     def structured_traceback(self, etype, evalue, tb, tb_offset=None,
                              context=5, mode=None):
         """Return a list of traceback frames.
 
         Must be implemented by each class.
         """
         raise NotImplementedError()
 
 
 #---------------------------------------------------------------------------
 class ListTB(TBTools):
     """Print traceback information from a traceback list, with optional color.
         
     Calling: requires 3 arguments:
       (etype, evalue, elist)
     as would be obtained by:
       etype, evalue, tb = sys.exc_info()
       if tb:
         elist = traceback.extract_tb(tb)
       else:
         elist = None
 
     It can thus be used by programs which need to process the traceback before
     printing (such as console replacements based on the code module from the
     standard library).
 
     Because they are meant to be called without a full traceback (only a
     list), instances of this class can't call the interactive pdb debugger."""
 
     def __init__(self,color_scheme = 'NoColor'):
         TBTools.__init__(self,color_scheme = color_scheme,call_pdb=0)
         
     def __call__(self, etype, value, elist):
         io.Term.cout.flush()
         io.Term.cerr.write(self.text(etype, value, elist))
         io.Term.cerr.write('\n')
 
     def structured_traceback(self, etype, value, elist, tb_offset=None,
                              context=5):
         """Return a color formatted string with the traceback info.
 
         Parameters
         ----------
         etype : exception type
           Type of the exception raised.
 
         value : object
           Data stored in the exception
 
         elist : list
           List of frames, see class docstring for details.
 
         tb_offset : int, optional
           Number of frames in the traceback to skip.  If not given, the
           instance value is used (set in constructor).
           
         context : int, optional
           Number of lines of context information to print.
 
         Returns
         -------
         String with formatted exception.
         """
         tb_offset = self.tb_offset if tb_offset is None else tb_offset
         Colors = self.Colors
         out_list = []
         if elist:
 
             if tb_offset and len(elist) > tb_offset:
                 elist = elist[tb_offset:]
             
             out_list.append('Traceback %s(most recent call last)%s:' %
                                 (Colors.normalEm, Colors.Normal) + '\n')
             out_list.extend(self._format_list(elist))
         # The exception info should be a single entry in the list.
         lines = ''.join(self._format_exception_only(etype, value))
         out_list.append(lines)
 
         # Note: this code originally read:
         
         ## for line in lines[:-1]:
         ##     out_list.append(" "+line)
         ## out_list.append(lines[-1])
 
         # This means it was indenting everything but the last line by a little
         # bit.  I've disabled this for now, but if we see ugliness somewhre we
         # can restore it.
         
         return out_list
 
     def _format_list(self, extracted_list):
         """Format a list of traceback entry tuples for printing.
 
         Given a list of tuples as returned by extract_tb() or
         extract_stack(), return a list of strings ready for printing.
         Each string in the resulting list corresponds to the item with the
         same index in the argument list.  Each string ends in a newline;
         the strings may contain internal newlines as well, for those items
         whose source text line is not None.
         
         Lifted almost verbatim from traceback.py
         """
 
         Colors = self.Colors
         list = []
         for filename, lineno, name, line in extracted_list[:-1]:
             item = '  File %s"%s"%s, line %s%d%s, in %s%s%s\n' % \
                     (Colors.filename, filename, Colors.Normal, 
                      Colors.lineno, lineno, Colors.Normal,
                      Colors.name, name, Colors.Normal)
             if line:
                 item = item + '    %s\n' % line.strip()
             list.append(item)
         # Emphasize the last entry
         filename, lineno, name, line = extracted_list[-1]
         item = '%s  File %s"%s"%s, line %s%d%s, in %s%s%s%s\n' % \
                 (Colors.normalEm,
                  Colors.filenameEm, filename, Colors.normalEm,
                  Colors.linenoEm, lineno, Colors.normalEm,
                  Colors.nameEm, name, Colors.normalEm,
                  Colors.Normal)
         if line:
             item = item + '%s    %s%s\n' % (Colors.line, line.strip(),
                                             Colors.Normal)
         list.append(item)
         #from pprint import pformat; print 'LISTTB', pformat(list) # dbg
         return list
         
     def _format_exception_only(self, etype, value):
         """Format the exception part of a traceback.
 
         The arguments are the exception type and value such as given by
         sys.exc_info()[:2]. The return value is a list of strings, each ending
         in a newline.  Normally, the list contains a single string; however,
         for SyntaxError exceptions, it contains several lines that (when
         printed) display detailed information about where the syntax error
         occurred.  The message indicating which exception occurred is the
         always last string in the list.
         
         Also lifted nearly verbatim from traceback.py
         """
 
         have_filedata = False
         Colors = self.Colors
         list = []
         try:
             stype = Colors.excName + etype.__name__ + Colors.Normal
         except AttributeError:
             stype = etype  # String exceptions don't get special coloring
         if value is None:
             list.append( str(stype) + '\n')
         else:
             if etype is SyntaxError:
                 try:
                     msg, (filename, lineno, offset, line) = value
                 except:
                     have_filedata = False
                 else:
                     have_filedata = True
                     #print 'filename is',filename  # dbg
                     if not filename: filename = "<string>"
                     list.append('%s  File %s"%s"%s, line %s%d%s\n' % \
                             (Colors.normalEm,
                              Colors.filenameEm, filename, Colors.normalEm,
                              Colors.linenoEm, lineno, Colors.Normal  ))
                     if line is not None:
                         i = 0
                         while i < len(line) and line[i].isspace():
                             i = i+1
                         list.append('%s    %s%s\n' % (Colors.line,
                                                       line.strip(), 
                                                       Colors.Normal))
                         if offset is not None:
                             s = '    '
                             for c in line[i:offset-1]:
                                 if c.isspace():
                                     s = s + c
                                 else:
                                     s = s + ' '
                             list.append('%s%s^%s\n' % (Colors.caret, s,
                                                        Colors.Normal) )
                         value = msg
             s = self._some_str(value)
             if s:
                 list.append('%s%s:%s %s\n' % (str(stype), Colors.excName,
                                               Colors.Normal, s))
             else:
                 list.append('%s\n' % str(stype))
 
         # sync with user hooks
         if have_filedata:
             ipinst = ipapi.get()
             if ipinst is not None:
                 ipinst.hooks.synchronize_with_editor(filename, lineno, 0)
 
         return list
 
     def get_exception_only(self, etype, value):
         """Only print the exception type and message, without a traceback.
         
         Parameters
         ----------
         etype : exception type
         value : exception value
         """
         return ListTB.structured_traceback(self, etype, value, [])
 
 
     def show_exception_only(self, etype, value):
         """Only print the exception type and message, without a traceback.
         
         Parameters
         ----------
         etype : exception type
         value : exception value
         """
         # This method needs to use __call__ from *this* class, not the one from
         # a subclass whose signature or behavior may be different
         if self.out_stream == 'stdout':
             ostream = sys.stdout
         else:
             ostream = io.Term.cerr
         ostream.flush()
         ostream.write('\n'.join(self.get_exception_only(etype, evalue)))
         ostream.flush()
 
     def _some_str(self, value):
         # Lifted from traceback.py
         try:
             return str(value)
         except:
             return '<unprintable %s object>' % type(value).__name__
 
 #----------------------------------------------------------------------------
 class VerboseTB(TBTools):
     """A port of Ka-Ping Yee's cgitb.py module that outputs color text instead
     of HTML.  Requires inspect and pydoc.  Crazy, man.
 
     Modified version which optionally strips the topmost entries from the
     traceback, to be used with alternate interpreters (because their own code
     would appear in the traceback)."""
 
     def __init__(self,color_scheme = 'Linux',tb_offset=0,long_header=0,
                  call_pdb = 0, include_vars=1):
         """Specify traceback offset, headers and color scheme.
 
         Define how many frames to drop from the tracebacks. Calling it with
         tb_offset=1 allows use of this handler in interpreters which will have
         their own code at the top of the traceback (VerboseTB will first
         remove that frame before printing the traceback info)."""
         TBTools.__init__(self,color_scheme=color_scheme,call_pdb=call_pdb)
         self.tb_offset = tb_offset
         self.long_header = long_header
         self.include_vars = include_vars
 
     def structured_traceback(self, etype, evalue, etb, tb_offset=None,
                              context=5):
         """Return a nice text document describing the traceback."""
 
         tb_offset = self.tb_offset if tb_offset is None else tb_offset
 
         # some locals
         try:
             etype = etype.__name__
         except AttributeError:
             pass
         Colors        = self.Colors   # just a shorthand + quicker name lookup
         ColorsNormal  = Colors.Normal  # used a lot
         col_scheme    = self.color_scheme_table.active_scheme_name
         indent        = ' '*INDENT_SIZE
         em_normal     = '%s\n%s%s' % (Colors.valEm, indent,ColorsNormal)
         undefined     = '%sundefined%s' % (Colors.em, ColorsNormal)
         exc = '%s%s%s' % (Colors.excName,etype,ColorsNormal)
 
         # some internal-use functions
         def text_repr(value):
             """Hopefully pretty robust repr equivalent."""
             # this is pretty horrible but should always return *something*
             try:
                 return pydoc.text.repr(value)
             except KeyboardInterrupt:
                 raise
             except:
                 try:
                     return repr(value)
                 except KeyboardInterrupt:
                     raise
                 except:
                     try:
                         # all still in an except block so we catch
                         # getattr raising
                         name = getattr(value, '__name__', None)
                         if name:
                             # ick, recursion
                             return text_repr(name)
                         klass = getattr(value, '__class__', None)
                         if klass:
                             return '%s instance' % text_repr(klass)
                     except KeyboardInterrupt:
                         raise
                     except:
                         return 'UNRECOVERABLE REPR FAILURE'
         def eqrepr(value, repr=text_repr): return '=%s' % repr(value)
         def nullrepr(value, repr=text_repr): return ''
 
         # meat of the code begins
         try:
             etype = etype.__name__
         except AttributeError:
             pass
 
         if self.long_header:
             # Header with the exception type, python version, and date
             pyver = 'Python ' + string.split(sys.version)[0] + ': ' + sys.executable
             date = time.ctime(time.time())
             
             head = '%s%s%s\n%s%s%s\n%s' % (Colors.topline, '-'*75, ColorsNormal,
                                            exc, ' '*(75-len(str(etype))-len(pyver)),
                                            pyver, string.rjust(date, 75) )
             head += "\nA problem occured executing Python code.  Here is the sequence of function"\
                     "\ncalls leading up to the error, with the most recent (innermost) call last."
         else:
             # Simplified header
             head = '%s%s%s\n%s%s' % (Colors.topline, '-'*75, ColorsNormal,exc,
                                      string.rjust('Traceback (most recent call last)',
                                                   75 - len(str(etype)) ) )
         frames = []
         # Flush cache before calling inspect.  This helps alleviate some of the
         # problems with python 2.3's inspect.py.
         linecache.checkcache()
         # Drop topmost frames if requested
         try:
             # Try the default getinnerframes and Alex's: Alex's fixes some
             # problems, but it generates empty tracebacks for console errors
             # (5 blanks lines) where none should be returned.
             #records = inspect.getinnerframes(etb, context)[tb_offset:]
             #print 'python records:', records # dbg
             records = _fixed_getinnerframes(etb, context, tb_offset)
             #print 'alex   records:', records # dbg
         except:
 
             # FIXME: I've been getting many crash reports from python 2.3
             # users, traceable to inspect.py.  If I can find a small test-case
             # to reproduce this, I should either write a better workaround or
             # file a bug report against inspect (if that's the real problem).
             # So far, I haven't been able to find an isolated example to
             # reproduce the problem.
             inspect_error()
             traceback.print_exc(file=io.Term.cerr)
             info('\nUnfortunately, your original traceback can not be constructed.\n')
             return ''
 
         # build some color string templates outside these nested loops
         tpl_link       = '%s%%s%s' % (Colors.filenameEm,ColorsNormal)
         tpl_call       = 'in %s%%s%s%%s%s' % (Colors.vName, Colors.valEm,
                                               ColorsNormal)
         tpl_call_fail  = 'in %s%%s%s(***failed resolving arguments***)%s' % \
                          (Colors.vName, Colors.valEm, ColorsNormal)
         tpl_local_var  = '%s%%s%s' % (Colors.vName, ColorsNormal)
         tpl_global_var = '%sglobal%s %s%%s%s' % (Colors.em, ColorsNormal,
                                                  Colors.vName, ColorsNormal)
         tpl_name_val   = '%%s %s= %%s%s' % (Colors.valEm, ColorsNormal)
         tpl_line       = '%s%%s%s %%s' % (Colors.lineno, ColorsNormal)
         tpl_line_em    = '%s%%s%s %%s%s' % (Colors.linenoEm,Colors.line,
                                             ColorsNormal)
 
         # now, loop over all records printing context and info
         abspath = os.path.abspath
         for frame, file, lnum, func, lines, index in records:
             #print '*** record:',file,lnum,func,lines,index  # dbg
             try:
                 file = file and abspath(file) or '?'
             except OSError:
                 # if file is '<console>' or something not in the filesystem,
                 # the abspath call will throw an OSError.  Just ignore it and
                 # keep the original file string.
                 pass
             link = tpl_link % file
             try:
                 args, varargs, varkw, locals = inspect.getargvalues(frame)
             except:
                 # This can happen due to a bug in python2.3.  We should be
                 # able to remove this try/except when 2.4 becomes a
                 # requirement.  Bug details at http://python.org/sf/1005466
                 inspect_error()
                 traceback.print_exc(file=io.Term.cerr)
                 info("\nIPython's exception reporting continues...\n")
                 
             if func == '?':
                 call = ''
             else:
                 # Decide whether to include variable details or not
                 var_repr = self.include_vars and eqrepr or nullrepr
                 try:
                     call = tpl_call % (func,inspect.formatargvalues(args,
                                                 varargs, varkw,
                                                 locals,formatvalue=var_repr))
                 except KeyError:
                     # Very odd crash from inspect.formatargvalues().  The
                     # scenario under which it appeared was a call to
                     # view(array,scale) in NumTut.view.view(), where scale had
                     # been defined as a scalar (it should be a tuple). Somehow
                     # inspect messes up resolving the argument list of view()
                     # and barfs out. At some point I should dig into this one
                     # and file a bug report about it.
                     inspect_error()
                     traceback.print_exc(file=io.Term.cerr)
                     info("\nIPython's exception reporting continues...\n")
                     call = tpl_call_fail % func
 
             # Initialize a list of names on the current line, which the
             # tokenizer below will populate.
             names = []
 
             def tokeneater(token_type, token, start, end, line):
                 """Stateful tokeneater which builds dotted names.
 
                 The list of names it appends to (from the enclosing scope) can
                 contain repeated composite names.  This is unavoidable, since
                 there is no way to disambguate partial dotted structures until
                 the full list is known.  The caller is responsible for pruning
                 the final list of duplicates before using it."""
                 
                 # build composite names
                 if token == '.':
                     try:
                         names[-1] += '.'
                         # store state so the next token is added for x.y.z names
                         tokeneater.name_cont = True
                         return
                     except IndexError:
                         pass
                 if token_type == tokenize.NAME and token not in keyword.kwlist:
                     if tokeneater.name_cont:
                         # Dotted names
                         names[-1] += token
                         tokeneater.name_cont = False
                     else:
                         # Regular new names.  We append everything, the caller
                         # will be responsible for pruning the list later.  It's
                         # very tricky to try to prune as we go, b/c composite
                         # names can fool us.  The pruning at the end is easy
                         # to do (or the caller can print a list with repeated
                         # names if so desired.
                         names.append(token)
                 elif token_type == tokenize.NEWLINE:
                     raise IndexError
             # we need to store a bit of state in the tokenizer to build
             # dotted names
             tokeneater.name_cont = False
 
             def linereader(file=file, lnum=[lnum], getline=linecache.getline):
                 line = getline(file, lnum[0])
                 lnum[0] += 1
                 return line
 
             # Build the list of names on this line of code where the exception
             # occurred.
             try:
                 # This builds the names list in-place by capturing it from the
                 # enclosing scope.
                 tokenize.tokenize(linereader, tokeneater)
             except IndexError:
                 # signals exit of tokenizer
                 pass
             except tokenize.TokenError,msg:
                 _m = ("An unexpected error occurred while tokenizing input\n"
                       "The following traceback may be corrupted or invalid\n"
                       "The error message is: %s\n" % msg)
                 error(_m)
             
             # prune names list of duplicates, but keep the right order
             unique_names = uniq_stable(names)
 
             # Start loop over vars
             lvals = []
             if self.include_vars:
                 for name_full in unique_names:
                     name_base = name_full.split('.',1)[0]
                     if name_base in frame.f_code.co_varnames:
                         if locals.has_key(name_base):
                             try:
                                 value = repr(eval(name_full,locals))
                             except:
                                 value = undefined
                         else:
                             value = undefined
                         name = tpl_local_var % name_full
                     else:
                         if frame.f_globals.has_key(name_base):
                             try:
                                 value = repr(eval(name_full,frame.f_globals))
                             except:
                                 value = undefined
                         else:
                             value = undefined
                         name = tpl_global_var % name_full
                     lvals.append(tpl_name_val % (name,value))
             if lvals:
                 lvals = '%s%s' % (indent,em_normal.join(lvals))
             else:
                 lvals = ''
 
             level = '%s %s\n' % (link,call)
 
             if index is None:
                 frames.append(level)
             else:
                 frames.append('%s%s' % (level,''.join(
                     _format_traceback_lines(lnum,index,lines,Colors,lvals,
                                             col_scheme))))
 
         # Get (safely) a string form of the exception info
         try:
             etype_str,evalue_str = map(str,(etype,evalue))
         except:
             # User exception is improperly defined.
             etype,evalue = str,sys.exc_info()[:2]
             etype_str,evalue_str = map(str,(etype,evalue))
         # ... and format it
         exception = ['%s%s%s: %s' % (Colors.excName, etype_str,
                                      ColorsNormal, evalue_str)]
         if type(evalue) is types.InstanceType:
             try:
                 names = [w for w in dir(evalue) if isinstance(w, basestring)]
             except:
                 # Every now and then, an object with funny inernals blows up
                 # when dir() is called on it.  We do the best we can to report
                 # the problem and continue
                 _m = '%sException reporting error (object with broken dir())%s:'
                 exception.append(_m % (Colors.excName,ColorsNormal))
                 etype_str,evalue_str = map(str,sys.exc_info()[:2])
                 exception.append('%s%s%s: %s' % (Colors.excName,etype_str,
                                      ColorsNormal, evalue_str))
                 names = []
             for name in names:
                 value = text_repr(getattr(evalue, name))
                 exception.append('\n%s%s = %s' % (indent, name, value))
 
         # vds: >>
         if records:
              filepath, lnum = records[-1][1:3]
              #print "file:", str(file), "linenb", str(lnum) # dbg
              filepath = os.path.abspath(filepath)
              ipinst = ipapi.get()
              if ipinst is not None:
                  ipinst.hooks.synchronize_with_editor(filepath, lnum, 0)
         # vds: <<
                 
         # return all our info assembled as a single string
         # return '%s\n\n%s\n%s' % (head,'\n'.join(frames),''.join(exception[0]) )
         return [head] + frames + [''.join(exception[0])]
     
     def debugger(self,force=False):
         """Call up the pdb debugger if desired, always clean up the tb
         reference.
 
         Keywords:
 
           - force(False): by default, this routine checks the instance call_pdb
           flag and does not actually invoke the debugger if the flag is false.
           The 'force' option forces the debugger to activate even if the flag
           is false.
 
         If the call_pdb flag is set, the pdb interactive debugger is
         invoked. In all cases, the self.tb reference to the current traceback
         is deleted to prevent lingering references which hamper memory
         management.
 
         Note that each call to pdb() does an 'import readline', so if your app
         requires a special setup for the readline completers, you'll have to
         fix that by hand after invoking the exception handler."""
 
         if force or self.call_pdb:
             if self.pdb is None:
                 self.pdb = debugger.Pdb(
                     self.color_scheme_table.active_scheme_name)
             # the system displayhook may have changed, restore the original
             # for pdb
             display_trap = DisplayTrap(hook=sys.__displayhook__)
             with display_trap:
                 self.pdb.reset()
                 # Find the right frame so we don't pop up inside ipython itself
                 if hasattr(self,'tb') and self.tb is not None:
                     etb = self.tb
                 else:
                     etb = self.tb = sys.last_traceback
                 while self.tb is not None and self.tb.tb_next is not None:
                     self.tb = self.tb.tb_next
                 if etb and etb.tb_next:
                     etb = etb.tb_next
                 self.pdb.botframe = etb.tb_frame
                 self.pdb.interaction(self.tb.tb_frame, self.tb)
 
         if hasattr(self,'tb'):
             del self.tb
 
     def handler(self, info=None):
         (etype, evalue, etb) = info or sys.exc_info()
         self.tb = etb
         io.Term.cout.flush()
         io.Term.cerr.write(self.text(etype, evalue, etb))
         io.Term.cerr.write('\n')
 
     # Changed so an instance can just be called as VerboseTB_inst() and print
     # out the right info on its own.
     def __call__(self, etype=None, evalue=None, etb=None):
         """This hook can replace sys.excepthook (for Python 2.1 or higher)."""
         if etb is None:
             self.handler()
         else:
             self.handler((etype, evalue, etb))
         try:
             self.debugger()
         except KeyboardInterrupt:
             print "\nKeyboardInterrupt"
 
 #----------------------------------------------------------------------------
 class FormattedTB(VerboseTB, ListTB):
     """Subclass ListTB but allow calling with a traceback.
 
     It can thus be used as a sys.excepthook for Python > 2.1.
 
     Also adds 'Context' and 'Verbose' modes, not available in ListTB.
 
     Allows a tb_offset to be specified. This is useful for situations where
     one needs to remove a number of topmost frames from the traceback (such as
     occurs with python programs that themselves execute other python code,
     like Python shells).  """
     
     def __init__(self, mode='Plain', color_scheme='Linux',
                  tb_offset=0, long_header=0, call_pdb=0, include_vars=0):
 
         # NEVER change the order of this list. Put new modes at the end:
         self.valid_modes = ['Plain','Context','Verbose']
         self.verbose_modes = self.valid_modes[1:3]
 
         VerboseTB.__init__(self,color_scheme,tb_offset,long_header,
                            call_pdb=call_pdb,include_vars=include_vars)
+
+        # Different types of tracebacks are joined with different separators to
+        # form a single string.  They are taken from this dict
+        self._join_chars = dict(Plain='', Context='\n', Verbose='\n')
+        # set_mode also sets the tb_join_char attribute
         self.set_mode(mode)
         
     def _extract_tb(self,tb):
         if tb:
             return traceback.extract_tb(tb)
         else:
             return None
 
-    def structured_traceback(self, etype, value, tb, tb_offset=None,
-                             context=5, mode=None):
+    def structured_traceback(self, etype, value, tb, tb_offset=None, context=5):
         tb_offset = self.tb_offset if tb_offset is None else tb_offset
-        mode = self.mode if mode is None else mode
+        mode = self.mode
         if mode in self.verbose_modes:
             # Verbose modes need a full traceback
             return VerboseTB.structured_traceback(
                 self, etype, value, tb, tb_offset, context
             )
         else:
             # We must check the source cache because otherwise we can print
             # out-of-date source code.
             linecache.checkcache()
             # Now we can extract and format the exception
             elist = self._extract_tb(tb)
             return ListTB.structured_traceback(
                 self, etype, value, elist, tb_offset, context
             )
 
-    def text(self, etype, value, tb, tb_offset=None, context=5, mode=None):
-        """Return formatted traceback.
-
-        If the optional mode parameter is given, it overrides the current
-        mode."""
-
-        mode = self.mode if mode is None else mode
-        tb_list = self.structured_traceback(etype, value, tb, tb_offset,
-                                            context, mode)
-        return '\n'.join(tb_list)
+    def stb2text(self, stb):
+        """Convert a structured traceback (a list) to a string."""
+        return self.tb_join_char.join(stb)
         
 
     def set_mode(self,mode=None):
         """Switch to the desired mode.
 
         If mode is not specified, cycles through the available modes."""
 
         if not mode:
             new_idx = ( self.valid_modes.index(self.mode) + 1 ) % \
                       len(self.valid_modes)
             self.mode = self.valid_modes[new_idx]
         elif mode not in self.valid_modes:
             raise ValueError, 'Unrecognized mode in FormattedTB: <'+mode+'>\n'\
                   'Valid modes: '+str(self.valid_modes)
         else:
             self.mode = mode
         # include variable details only in 'Verbose' mode
         self.include_vars = (self.mode == self.valid_modes[2])
+        # Set the join character for generating text tracebacks
+        self.tb_join_char = self._join_chars[mode]
 
     # some convenient shorcuts
     def plain(self):
         self.set_mode(self.valid_modes[0])
 
     def context(self):
         self.set_mode(self.valid_modes[1])
 
     def verbose(self):
         self.set_mode(self.valid_modes[2])
 
 #----------------------------------------------------------------------------
 class AutoFormattedTB(FormattedTB):
     """A traceback printer which can be called on the fly.
 
     It will find out about exceptions by itself.
 
     A brief example:
     
     AutoTB = AutoFormattedTB(mode = 'Verbose',color_scheme='Linux')
     try:
       ...
     except:
       AutoTB()  # or AutoTB(out=logfile) where logfile is an open file object
     """
     
     def __call__(self,etype=None,evalue=None,etb=None,
                  out=None,tb_offset=None):
         """Print out a formatted exception traceback.
 
         Optional arguments:
           - out: an open file-like object to direct output to.
 
           - tb_offset: the number of frames to skip over in the stack, on a
           per-call basis (this overrides temporarily the instance's tb_offset
           given at initialization time.  """
 
         if out is None:
             if self.out_stream == 'stdout':
                 out = sys.stdout
             else:
                 out = io.Term.cerr
         out.flush()
         out.write(self.text(etype, evalue, etb, tb_offset))
         out.write('\n')
         out.flush()
         # FIXME: we should remove the auto pdb behavior from here and leave
         # that to the clients.
         try:
             self.debugger()
         except KeyboardInterrupt:
             print "\nKeyboardInterrupt"
 
     def structured_traceback(self, etype=None, value=None, tb=None,
-                             tb_offset=None, context=5, mode=None):
+                             tb_offset=None, context=5):
         if etype is None:
             etype,value,tb = sys.exc_info()
         self.tb = tb
         return FormattedTB.structured_traceback(
-            self, etype, value, tb, tb_offset, context, mode )
+            self, etype, value, tb, tb_offset, context)
 
 #---------------------------------------------------------------------------
 
 # A simple class to preserve Nathan's original functionality.
 class ColorTB(FormattedTB):
     """Shorthand to initialize a FormattedTB in Linux colors mode."""
     def __init__(self,color_scheme='Linux',call_pdb=0):
         FormattedTB.__init__(self,color_scheme=color_scheme,
                              call_pdb=call_pdb)
 
 
 class SyntaxTB(ListTB):
     """Extension which holds some state: the last exception value"""
 
     def __init__(self,color_scheme = 'NoColor'):
         ListTB.__init__(self,color_scheme)
         self.last_syntax_error = None
 
     def __call__(self, etype, value, elist):
         self.last_syntax_error = value
         ListTB.__call__(self,etype,value,elist)
 
     def clear_err_state(self):
         """Return the current error state and clear it"""
         e = self.last_syntax_error
         self.last_syntax_error = None
         return e
 
-    def text(self, etype, value, tb, tb_offset=None, context=5):
-        """Return formatted traceback.
+    def stb2text(self, stb):
+        """Convert a structured traceback (a list) to a string."""
+        return ''.join(stb)
 
-        Subclasses may override this if they add extra arguments.
-        """
-        tb_list = self.structured_traceback(etype, value, tb,
-                                            tb_offset, context)
-        return ''.join(tb_list)
 
 #----------------------------------------------------------------------------
 # module testing (minimal)
 if __name__ == "__main__":
     def spam(c, (d, e)):
         x = c + d
         y = c * d
         foo(x, y)
 
     def foo(a, b, bar=1):
         eggs(a, b + bar)
 
     def eggs(f, g, z=globals()):
         h = f + g
         i = f - g
         return h / i
 
     print ''
     print '*** Before ***'
     try:
         print spam(1, (2, 3))
     except:
         traceback.print_exc()
     print ''
     
     handler = ColorTB()
     print '*** ColorTB ***'
     try:
         print spam(1, (2, 3))
     except:
         apply(handler, sys.exc_info() )
     print ''
     
     handler = VerboseTB()
     print '*** VerboseTB ***'
     try:
         print spam(1, (2, 3))
     except:
         apply(handler, sys.exc_info() )
     print ''
     

IPython/frontend/qt/console/frontend_widget.py

@@ -1,377 +1,384 b''
 # Standard library imports
 import signal
 import sys
 
 # System library imports
 from pygments.lexers import PythonLexer
 from PyQt4 import QtCore, QtGui
 import zmq
 
 # Local imports
 from IPython.core.inputsplitter import InputSplitter
 from IPython.frontend.qt.base_frontend_mixin import BaseFrontendMixin
 from call_tip_widget import CallTipWidget
 from completion_lexer import CompletionLexer
 from console_widget import HistoryConsoleWidget
 from pygments_highlighter import PygmentsHighlighter
 
 
 class FrontendHighlighter(PygmentsHighlighter):
     """ A PygmentsHighlighter that can be turned on and off and that ignores
         prompts.
     """
 
     def __init__(self, frontend):
         super(FrontendHighlighter, self).__init__(frontend._control.document())
         self._current_offset = 0
         self._frontend = frontend
         self.highlighting_on = False
 
     def highlightBlock(self, qstring):
         """ Highlight a block of text. Reimplemented to highlight selectively.
         """
         if not self.highlighting_on:
             return
 
         # The input to this function is unicode string that may contain
         # paragraph break characters, non-breaking spaces, etc. Here we acquire
         # the string as plain text so we can compare it.
         current_block = self.currentBlock()
         string = self._frontend._get_block_plain_text(current_block)
 
         # Decide whether to check for the regular or continuation prompt.
         if current_block.contains(self._frontend._prompt_pos):
             prompt = self._frontend._prompt
         else:
             prompt = self._frontend._continuation_prompt
 
         # Don't highlight the part of the string that contains the prompt.
         if string.startswith(prompt):
             self._current_offset = len(prompt)
             qstring.remove(0, len(prompt))
         else:
             self._current_offset = 0
 
         PygmentsHighlighter.highlightBlock(self, qstring)
 
     def rehighlightBlock(self, block):
         """ Reimplemented to temporarily enable highlighting if disabled.
         """
         old = self.highlighting_on
         self.highlighting_on = True
         super(FrontendHighlighter, self).rehighlightBlock(block)
         self.highlighting_on = old
 
     def setFormat(self, start, count, format):
         """ Reimplemented to highlight selectively.
         """
         start += self._current_offset
         PygmentsHighlighter.setFormat(self, start, count, format)
 
 
 class FrontendWidget(HistoryConsoleWidget, BaseFrontendMixin):
     """ A Qt frontend for a generic Python kernel.
     """
    
     # Emitted when an 'execute_reply' has been received from the kernel and
     # processed by the FrontendWidget.
     executed = QtCore.pyqtSignal(object)
 
     # Protected class attributes.
     _highlighter_class = FrontendHighlighter
     _input_splitter_class = InputSplitter
 
     #---------------------------------------------------------------------------
     # 'object' interface
     #---------------------------------------------------------------------------
     
     def __init__(self, *args, **kw):
         super(FrontendWidget, self).__init__(*args, **kw)
 
         # FrontendWidget protected variables.
         self._call_tip_widget = CallTipWidget(self._control)
         self._completion_lexer = CompletionLexer(PythonLexer())
         self._hidden = False
         self._highlighter = self._highlighter_class(self)
         self._input_splitter = self._input_splitter_class(input_mode='replace')
         self._kernel_manager = None
 
         # Configure the ConsoleWidget.
         self.tab_width = 4
         self._set_continuation_prompt('... ')
 
         # Connect signal handlers.
         document = self._control.document()
         document.contentsChange.connect(self._document_contents_change)
 
     #---------------------------------------------------------------------------
     # 'ConsoleWidget' abstract interface
     #---------------------------------------------------------------------------
 
     def _is_complete(self, source, interactive):
         """ Returns whether 'source' can be completely processed and a new
             prompt created. When triggered by an Enter/Return key press,
             'interactive' is True; otherwise, it is False.
         """
         complete = self._input_splitter.push(source.expandtabs(4))
         if interactive:
             complete = not self._input_splitter.push_accepts_more()
         return complete
 
     def _execute(self, source, hidden):
         """ Execute 'source'. If 'hidden', do not show any output.
         """
         self.kernel_manager.xreq_channel.execute(source)
         self._hidden = hidden
 
     def _execute_interrupt(self):
         """ Attempts to stop execution. Returns whether this method has an
             implementation.
         """
         self._interrupt_kernel()
         return True
         
     def _prompt_started_hook(self):
         """ Called immediately after a new prompt is displayed.
         """
         if not self._reading:
             self._highlighter.highlighting_on = True
 
     def _prompt_finished_hook(self):
         """ Called immediately after a prompt is finished, i.e. when some input
             will be processed and a new prompt displayed.
         """
         if not self._reading:
             self._highlighter.highlighting_on = False
 
     def _tab_pressed(self):
         """ Called when the tab key is pressed. Returns whether to continue
             processing the event.
         """
         self._keep_cursor_in_buffer()
         cursor = self._get_cursor()
         return not self._complete()
 
     #---------------------------------------------------------------------------
     # 'ConsoleWidget' protected interface
     #---------------------------------------------------------------------------
 
     def _show_continuation_prompt(self):
         """ Reimplemented for auto-indentation.
         """
         super(FrontendWidget, self)._show_continuation_prompt()
         spaces = self._input_splitter.indent_spaces
         self._append_plain_text('\t' * (spaces / self.tab_width))
         self._append_plain_text(' ' * (spaces % self.tab_width))
 
     #---------------------------------------------------------------------------
     # 'BaseFrontendMixin' abstract interface
     #---------------------------------------------------------------------------
 
     def _handle_complete_reply(self, rep):
         """ Handle replies for tab completion.
         """
         cursor = self._get_cursor()
         if rep['parent_header']['msg_id'] == self._complete_id and \
                 cursor.position() == self._complete_pos:
             text = '.'.join(self._get_context())
             cursor.movePosition(QtGui.QTextCursor.Left, n=len(text))
             self._complete_with_items(cursor, rep['content']['matches'])
 
     def _handle_execute_reply(self, msg):
         """ Handles replies for code execution.
         """
         if not self._hidden:
             # Make sure that all output from the SUB channel has been processed
             # before writing a new prompt.
             self.kernel_manager.sub_channel.flush()
 
             content = msg['content']
             status = content['status']
             if status == 'ok':
                 self._process_execute_ok(msg)
             elif status == 'error':
                 self._process_execute_error(msg)
             elif status == 'abort':
                 self._process_execute_abort(msg)
 
             self._show_interpreter_prompt_for_reply(msg)
             self.executed.emit(msg)
 
     def _handle_input_request(self, msg):
         """ Handle requests for raw_input.
         """
         if self._hidden:
             raise RuntimeError('Request for raw input during hidden execution.')
 
         # Make sure that all output from the SUB channel has been processed
         # before entering readline mode.
         self.kernel_manager.sub_channel.flush()
 
         def callback(line):
             self.kernel_manager.rep_channel.input(line)
         self._readline(msg['content']['prompt'], callback=callback)
 
     def _handle_object_info_reply(self, rep):
         """ Handle replies for call tips.
         """
         cursor = self._get_cursor()
         if rep['parent_header']['msg_id'] == self._call_tip_id and \
                 cursor.position() == self._call_tip_pos:
             doc = rep['content']['docstring']
             if doc:
                 self._call_tip_widget.show_docstring(doc)
 
     def _handle_pyout(self, msg):
         """ Handle display hook output.
         """
         if not self._hidden and self._is_from_this_session(msg):
             self._append_plain_text(msg['content']['data'] + '\n')
 
     def _handle_stream(self, msg):
         """ Handle stdout, stderr, and stdin.
         """
         if not self._hidden and self._is_from_this_session(msg):
             self._append_plain_text(msg['content']['data'])
             self._control.moveCursor(QtGui.QTextCursor.End)
     
     def _started_channels(self):
         """ Called when the KernelManager channels have started listening or 
             when the frontend is assigned an already listening KernelManager.
         """
         self._reset()
         self._append_plain_text(self._get_banner())
         self._show_interpreter_prompt()
 
     def _stopped_channels(self):
         """ Called when the KernelManager channels have stopped listening or
             when a listening KernelManager is removed from the frontend.
         """
         # FIXME: Print a message here?
         pass
 
     #---------------------------------------------------------------------------
     # 'FrontendWidget' interface
     #---------------------------------------------------------------------------
 
     def execute_file(self, path, hidden=False):
         """ Attempts to execute file with 'path'. If 'hidden', no output is
             shown.
         """
         self.execute('execfile("%s")' % path, hidden=hidden)
 
     #---------------------------------------------------------------------------
     # 'FrontendWidget' protected interface
     #---------------------------------------------------------------------------
 
     def _call_tip(self):
         """ Shows a call tip, if appropriate, at the current cursor location.
         """
         # Decide if it makes sense to show a call tip
         cursor = self._get_cursor()
         cursor.movePosition(QtGui.QTextCursor.Left)
         document = self._control.document()
         if document.characterAt(cursor.position()).toAscii() != '(':
             return False
         context = self._get_context(cursor)
         if not context:
             return False
 
         # Send the metadata request to the kernel
         name = '.'.join(context)
         self._call_tip_id = self.kernel_manager.xreq_channel.object_info(name)
         self._call_tip_pos = self._get_cursor().position()
         return True
 
     def _complete(self):
         """ Performs completion at the current cursor location.
         """
         # Decide if it makes sense to do completion
         context = self._get_context()
         if not context:
             return False
 
         # Send the completion request to the kernel
         text = '.'.join(context)
+        
+        # FIXME - Evan: we need the position of the cursor in the current input
+        # buffer.  I tried this line below but the numbers I get are bogus. -
+        # Not sure what to do.  fperez.
+        cursor_pos = self._get_cursor().position()
+        
         self._complete_id = self.kernel_manager.xreq_channel.complete(
-            text, self._get_input_buffer_cursor_line(), self.input_buffer)
+            text, self._get_input_buffer_cursor_line(), cursor_pos,
+            self.input_buffer)
         self._complete_pos = self._get_cursor().position()
         return True
 
     def _get_banner(self):
         """ Gets a banner to display at the beginning of a session.
         """
         banner = 'Python %s on %s\nType "help", "copyright", "credits" or ' \
             '"license" for more information.'
         return banner % (sys.version, sys.platform)
 
     def _get_context(self, cursor=None):
         """ Gets the context at the current cursor location.
         """
         if cursor is None:
             cursor = self._get_cursor()
         cursor.movePosition(QtGui.QTextCursor.StartOfBlock, 
                             QtGui.QTextCursor.KeepAnchor)
         text = str(cursor.selection().toPlainText())
         return self._completion_lexer.get_context(text)
 
     def _interrupt_kernel(self):
         """ Attempts to the interrupt the kernel.
         """
         if self.kernel_manager.has_kernel:
             self.kernel_manager.signal_kernel(signal.SIGINT)
         else:
             self._append_plain_text('Kernel process is either remote or '
                                     'unspecified. Cannot interrupt.\n')
 
     def _process_execute_abort(self, msg):
         """ Process a reply for an aborted execution request.
         """
         self._append_plain_text("ERROR: execution aborted\n")
 
     def _process_execute_error(self, msg):
         """ Process a reply for an execution request that resulted in an error.
         """
         content = msg['content']
         traceback = ''.join(content['traceback'])
         self._append_plain_text(traceback)
 
     def _process_execute_ok(self, msg):
         """ Process a reply for a successful execution equest.
         """
         payload = msg['content']['payload']
         for item in payload:
             if not self._process_execute_payload(item):
                 warning = 'Received unknown payload of type %s\n'
                 self._append_plain_text(warning % repr(item['source']))
 
     def _process_execute_payload(self, item):
         """ Process a single payload item from the list of payload items in an
             execution reply. Returns whether the payload was handled.
         """
         # The basic FrontendWidget doesn't handle payloads, as they are a
         # mechanism for going beyond the standard Python interpreter model.
         return False
 
     def _show_interpreter_prompt(self):
         """ Shows a prompt for the interpreter.
         """
         self._show_prompt('>>> ')
 
     def _show_interpreter_prompt_for_reply(self, msg):
         """ Shows a prompt for the interpreter given an 'execute_reply' message.
         """
         self._show_interpreter_prompt()
 
     #------ Signal handlers ----------------------------------------------------
 
     def _document_contents_change(self, position, removed, added):
         """ Called whenever the document's content changes. Display a call tip
             if appropriate.
         """
         # Calculate where the cursor should be *after* the change:
         position += added
 
         document = self._control.document()
         if position == self._get_cursor().position():
             self._call_tip()

IPython/zmq/ipkernel.py

@@ -1,387 +1,398 b''
 #!/usr/bin/env python
 """A simple interactive kernel that talks to a frontend over 0MQ.
 
 Things to do:
 
 * Implement `set_parent` logic. Right before doing exec, the Kernel should
   call set_parent on all the PUB objects with the message about to be executed.
 * Implement random port and security key logic.
 * Implement control messages.
 * Implement event loop and poll version.
 """
 
 #-----------------------------------------------------------------------------
 # Imports
 #-----------------------------------------------------------------------------
 
 # Standard library imports.
 import __builtin__
 import sys
 import time
 import traceback
 
 # System library imports.
 import zmq
 
 # Local imports.
 from IPython.config.configurable import Configurable
 from IPython.utils.traitlets import Instance
 from completer import KernelCompleter
 from entry_point import base_launch_kernel, make_argument_parser, make_kernel, \
     start_kernel
 from iostream import OutStream
 from session import Session, Message
 from zmqshell import ZMQInteractiveShell
 
 #-----------------------------------------------------------------------------
 # Main kernel class
 #-----------------------------------------------------------------------------
 
 class Kernel(Configurable):
 
     #---------------------------------------------------------------------------
     # Kernel interface
     #---------------------------------------------------------------------------
 
     shell = Instance('IPython.core.interactiveshell.InteractiveShellABC')
     session = Instance(Session)
     reply_socket = Instance('zmq.Socket')
     pub_socket = Instance('zmq.Socket')
     req_socket = Instance('zmq.Socket')
 
     # Maps user-friendly backend names to matplotlib backend identifiers.
     _pylab_map = { 'tk': 'TkAgg',
                    'gtk': 'GTKAgg',
                    'wx': 'WXAgg',
                    'qt': 'Qt4Agg', # qt3 not supported
                    'qt4': 'Qt4Agg',
                    'payload-svg' : \
                        'module://IPython.zmq.pylab.backend_payload_svg' }
 
     def __init__(self, **kwargs):
         super(Kernel, self).__init__(**kwargs)
 
         # Initialize the InteractiveShell subclass
         self.shell = ZMQInteractiveShell.instance()
         self.shell.displayhook.session = self.session
         self.shell.displayhook.pub_socket = self.pub_socket
 
         # TMP - hack while developing
         self.shell._reply_content = None
 
         # Build dict of handlers for message types
         msg_types = [ 'execute_request', 'complete_request', 
                       'object_info_request', 'prompt_request',
                       'history_request' ]
         self.handlers = {}
         for msg_type in msg_types:
             self.handlers[msg_type] = getattr(self, msg_type)
 
     def activate_pylab(self, backend=None, import_all=True):
         """ Activates pylab in this kernel's namespace.
 
         Parameters:
         -----------
         backend : str, optional
             A valid backend name.
 
         import_all : bool, optional
             If true, an 'import *' is done from numpy and pylab.
         """
         # FIXME: This is adapted from IPython.lib.pylabtools.pylab_activate.
         #        Common functionality should be refactored.
 
         # We must set the desired backend before importing pylab.
         import matplotlib
         if backend:
             backend_id = self._pylab_map[backend]
             if backend_id.startswith('module://'):
                 # Work around bug in matplotlib: matplotlib.use converts the
                 # backend_id to lowercase even if a module name is specified!
                 matplotlib.rcParams['backend'] = backend_id
             else:
                 matplotlib.use(backend_id)
 
         # Import numpy as np/pyplot as plt are conventions we're trying to
         # somewhat standardize on. Making them available to users by default
         # will greatly help this.
         exec ("import numpy\n"
               "import matplotlib\n"
               "from matplotlib import pylab, mlab, pyplot\n"
               "np = numpy\n"
               "plt = pyplot\n"
               ) in self.shell.user_ns
 
         if import_all:
             exec("from matplotlib.pylab import *\n"
                  "from numpy import *\n") in self.shell.user_ns
 
         matplotlib.interactive(True)
 
     def start(self):
         """ Start the kernel main loop.
         """
         while True:
             ident = self.reply_socket.recv()
             assert self.reply_socket.rcvmore(), "Missing message part."
             msg = self.reply_socket.recv_json()
             omsg = Message(msg)
             print>>sys.__stdout__
             print>>sys.__stdout__, omsg
             handler = self.handlers.get(omsg.msg_type, None)
             if handler is None:
                 print >> sys.__stderr__, "UNKNOWN MESSAGE TYPE:", omsg
             else:
                 handler(ident, omsg)
 
     #---------------------------------------------------------------------------
     # Kernel request handlers
     #---------------------------------------------------------------------------
 
     def execute_request(self, ident, parent):
         try:
             code = parent[u'content'][u'code']
         except:
             print>>sys.__stderr__, "Got bad msg: "
             print>>sys.__stderr__, Message(parent)
             return
         pyin_msg = self.session.msg(u'pyin',{u'code':code}, parent=parent)
         self.pub_socket.send_json(pyin_msg)
 
         try:
             # Replace raw_input. Note that is not sufficient to replace 
             # raw_input in the user namespace.
             raw_input = lambda prompt='': self._raw_input(prompt, ident, parent)
             __builtin__.raw_input = raw_input
 
             # Set the parent message of the display hook and out streams.
             self.shell.displayhook.set_parent(parent)
             sys.stdout.set_parent(parent)
             sys.stderr.set_parent(parent)
 
             # FIXME: runlines calls the exception handler itself.  We should
             # clean this up.
             self.shell._reply_content = None
             self.shell.runlines(code)
         except:
             # FIXME: this code right now isn't being used yet by default,
             # because the runlines() call above directly fires off exception
             # reporting.  This code, therefore, is only active in the scenario
             # where runlines itself has an unhandled exception.  We need to
             # uniformize this, for all exception construction to come from a
             # single location in the codbase.
             etype, evalue, tb = sys.exc_info()
             tb_list = traceback.format_exception(etype, evalue, tb)
             reply_content = self.shell._showtraceback(etype, evalue, tb_list)
         else:
             payload = self.shell.payload_manager.read_payload()
             # Be agressive about clearing the payload because we don't want
             # it to sit in memory until the next execute_request comes in.
             self.shell.payload_manager.clear_payload()
             reply_content = { 'status' : 'ok', 'payload' : payload }
 
         # Compute the prompt information
         prompt_number = self.shell.displayhook.prompt_count
         reply_content['prompt_number'] = prompt_number        
         prompt_string = self.shell.displayhook.prompt1.peek_next_prompt()
         next_prompt = {'prompt_string' : prompt_string,
                        'prompt_number' : prompt_number+1,
                        'input_sep'     : self.shell.displayhook.input_sep}
         reply_content['next_prompt'] = next_prompt
 
         # TMP - fish exception info out of shell, possibly left there by
         # runlines
         if self.shell._reply_content is not None:
             reply_content.update(self.shell._reply_content)
 
         # Flush output before sending the reply.
         sys.stderr.flush()
         sys.stdout.flush()
 
         # Send the reply.
         reply_msg = self.session.msg(u'execute_reply', reply_content, parent)
         print>>sys.__stdout__, Message(reply_msg)
         self.reply_socket.send(ident, zmq.SNDMORE)
         self.reply_socket.send_json(reply_msg)
         if reply_msg['content']['status'] == u'error':
             self._abort_queue()
 
     def complete_request(self, ident, parent):
         matches = {'matches' : self._complete(parent),
                    'status' : 'ok'}
         completion_msg = self.session.send(self.reply_socket, 'complete_reply',
                                            matches, parent, ident)
         print >> sys.__stdout__, completion_msg
 
     def object_info_request(self, ident, parent):
         context = parent['content']['oname'].split('.')
         object_info = self._object_info(context)
         msg = self.session.send(self.reply_socket, 'object_info_reply',
                                 object_info, parent, ident)
         print >> sys.__stdout__, msg
 
     def prompt_request(self, ident, parent):
         prompt_number = self.shell.displayhook.prompt_count
         prompt_string = self.shell.displayhook.prompt1.peek_next_prompt()
         content = {'prompt_string' : prompt_string,
                    'prompt_number' : prompt_number+1}
         msg = self.session.send(self.reply_socket, 'prompt_reply',
                                 content, parent, ident)
         print >> sys.__stdout__, msg
 
     def history_request(self, ident, parent):
         output = parent['content'].get('output', True)
         index = parent['content'].get('index')
         raw = parent['content'].get('raw', False)
         hist = self.shell.get_history(index=index, raw=raw, output=output)
         content = {'history' : hist}
         msg = self.session.send(self.reply_socket, 'history_reply',
                                 content, parent, ident)
         print >> sys.__stdout__, msg
         
     #---------------------------------------------------------------------------
     # Protected interface
     #---------------------------------------------------------------------------
 
     def _abort_queue(self):
         while True:
             try:
                 ident = self.reply_socket.recv(zmq.NOBLOCK)
             except zmq.ZMQError, e:
                 if e.errno == zmq.EAGAIN:
                     break
             else:
                 assert self.reply_socket.rcvmore(), "Unexpected missing message part."
                 msg = self.reply_socket.recv_json()
             print>>sys.__stdout__, "Aborting:"
             print>>sys.__stdout__, Message(msg)
             msg_type = msg['msg_type']
             reply_type = msg_type.split('_')[0] + '_reply'
             reply_msg = self.session.msg(reply_type, {'status' : 'aborted'}, msg)
             print>>sys.__stdout__, Message(reply_msg)
             self.reply_socket.send(ident,zmq.SNDMORE)
             self.reply_socket.send_json(reply_msg)
             # We need to wait a bit for requests to come in. This can probably
             # be set shorter for true asynchronous clients.
             time.sleep(0.1)
 
     def _raw_input(self, prompt, ident, parent):
         # Flush output before making the request.
         sys.stderr.flush()
         sys.stdout.flush()
 
         # Send the input request.
         content = dict(prompt=prompt)
         msg = self.session.msg(u'input_request', content, parent)
         self.req_socket.send_json(msg)
 
         # Await a response.
         reply = self.req_socket.recv_json()
         try:
             value = reply['content']['value']
         except:
             print>>sys.__stderr__, "Got bad raw_input reply: "
             print>>sys.__stderr__, Message(parent)
             value = ''
         return value
     
     def _complete(self, msg):
-        return self.shell.complete(msg.content.line)
+        #from IPython.utils.io import rprint  # dbg
+        #rprint('\n\n**MSG**\n\n', msg)  # dbg
+        #import traceback; rprint(''.join(traceback.format_stack())) # dbg
+        c = msg['content']
+        try:
+            cpos = int(c['cursor_pos'])
+        except:
+            # If we don't get something that we can convert to an integer, at
+            # leasat attempt the completion guessing the cursor is at the end
+            # of the text
+            cpos = len(c['text'])
+        return self.shell.complete(c['text'], c['line'], cpos)
 
     def _object_info(self, context):
         symbol, leftover = self._symbol_from_context(context)
         if symbol is not None and not leftover:
             doc = getattr(symbol, '__doc__', '')
         else:
             doc = ''
         object_info = dict(docstring = doc)
         return object_info
 
     def _symbol_from_context(self, context):
         if not context:
             return None, context
 
         base_symbol_string = context[0]
         symbol = self.shell.user_ns.get(base_symbol_string, None)
         if symbol is None:
             symbol = __builtin__.__dict__.get(base_symbol_string, None)
         if symbol is None:
             return None, context
 
         context = context[1:]
         for i, name in enumerate(context):
             new_symbol = getattr(symbol, name, None)
             if new_symbol is None:
                 return symbol, context[i:]
             else:
                 symbol = new_symbol
 
         return symbol, []
 
 #-----------------------------------------------------------------------------
 # Kernel main and launch functions
 #-----------------------------------------------------------------------------
 
 def launch_kernel(xrep_port=0, pub_port=0, req_port=0, independent=False,
                   pylab=False):
     """ Launches a localhost kernel, binding to the specified ports.
 
     Parameters
     ----------
     xrep_port : int, optional
         The port to use for XREP channel.
 
     pub_port : int, optional
         The port to use for the SUB channel.
 
     req_port : int, optional
         The port to use for the REQ (raw input) channel.
 
     independent : bool, optional (default False) 
         If set, the kernel process is guaranteed to survive if this process
         dies. If not set, an effort is made to ensure that the kernel is killed
         when this process dies. Note that in this case it is still good practice
         to kill kernels manually before exiting.
 
     pylab : bool or string, optional (default False)
         If not False, the kernel will be launched with pylab enabled. If a
         string is passed, matplotlib will use the specified backend. Otherwise,
         matplotlib's default backend will be used.
 
     Returns
     -------
     A tuple of form:
         (kernel_process, xrep_port, pub_port, req_port)
     where kernel_process is a Popen object and the ports are integers.
     """
     extra_arguments = []
     if pylab:
         extra_arguments.append('--pylab')
         if isinstance(pylab, basestring):
             extra_arguments.append(pylab)
     return base_launch_kernel('from IPython.zmq.ipkernel import main; main()',
                               xrep_port, pub_port, req_port, independent, 
                               extra_arguments)
 
 def main():
     """ The IPython kernel main entry point.
     """
     parser = make_argument_parser()
     parser.add_argument('--pylab', type=str, metavar='GUI', nargs='?', 
                         const='auto', help = \
 "Pre-load matplotlib and numpy for interactive use. If GUI is not \
 given, the GUI backend is matplotlib's, otherwise use one of: \
 ['tk', 'gtk', 'qt', 'wx', 'payload-svg'].")
     namespace = parser.parse_args()
 
     kernel = make_kernel(namespace, Kernel, OutStream)
     if namespace.pylab:
         if namespace.pylab == 'auto':
             kernel.activate_pylab()
         else:
             kernel.activate_pylab(namespace.pylab)
     
     start_kernel(namespace, kernel)
 
 if __name__ == '__main__':
     main()

IPython/zmq/kernelmanager.py

@@ -1,578 +1,578 b''
 """Base classes to manage the interaction with a running kernel.
 
 Todo
 ====
 
 * Create logger to handle debugging and console messages.
 """
 
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2008-2010  The IPython Development Team
 #
 #  Distributed under the terms of the BSD License.  The full license is in
 #  the file COPYING, distributed as part of this software.
 #-----------------------------------------------------------------------------
 
 #-----------------------------------------------------------------------------
 # Imports
 #-----------------------------------------------------------------------------
 
 # Standard library imports.
 from Queue import Queue, Empty
 from subprocess import Popen
 from threading import Thread
 import time
 
 # System library imports.
 import zmq
 from zmq import POLLIN, POLLOUT, POLLERR
 from zmq.eventloop import ioloop
 
 # Local imports.
 from IPython.utils.traitlets import HasTraits, Any, Instance, Type, TCPAddress
 from session import Session
 
 #-----------------------------------------------------------------------------
 # Constants and exceptions
 #-----------------------------------------------------------------------------
 
 LOCALHOST = '127.0.0.1'
 
 class InvalidPortNumber(Exception):
     pass
 
 #-----------------------------------------------------------------------------
 # ZMQ Socket Channel classes
 #-----------------------------------------------------------------------------
 
 class ZmqSocketChannel(Thread):
     """The base class for the channels that use ZMQ sockets.
     """
     context = None
     session = None
     socket = None
     ioloop = None
     iostate = None
     _address = None
 
     def __init__(self, context, session, address):
         """Create a channel
 
         Parameters
         ----------
         context : :class:`zmq.Context`
             The ZMQ context to use.
         session : :class:`session.Session`
             The session to use.
         address : tuple
             Standard (ip, port) tuple that the kernel is listening on.
         """
         super(ZmqSocketChannel, self).__init__()
         self.daemon = True
 
         self.context = context
         self.session = session
         if address[1] == 0:
             message = 'The port number for a channel cannot be 0.'
             raise InvalidPortNumber(message)
         self._address = address
 
     def stop(self):
         """Stop the channel's activity.
 
         This calls :method:`Thread.join` and returns when the thread
         terminates. :class:`RuntimeError` will be raised if 
         :method:`self.start` is called again.
         """
         self.join()
 
     @property
     def address(self):
         """Get the channel's address as an (ip, port) tuple.
         
         By the default, the address is (localhost, 0), where 0 means a random
         port.
         """
         return self._address
 
     def add_io_state(self, state):
         """Add IO state to the eventloop.
 
         Parameters
         ----------
         state : zmq.POLLIN|zmq.POLLOUT|zmq.POLLERR
             The IO state flag to set.
 
         This is thread safe as it uses the thread safe IOLoop.add_callback.
         """
         def add_io_state_callback():
             if not self.iostate & state:
                 self.iostate = self.iostate | state
                 self.ioloop.update_handler(self.socket, self.iostate)
         self.ioloop.add_callback(add_io_state_callback)
 
     def drop_io_state(self, state):
         """Drop IO state from the eventloop.
 
         Parameters
         ----------
         state : zmq.POLLIN|zmq.POLLOUT|zmq.POLLERR
             The IO state flag to set.
 
         This is thread safe as it uses the thread safe IOLoop.add_callback.
         """
         def drop_io_state_callback():
             if self.iostate & state:
                 self.iostate = self.iostate & (~state)
                 self.ioloop.update_handler(self.socket, self.iostate)
         self.ioloop.add_callback(drop_io_state_callback)
 
 
 class XReqSocketChannel(ZmqSocketChannel):
     """The XREQ channel for issues request/replies to the kernel.
     """
 
     command_queue = None
 
     def __init__(self, context, session, address):
         self.command_queue = Queue()
         super(XReqSocketChannel, self).__init__(context, session, address)
 
     def run(self):
         """The thread's main activity.  Call start() instead."""
         self.socket = self.context.socket(zmq.XREQ)
         self.socket.setsockopt(zmq.IDENTITY, self.session.session)
         self.socket.connect('tcp://%s:%i' % self.address)
         self.ioloop = ioloop.IOLoop()
         self.iostate = POLLERR|POLLIN
         self.ioloop.add_handler(self.socket, self._handle_events, 
                                 self.iostate)
         self.ioloop.start()
 
     def stop(self):
         self.ioloop.stop()
         super(XReqSocketChannel, self).stop()
 
     def call_handlers(self, msg):
         """This method is called in the ioloop thread when a message arrives.
 
         Subclasses should override this method to handle incoming messages.
         It is important to remember that this method is called in the thread
         so that some logic must be done to ensure that the application leve
         handlers are called in the application thread.
         """
         raise NotImplementedError('call_handlers must be defined in a subclass.')
 
     def execute(self, code):
         """Execute code in the kernel.
 
         Parameters
         ----------
         code : str
             A string of Python code.
 
         Returns
         -------
         The msg_id of the message sent.
         """
         # Create class for content/msg creation. Related to, but possibly
         # not in Session.
         content = dict(code=code)
         msg = self.session.msg('execute_request', content)
         self._queue_request(msg)
         return msg['header']['msg_id']
 
-    def complete(self, text, line, block=None):
+    def complete(self, text, line, cursor_pos, block=None):
         """Tab complete text, line, block in the kernel's namespace.
 
         Parameters
         ----------
         text : str
             The text to complete.
         line : str
             The full line of text that is the surrounding context for the 
             text to complete.
         block : str
             The full block of code in which the completion is being requested.
 
         Returns
         -------
         The msg_id of the message sent.
         """
-        content = dict(text=text, line=line)
+        content = dict(text=text, line=line, block=block, cursor_pos=cursor_pos)
         msg = self.session.msg('complete_request', content)
         self._queue_request(msg)
         return msg['header']['msg_id']
 
     def object_info(self, oname):
         """Get metadata information about an object.
 
         Parameters
         ----------
         oname : str
             A string specifying the object name.
         
         Returns
         -------
         The msg_id of the message sent.
         """
         content = dict(oname=oname)
         msg = self.session.msg('object_info_request', content)
         self._queue_request(msg)
         return msg['header']['msg_id']
 
     def _handle_events(self, socket, events):
         if events & POLLERR:
             self._handle_err()
         if events & POLLOUT:
             self._handle_send()
         if events & POLLIN:
             self._handle_recv()
 
     def _handle_recv(self):
         msg = self.socket.recv_json()
         self.call_handlers(msg)
 
     def _handle_send(self):
         try:
             msg = self.command_queue.get(False)
         except Empty:
             pass
         else:
             self.socket.send_json(msg)
         if self.command_queue.empty():
             self.drop_io_state(POLLOUT)
 
     def _handle_err(self):
         # We don't want to let this go silently, so eventually we should log.
         raise zmq.ZMQError()
 
     def _queue_request(self, msg):
         self.command_queue.put(msg)
         self.add_io_state(POLLOUT)
 
 
 class SubSocketChannel(ZmqSocketChannel):
     """The SUB channel which listens for messages that the kernel publishes.
     """
 
     def __init__(self, context, session, address):
         super(SubSocketChannel, self).__init__(context, session, address)
 
     def run(self):
         """The thread's main activity.  Call start() instead."""
         self.socket = self.context.socket(zmq.SUB)
         self.socket.setsockopt(zmq.SUBSCRIBE,'')
         self.socket.setsockopt(zmq.IDENTITY, self.session.session)
         self.socket.connect('tcp://%s:%i' % self.address)
         self.ioloop = ioloop.IOLoop()
         self.iostate = POLLIN|POLLERR
         self.ioloop.add_handler(self.socket, self._handle_events, 
                                 self.iostate)
         self.ioloop.start()
 
     def stop(self):
         self.ioloop.stop()
         super(SubSocketChannel, self).stop()
 
     def call_handlers(self, msg):
         """This method is called in the ioloop thread when a message arrives.
 
         Subclasses should override this method to handle incoming messages.
         It is important to remember that this method is called in the thread
         so that some logic must be done to ensure that the application leve
         handlers are called in the application thread.
         """
         raise NotImplementedError('call_handlers must be defined in a subclass.')
 
     def flush(self, timeout=1.0):
         """Immediately processes all pending messages on the SUB channel.
 
         Callers should use this method to ensure that :method:`call_handlers`
         has been called for all messages that have been received on the
         0MQ SUB socket of this channel.
 
         This method is thread safe.
 
         Parameters
         ----------
         timeout : float, optional
             The maximum amount of time to spend flushing, in seconds. The
             default is one second.
         """
         # We do the IOLoop callback process twice to ensure that the IOLoop
         # gets to perform at least one full poll.
         stop_time = time.time() + timeout
         for i in xrange(2):
             self._flushed = False
             self.ioloop.add_callback(self._flush)
             while not self._flushed and time.time() < stop_time:
                 time.sleep(0.01)
 
     def _handle_events(self, socket, events):
         # Turn on and off POLLOUT depending on if we have made a request
         if events & POLLERR:
             self._handle_err()
         if events & POLLIN:
             self._handle_recv()
 
     def _handle_err(self):
         # We don't want to let this go silently, so eventually we should log.
         raise zmq.ZMQError()
 
     def _handle_recv(self):
         # Get all of the messages we can
         while True:
             try:
                 msg = self.socket.recv_json(zmq.NOBLOCK)
             except zmq.ZMQError:
                 # Check the errno?
                 # Will this trigger POLLERR?
                 break
             else:
                 self.call_handlers(msg)
 
     def _flush(self):
         """Callback for :method:`self.flush`."""
         self._flushed = True
 
 
 class RepSocketChannel(ZmqSocketChannel):
     """A reply channel to handle raw_input requests that the kernel makes."""
 
     msg_queue = None
 
     def __init__(self, context, session, address):
         self.msg_queue = Queue()
         super(RepSocketChannel, self).__init__(context, session, address)
 
     def run(self):
         """The thread's main activity.  Call start() instead."""
         self.socket = self.context.socket(zmq.XREQ)
         self.socket.setsockopt(zmq.IDENTITY, self.session.session)
         self.socket.connect('tcp://%s:%i' % self.address)
         self.ioloop = ioloop.IOLoop()
         self.iostate = POLLERR|POLLIN
         self.ioloop.add_handler(self.socket, self._handle_events, 
                                 self.iostate)
         self.ioloop.start()
 
     def stop(self):
         self.ioloop.stop()
         super(RepSocketChannel, self).stop()
 
     def call_handlers(self, msg):
         """This method is called in the ioloop thread when a message arrives.
 
         Subclasses should override this method to handle incoming messages.
         It is important to remember that this method is called in the thread
         so that some logic must be done to ensure that the application leve
         handlers are called in the application thread.
         """
         raise NotImplementedError('call_handlers must be defined in a subclass.')
 
     def input(self, string):
         """Send a string of raw input to the kernel."""
         content = dict(value=string)
         msg = self.session.msg('input_reply', content)
         self._queue_reply(msg)
 
     def _handle_events(self, socket, events):
         if events & POLLERR:
             self._handle_err()
         if events & POLLOUT:
             self._handle_send()
         if events & POLLIN:
             self._handle_recv()
 
     def _handle_recv(self):
         msg = self.socket.recv_json()
         self.call_handlers(msg)
 
     def _handle_send(self):
         try:
             msg = self.msg_queue.get(False)
         except Empty:
             pass
         else:
             self.socket.send_json(msg)
         if self.msg_queue.empty():
             self.drop_io_state(POLLOUT)
 
     def _handle_err(self):
         # We don't want to let this go silently, so eventually we should log.
         raise zmq.ZMQError()
 
     def _queue_reply(self, msg):
         self.msg_queue.put(msg)
         self.add_io_state(POLLOUT)
 
 
 #-----------------------------------------------------------------------------
 # Main kernel manager class
 #-----------------------------------------------------------------------------
 
 class KernelManager(HasTraits):
     """ Manages a kernel for a frontend.
 
     The SUB channel is for the frontend to receive messages published by the
     kernel.
         
     The REQ channel is for the frontend to make requests of the kernel.
     
     The REP channel is for the kernel to request stdin (raw_input) from the
     frontend.
     """
     # The PyZMQ Context to use for communication with the kernel.
     context = Instance(zmq.Context,(),{})
 
     # The Session to use for communication with the kernel.
     session = Instance(Session,(),{})
 
     # The kernel process with which the KernelManager is communicating.
     kernel = Instance(Popen)
 
     # The addresses for the communication channels. 
     xreq_address = TCPAddress((LOCALHOST, 0))
     sub_address = TCPAddress((LOCALHOST, 0))
     rep_address = TCPAddress((LOCALHOST, 0))
 
     # The classes to use for the various channels.
     xreq_channel_class = Type(XReqSocketChannel)
     sub_channel_class = Type(SubSocketChannel)
     rep_channel_class = Type(RepSocketChannel)
     
     # Protected traits.
     _xreq_channel = Any
     _sub_channel = Any
     _rep_channel = Any
 
     #--------------------------------------------------------------------------
     # Channel management methods:
     #--------------------------------------------------------------------------
 
     def start_channels(self):
         """Starts the channels for this kernel.
 
         This will create the channels if they do not exist and then start
         them. If port numbers of 0 are being used (random ports) then you
         must first call :method:`start_kernel`. If the channels have been
         stopped and you call this, :class:`RuntimeError` will be raised.
         """
         self.xreq_channel.start()
         self.sub_channel.start()
         self.rep_channel.start()
 
     def stop_channels(self):
         """Stops the channels for this kernel.
 
         This stops the channels by joining their threads. If the channels
         were not started, :class:`RuntimeError` will be raised.
         """
         self.xreq_channel.stop()
         self.sub_channel.stop()
         self.rep_channel.stop()
 
     @property
     def channels_running(self):
         """Are all of the channels created and running?"""
         return self.xreq_channel.is_alive() \
             and self.sub_channel.is_alive() \
             and self.rep_channel.is_alive()
 
     #--------------------------------------------------------------------------
     # Kernel process management methods:
     #--------------------------------------------------------------------------
 
     def start_kernel(self, ipython=True, **kw):
         """Starts a kernel process and configures the manager to use it.
 
         If random ports (port=0) are being used, this method must be called
         before the channels are created.
 
         Parameters:
         -----------
         ipython : bool, optional (default True)
              Whether to use an IPython kernel instead of a plain Python kernel.
         """
         xreq, sub, rep = self.xreq_address, self.sub_address, self.rep_address
         if xreq[0] != LOCALHOST or sub[0] != LOCALHOST or rep[0] != LOCALHOST:
             raise RuntimeError("Can only launch a kernel on localhost."
                                "Make sure that the '*_address' attributes are "
                                "configured properly.")
 
         if ipython:
             from ipkernel import launch_kernel as launch
         else:
             from pykernel import launch_kernel as launch
         self.kernel, xrep, pub, req = launch(xrep_port=xreq[1], pub_port=sub[1],
                                              req_port=rep[1], **kw)
         self.xreq_address = (LOCALHOST, xrep)
         self.sub_address = (LOCALHOST, pub)
         self.rep_address = (LOCALHOST, req)
 
     @property
     def has_kernel(self):
         """Returns whether a kernel process has been specified for the kernel
         manager.
         """
         return self.kernel is not None
 
     def kill_kernel(self):
         """ Kill the running kernel. """
         if self.kernel is not None:
             self.kernel.kill()
             self.kernel = None
         else:
             raise RuntimeError("Cannot kill kernel. No kernel is running!")
 
     def signal_kernel(self, signum):
         """ Sends a signal to the kernel. """
         if self.kernel is not None:
             self.kernel.send_signal(signum)
         else:
             raise RuntimeError("Cannot signal kernel. No kernel is running!")
 
     @property
     def is_alive(self):
         """Is the kernel process still running?"""
         if self.kernel is not None:
             if self.kernel.poll() is None:
                 return True
             else:
                 return False
         else:
             # We didn't start the kernel with this KernelManager so we don't
             # know if it is running. We should use a heartbeat for this case.
             return True
 
     #--------------------------------------------------------------------------
     # Channels used for communication with the kernel:
     #--------------------------------------------------------------------------
 
     @property
     def xreq_channel(self):
         """Get the REQ socket channel object to make requests of the kernel."""
         if self._xreq_channel is None:
             self._xreq_channel = self.xreq_channel_class(self.context, 
                                                          self.session,
                                                          self.xreq_address)
         return self._xreq_channel
 
     @property
     def sub_channel(self):
         """Get the SUB socket channel object."""
         if self._sub_channel is None:
             self._sub_channel = self.sub_channel_class(self.context,
                                                        self.session,
                                                        self.sub_address)
         return self._sub_channel
 
     @property
     def rep_channel(self):
         """Get the REP socket channel object to handle stdin (raw_input)."""
         if self._rep_channel is None:
             self._rep_channel = self.rep_channel_class(self.context, 
                                                        self.session,
                                                        self.rep_address)
         return self._rep_channel