Merge pull request #4459 from minrk/bad-repr

protect against broken repr in lib.pretty

IPython/core/formatters.py

@@ -1,657 +1,654 b''
 # -*- coding: utf-8 -*-
 """Display formatters.
 
 Inheritance diagram:
 
 .. inheritance-diagram:: IPython.core.formatters
    :parts: 3
 
 Authors:
 
 * Robert Kern
 * Brian Granger
 """
 #-----------------------------------------------------------------------------
 # Copyright (C) 2010-2011, IPython Development Team.
 #
 # Distributed under the terms of the Modified BSD License.
 #
 # The full license is in the file COPYING.txt, distributed with this software.
 #-----------------------------------------------------------------------------
 
 #-----------------------------------------------------------------------------
 # Imports
 #-----------------------------------------------------------------------------
 
 # Stdlib imports
 import abc
 import sys
 import warnings
 
 # Our own imports
 from IPython.config.configurable import Configurable
 from IPython.lib import pretty
 from IPython.utils.traitlets import (
     Bool, Dict, Integer, Unicode, CUnicode, ObjectName, List,
 )
 from IPython.utils.py3compat import unicode_to_str, with_metaclass, PY3
 
 if PY3:
     from io import StringIO
 else:
     from StringIO import StringIO
 
 
 #-----------------------------------------------------------------------------
 # The main DisplayFormatter class
 #-----------------------------------------------------------------------------
 
 
 class DisplayFormatter(Configurable):
 
     # When set to true only the default plain text formatter will be used.
     plain_text_only = Bool(False, config=True)
     def _plain_text_only_changed(self, name, old, new):
         warnings.warn("""DisplayFormatter.plain_text_only is deprecated.
         
         Use DisplayFormatter.active_types = ['text/plain']
         for the same effect.
         """, DeprecationWarning)
         if new:
             self.active_types = ['text/plain']
         else:
             self.active_types = self.format_types
     
     active_types = List(Unicode, config=True,
         help="""List of currently active mime-types to display.
         You can use this to set a white-list for formats to display.
         
         Most users will not need to change this value.
         """)
     def _active_types_default(self):
         return self.format_types
     
     def _active_types_changed(self, name, old, new):
         for key, formatter in self.formatters.items():
             if key in new:
                 formatter.enabled = True
             else:
                 formatter.enabled = False
     
     # A dict of formatter whose keys are format types (MIME types) and whose
     # values are subclasses of BaseFormatter.
     formatters = Dict()
     def _formatters_default(self):
         """Activate the default formatters."""
         formatter_classes = [
             PlainTextFormatter,
             HTMLFormatter,
             SVGFormatter,
             PNGFormatter,
             JPEGFormatter,
             LatexFormatter,
             JSONFormatter,
             JavascriptFormatter
         ]
         d = {}
         for cls in formatter_classes:
             f = cls(parent=self)
             d[f.format_type] = f
         return d
 
     def format(self, obj, include=None, exclude=None):
         """Return a format data dict for an object.
 
         By default all format types will be computed.
 
         The following MIME types are currently implemented:
 
         * text/plain
         * text/html
         * text/latex
         * application/json
         * application/javascript
         * image/png
         * image/jpeg
         * image/svg+xml
 
         Parameters
         ----------
         obj : object
             The Python object whose format data will be computed.
         include : list or tuple, optional
             A list of format type strings (MIME types) to include in the
             format data dict. If this is set *only* the format types included
             in this list will be computed.
         exclude : list or tuple, optional
             A list of format type string (MIME types) to exclude in the format
             data dict. If this is set all format types will be computed,
             except for those included in this argument.
 
         Returns
         -------
         (format_dict, metadata_dict) : tuple of two dicts
         
             format_dict is a dictionary of key/value pairs, one of each format that was
             generated for the object. The keys are the format types, which
             will usually be MIME type strings and the values and JSON'able
             data structure containing the raw data for the representation in
             that format.
             
             metadata_dict is a dictionary of metadata about each mime-type output.
             Its keys will be a strict subset of the keys in format_dict.
         """
         format_dict = {}
         md_dict = {}
 
         for format_type, formatter in self.formatters.items():
             if include and format_type not in include:
                 continue
             if exclude and format_type in exclude:
                 continue
             
             md = None
             try:
                 data = formatter(obj)
             except:
                 # FIXME: log the exception
                 raise
             
             # formatters can return raw data or (data, metadata)
             if isinstance(data, tuple) and len(data) == 2:
                 data, md = data
             
             if data is not None:
                 format_dict[format_type] = data
             if md is not None:
                 md_dict[format_type] = md
             
         return format_dict, md_dict
 
     @property
     def format_types(self):
         """Return the format types (MIME types) of the active formatters."""
         return list(self.formatters.keys())
 
 
 #-----------------------------------------------------------------------------
 # Formatters for specific format types (text, html, svg, etc.)
 #-----------------------------------------------------------------------------
 
 
 class FormatterABC(with_metaclass(abc.ABCMeta, object)):
     """ Abstract base class for Formatters.
 
     A formatter is a callable class that is responsible for computing the
     raw format data for a particular format type (MIME type). For example,
     an HTML formatter would have a format type of `text/html` and would return
     the HTML representation of the object when called.
     """
 
     # The format type of the data returned, usually a MIME type.
     format_type = 'text/plain'
 
     # Is the formatter enabled...
     enabled = True
 
     @abc.abstractmethod
     def __call__(self, obj):
         """Return a JSON'able representation of the object.
 
         If the object cannot be formatted by this formatter, then return None
         """
         try:
             return repr(obj)
-        except TypeError:
+        except Exception:
             return None
 
 
 class BaseFormatter(Configurable):
     """A base formatter class that is configurable.
 
     This formatter should usually be used as the base class of all formatters.
     It is a traited :class:`Configurable` class and includes an extensible
     API for users to determine how their objects are formatted. The following
     logic is used to find a function to format an given object.
 
     1. The object is introspected to see if it has a method with the name
        :attr:`print_method`. If is does, that object is passed to that method
        for formatting.
     2. If no print method is found, three internal dictionaries are consulted
        to find print method: :attr:`singleton_printers`, :attr:`type_printers`
        and :attr:`deferred_printers`.
 
     Users should use these dictionaries to register functions that will be
     used to compute the format data for their objects (if those objects don't
     have the special print methods). The easiest way of using these
     dictionaries is through the :meth:`for_type` and :meth:`for_type_by_name`
     methods.
 
     If no function/callable is found to compute the format data, ``None`` is
     returned and this format type is not used.
     """
 
     format_type = Unicode('text/plain')
 
     enabled = Bool(True, config=True)
 
     print_method = ObjectName('__repr__')
 
     # The singleton printers.
     # Maps the IDs of the builtin singleton objects to the format functions.
     singleton_printers = Dict(config=True)
     def _singleton_printers_default(self):
         return {}
 
     # The type-specific printers.
     # Map type objects to the format functions.
     type_printers = Dict(config=True)
     def _type_printers_default(self):
         return {}
 
     # The deferred-import type-specific printers.
     # Map (modulename, classname) pairs to the format functions.
     deferred_printers = Dict(config=True)
     def _deferred_printers_default(self):
         return {}
 
     def __call__(self, obj):
         """Compute the format for an object."""
         if self.enabled:
             obj_id = id(obj)
             try:
                 obj_class = getattr(obj, '__class__', None) or type(obj)
                 # First try to find registered singleton printers for the type.
                 try:
                     printer = self.singleton_printers[obj_id]
                 except (TypeError, KeyError):
                     pass
                 else:
                     return printer(obj)
                 # Next look for type_printers.
                 for cls in pretty._get_mro(obj_class):
                     if cls in self.type_printers:
                         return self.type_printers[cls](obj)
                     else:
                         printer = self._in_deferred_types(cls)
                         if printer is not None:
                             return printer(obj)
                 # Finally look for special method names.
                 if hasattr(obj_class, self.print_method):
                     printer = getattr(obj_class, self.print_method)
                     return printer(obj)
                 return None
             except Exception:
                 pass
         else:
             return None
 
     def for_type(self, typ, func):
         """Add a format function for a given type.
 
         Parameters
         -----------
         typ : class
             The class of the object that will be formatted using `func`.
         func : callable
             The callable that will be called to compute the format data. The
             call signature of this function is simple, it must take the
             object to be formatted and return the raw data for the given
             format. Subclasses may use a different call signature for the
             `func` argument.
         """
         oldfunc = self.type_printers.get(typ, None)
         if func is not None:
             # To support easy restoration of old printers, we need to ignore
             # Nones.
             self.type_printers[typ] = func
         return oldfunc
 
     def for_type_by_name(self, type_module, type_name, func):
         """Add a format function for a type specified by the full dotted
         module and name of the type, rather than the type of the object.
 
         Parameters
         ----------
         type_module : str
             The full dotted name of the module the type is defined in, like
             ``numpy``.
         type_name : str
             The name of the type (the class name), like ``dtype``
         func : callable
             The callable that will be called to compute the format data. The
             call signature of this function is simple, it must take the
             object to be formatted and return the raw data for the given
             format. Subclasses may use a different call signature for the
             `func` argument.
         """
         key = (type_module, type_name)
         oldfunc = self.deferred_printers.get(key, None)
         if func is not None:
             # To support easy restoration of old printers, we need to ignore
             # Nones.
             self.deferred_printers[key] = func
         return oldfunc
 
     def _in_deferred_types(self, cls):
         """
         Check if the given class is specified in the deferred type registry.
 
         Returns the printer from the registry if it exists, and None if the
         class is not in the registry. Successful matches will be moved to the
         regular type registry for future use.
         """
         mod = getattr(cls, '__module__', None)
         name = getattr(cls, '__name__', None)
         key = (mod, name)
         printer = None
         if key in self.deferred_printers:
             # Move the printer over to the regular registry.
             printer = self.deferred_printers.pop(key)
             self.type_printers[cls] = printer
         return printer
 
 
 class PlainTextFormatter(BaseFormatter):
     """The default pretty-printer.
 
     This uses :mod:`IPython.lib.pretty` to compute the format data of
     the object. If the object cannot be pretty printed, :func:`repr` is used.
     See the documentation of :mod:`IPython.lib.pretty` for details on
     how to write pretty printers.  Here is a simple example::
 
         def dtype_pprinter(obj, p, cycle):
             if cycle:
                 return p.text('dtype(...)')
             if hasattr(obj, 'fields'):
                 if obj.fields is None:
                     p.text(repr(obj))
                 else:
                     p.begin_group(7, 'dtype([')
                     for i, field in enumerate(obj.descr):
                         if i > 0:
                             p.text(',')
                             p.breakable()
                         p.pretty(field)
                     p.end_group(7, '])')
     """
 
     # The format type of data returned.
     format_type = Unicode('text/plain')
 
     # This subclass ignores this attribute as it always need to return
     # something.
     enabled = Bool(True, config=False)
 
     # Look for a _repr_pretty_ methods to use for pretty printing.
     print_method = ObjectName('_repr_pretty_')
 
     # Whether to pretty-print or not.
     pprint = Bool(True, config=True)
 
     # Whether to be verbose or not.
     verbose = Bool(False, config=True)
 
     # The maximum width.
     max_width = Integer(79, config=True)
 
     # The newline character.
     newline = Unicode('\n', config=True)
 
     # format-string for pprinting floats
     float_format = Unicode('%r')
     # setter for float precision, either int or direct format-string
     float_precision = CUnicode('', config=True)
 
     def _float_precision_changed(self, name, old, new):
         """float_precision changed, set float_format accordingly.
 
         float_precision can be set by int or str.
         This will set float_format, after interpreting input.
         If numpy has been imported, numpy print precision will also be set.
 
         integer `n` sets format to '%.nf', otherwise, format set directly.
 
         An empty string returns to defaults (repr for float, 8 for numpy).
 
         This parameter can be set via the '%precision' magic.
         """
 
         if '%' in new:
             # got explicit format string
             fmt = new
             try:
                 fmt%3.14159
             except Exception:
                 raise ValueError("Precision must be int or format string, not %r"%new)
         elif new:
             # otherwise, should be an int
             try:
                 i = int(new)
                 assert i >= 0
             except ValueError:
                 raise ValueError("Precision must be int or format string, not %r"%new)
             except AssertionError:
                 raise ValueError("int precision must be non-negative, not %r"%i)
 
             fmt = '%%.%if'%i
             if 'numpy' in sys.modules:
                 # set numpy precision if it has been imported
                 import numpy
                 numpy.set_printoptions(precision=i)
         else:
             # default back to repr
             fmt = '%r'
             if 'numpy' in sys.modules:
                 import numpy
                 # numpy default is 8
                 numpy.set_printoptions(precision=8)
         self.float_format = fmt
 
     # Use the default pretty printers from IPython.lib.pretty.
     def _singleton_printers_default(self):
         return pretty._singleton_pprinters.copy()
 
     def _type_printers_default(self):
         d = pretty._type_pprinters.copy()
         d[float] = lambda obj,p,cycle: p.text(self.float_format%obj)
         return d
 
     def _deferred_printers_default(self):
         return pretty._deferred_type_pprinters.copy()
 
     #### FormatterABC interface ####
 
     def __call__(self, obj):
         """Compute the pretty representation of the object."""
         if not self.pprint:
-            try:
-                return repr(obj)
-            except TypeError:
-                return ''
+            return pretty._safe_repr(obj)
         else:
             # This uses use StringIO, as cStringIO doesn't handle unicode.
             stream = StringIO()
             # self.newline.encode() is a quick fix for issue gh-597. We need to
             # ensure that stream does not get a mix of unicode and bytestrings,
             # or it will cause trouble.
             printer = pretty.RepresentationPrinter(stream, self.verbose,
                 self.max_width, unicode_to_str(self.newline),
                 singleton_pprinters=self.singleton_printers,
                 type_pprinters=self.type_printers,
                 deferred_pprinters=self.deferred_printers)
             printer.pretty(obj)
             printer.flush()
             return stream.getvalue()
 
 
 class HTMLFormatter(BaseFormatter):
     """An HTML formatter.
 
     To define the callables that compute the HTML representation of your
     objects, define a :meth:`_repr_html_` method or use the :meth:`for_type`
     or :meth:`for_type_by_name` methods to register functions that handle
     this.
 
     The return value of this formatter should be a valid HTML snippet that
     could be injected into an existing DOM. It should *not* include the
     ```<html>`` or ```<body>`` tags.
     """
     format_type = Unicode('text/html')
 
     print_method = ObjectName('_repr_html_')
 
 
 class SVGFormatter(BaseFormatter):
     """An SVG formatter.
 
     To define the callables that compute the SVG representation of your
     objects, define a :meth:`_repr_svg_` method or use the :meth:`for_type`
     or :meth:`for_type_by_name` methods to register functions that handle
     this.
 
     The return value of this formatter should be valid SVG enclosed in
     ```<svg>``` tags, that could be injected into an existing DOM. It should
     *not* include the ```<html>`` or ```<body>`` tags.
     """
     format_type = Unicode('image/svg+xml')
 
     print_method = ObjectName('_repr_svg_')
 
 
 class PNGFormatter(BaseFormatter):
     """A PNG formatter.
 
     To define the callables that compute the PNG representation of your
     objects, define a :meth:`_repr_png_` method or use the :meth:`for_type`
     or :meth:`for_type_by_name` methods to register functions that handle
     this.
 
     The return value of this formatter should be raw PNG data, *not*
     base64 encoded.
     """
     format_type = Unicode('image/png')
 
     print_method = ObjectName('_repr_png_')
 
 
 class JPEGFormatter(BaseFormatter):
     """A JPEG formatter.
 
     To define the callables that compute the JPEG representation of your
     objects, define a :meth:`_repr_jpeg_` method or use the :meth:`for_type`
     or :meth:`for_type_by_name` methods to register functions that handle
     this.
 
     The return value of this formatter should be raw JPEG data, *not*
     base64 encoded.
     """
     format_type = Unicode('image/jpeg')
 
     print_method = ObjectName('_repr_jpeg_')
 
 
 class LatexFormatter(BaseFormatter):
     """A LaTeX formatter.
 
     To define the callables that compute the LaTeX representation of your
     objects, define a :meth:`_repr_latex_` method or use the :meth:`for_type`
     or :meth:`for_type_by_name` methods to register functions that handle
     this.
 
     The return value of this formatter should be a valid LaTeX equation,
     enclosed in either ```$```, ```$$``` or another LaTeX equation
     environment.
     """
     format_type = Unicode('text/latex')
 
     print_method = ObjectName('_repr_latex_')
 
 
 class JSONFormatter(BaseFormatter):
     """A JSON string formatter.
 
     To define the callables that compute the JSON string representation of
     your objects, define a :meth:`_repr_json_` method or use the :meth:`for_type`
     or :meth:`for_type_by_name` methods to register functions that handle
     this.
 
     The return value of this formatter should be a valid JSON string.
     """
     format_type = Unicode('application/json')
 
     print_method = ObjectName('_repr_json_')
 
 
 class JavascriptFormatter(BaseFormatter):
     """A Javascript formatter.
 
     To define the callables that compute the Javascript representation of
     your objects, define a :meth:`_repr_javascript_` method or use the
     :meth:`for_type` or :meth:`for_type_by_name` methods to register functions
     that handle this.
 
     The return value of this formatter should be valid Javascript code and
     should *not* be enclosed in ```<script>``` tags.
     """
     format_type = Unicode('application/javascript')
 
     print_method = ObjectName('_repr_javascript_')
 
 FormatterABC.register(BaseFormatter)
 FormatterABC.register(PlainTextFormatter)
 FormatterABC.register(HTMLFormatter)
 FormatterABC.register(SVGFormatter)
 FormatterABC.register(PNGFormatter)
 FormatterABC.register(JPEGFormatter)
 FormatterABC.register(LatexFormatter)
 FormatterABC.register(JSONFormatter)
 FormatterABC.register(JavascriptFormatter)
 
 
 def format_display_data(obj, include=None, exclude=None):
     """Return a format data dict for an object.
 
     By default all format types will be computed.
 
     The following MIME types are currently implemented:
 
     * text/plain
     * text/html
     * text/latex
     * application/json
     * application/javascript
     * image/png
     * image/jpeg
     * image/svg+xml
 
     Parameters
     ----------
     obj : object
         The Python object whose format data will be computed.
 
     Returns
     -------
     format_dict : dict
         A dictionary of key/value pairs, one or each format that was
         generated for the object. The keys are the format types, which
         will usually be MIME type strings and the values and JSON'able
         data structure containing the raw data for the representation in
         that format.
     include : list or tuple, optional
         A list of format type strings (MIME types) to include in the
         format data dict. If this is set *only* the format types included
         in this list will be computed.
     exclude : list or tuple, optional
         A list of format type string (MIME types) to exclue in the format
         data dict. If this is set all format types will be computed,
         except for those included in this argument.
     """
     from IPython.core.interactiveshell import InteractiveShell
 
     InteractiveShell.instance().display_formatter.format(
         obj,
         include,
         exclude
     )
 

IPython/lib/pretty.py

@@ -1,797 +1,851 b''
 # -*- coding: utf-8 -*-
 """
 Python advanced pretty printer.  This pretty printer is intended to
 replace the old `pprint` python module which does not allow developers
 to provide their own pretty print callbacks.
 
 This module is based on ruby's `prettyprint.rb` library by `Tanaka Akira`.
 
 
 Example Usage
 -------------
 
 To directly print the representation of an object use `pprint`::
 
     from pretty import pprint
     pprint(complex_object)
 
 To get a string of the output use `pretty`::
 
     from pretty import pretty
     string = pretty(complex_object)
 
 
 Extending
 ---------
 
 The pretty library allows developers to add pretty printing rules for their
 own objects.  This process is straightforward.  All you have to do is to
 add a `_repr_pretty_` method to your object and call the methods on the
 pretty printer passed::
 
     class MyObject(object):
 
         def _repr_pretty_(self, p, cycle):
             ...
 
 Depending on the python version you want to support you have two
 possibilities.  The following list shows the python 2.5 version and the
 compatibility one.
 
 
 Here the example implementation of a `_repr_pretty_` method for a list
 subclass for python 2.5 and higher (python 2.5 requires the with statement
 __future__ import)::
 
     class MyList(list):
 
         def _repr_pretty_(self, p, cycle):
             if cycle:
                 p.text('MyList(...)')
             else:
                 with p.group(8, 'MyList([', '])'):
                     for idx, item in enumerate(self):
                         if idx:
                             p.text(',')
                             p.breakable()
                         p.pretty(item)
 
 The `cycle` parameter is `True` if pretty detected a cycle.  You *have* to
 react to that or the result is an infinite loop.  `p.text()` just adds
 non breaking text to the output, `p.breakable()` either adds a whitespace
 or breaks here.  If you pass it an argument it's used instead of the
 default space.  `p.pretty` prettyprints another object using the pretty print
 method.
 
 The first parameter to the `group` function specifies the extra indentation
 of the next line.  In this example the next item will either be not
 breaked (if the items are short enough) or aligned with the right edge of
 the opening bracked of `MyList`.
 
 If you want to support python 2.4 and lower you can use this code::
 
     class MyList(list):
 
         def _repr_pretty_(self, p, cycle):
             if cycle:
                 p.text('MyList(...)')
             else:
                 p.begin_group(8, 'MyList([')
                 for idx, item in enumerate(self):
                     if idx:
                         p.text(',')
                         p.breakable()
                     p.pretty(item)
                 p.end_group(8, '])')
 
 If you just want to indent something you can use the group function
 without open / close parameters.  Under python 2.5 you can also use this
 code::
 
     with p.indent(2):
         ...
 
 Or under python2.4 you might want to modify ``p.indentation`` by hand but
 this is rather ugly.
 
 Inheritance diagram:
 
 .. inheritance-diagram:: IPython.lib.pretty
    :parts: 3
 
 :copyright: 2007 by Armin Ronacher.
             Portions (c) 2009 by Robert Kern.
 :license: BSD License.
 """
 from __future__ import print_function
 from contextlib import contextmanager
 import sys
 import types
 import re
 import datetime
 from collections import deque
 
 from IPython.utils.py3compat import PY3
 
 if PY3:
     from io import StringIO
 else:
     from StringIO import StringIO
 
 
 __all__ = ['pretty', 'pprint', 'PrettyPrinter', 'RepresentationPrinter',
     'for_type', 'for_type_by_name']
 
 
 _re_pattern_type = type(re.compile(''))
 
+def _failed_repr(obj, e):
+    """Render a failed repr, including the exception.
+    
+    Tries to get exception and type info
+    """
+    # get exception name
+    if e.__class__.__module__ in ('exceptions', 'builtins'):
+        ename = e.__class__.__name__
+    else:
+        ename = '{}.{}'.format(
+            e.__class__.__module__,
+            e.__class__.__name__,
+        )
+    # and exception string, which sometimes fails
+    # (usually due to unicode error message)
+    try:
+        estr = str(e)
+    except Exception:
+        estr = "unknown"
+    
+    # and class name
+    try:
+        klass = _safe_getattr(obj, '__class__', None) or type(obj)
+        mod = _safe_getattr(klass, '__module__', None)
+        if mod in (None, '__builtin__', 'builtins', 'exceptions'):
+            classname = klass.__name__
+        else:
+            classname = mod + '.' + klass.__name__
+    except Exception:
+        # this may be paranoid, but we already know repr is broken
+        classname = "unknown type"
+    
+    # the informative repr
+    return "<repr(<{} at 0x{:x}>) failed: {}: {}>".format(
+        classname, id(obj), ename, estr,
+    )
+
+def _safe_repr(obj):
+    """Don't assume repr is not broken."""
+    try:
+        return repr(obj)
+    except Exception as e:
+        return _failed_repr(obj, e)
+
+def _safe_getattr(obj, attr, default=None):
+    """Safe version of getattr.
+    
+    Same as getattr, but will return ``default`` on any Exception,
+    rather than raising.
+    """
+    try:
+        return getattr(obj, attr, default)
+    except Exception:
+        return default
 
 def pretty(obj, verbose=False, max_width=79, newline='\n'):
     """
     Pretty print the object's representation.
     """
     stream = StringIO()
     printer = RepresentationPrinter(stream, verbose, max_width, newline)
     printer.pretty(obj)
     printer.flush()
     return stream.getvalue()
 
 
 def pprint(obj, verbose=False, max_width=79, newline='\n'):
     """
     Like `pretty` but print to stdout.
     """
     printer = RepresentationPrinter(sys.stdout, verbose, max_width, newline)
     printer.pretty(obj)
     printer.flush()
     sys.stdout.write(newline)
     sys.stdout.flush()
 
 class _PrettyPrinterBase(object):
 
     @contextmanager
     def indent(self, indent):
         """with statement support for indenting/dedenting."""
         self.indentation += indent
         try:
             yield
         finally:
             self.indentation -= indent
 
     @contextmanager
     def group(self, indent=0, open='', close=''):
         """like begin_group / end_group but for the with statement."""
         self.begin_group(indent, open)
         try:
             yield
         finally:
             self.end_group(indent, close)
 
 class PrettyPrinter(_PrettyPrinterBase):
     """
     Baseclass for the `RepresentationPrinter` prettyprinter that is used to
     generate pretty reprs of objects.  Contrary to the `RepresentationPrinter`
     this printer knows nothing about the default pprinters or the `_repr_pretty_`
     callback method.
     """
 
     def __init__(self, output, max_width=79, newline='\n'):
         self.output = output
         self.max_width = max_width
         self.newline = newline
         self.output_width = 0
         self.buffer_width = 0
         self.buffer = deque()
 
         root_group = Group(0)
         self.group_stack = [root_group]
         self.group_queue = GroupQueue(root_group)
         self.indentation = 0
 
     def _break_outer_groups(self):
         while self.max_width < self.output_width + self.buffer_width:
             group = self.group_queue.deq()
             if not group:
                 return
             while group.breakables:
                 x = self.buffer.popleft()
                 self.output_width = x.output(self.output, self.output_width)
                 self.buffer_width -= x.width
             while self.buffer and isinstance(self.buffer[0], Text):
                 x = self.buffer.popleft()
                 self.output_width = x.output(self.output, self.output_width)
                 self.buffer_width -= x.width
 
     def text(self, obj):
         """Add literal text to the output."""
         width = len(obj)
         if self.buffer:
             text = self.buffer[-1]
             if not isinstance(text, Text):
                 text = Text()
                 self.buffer.append(text)
             text.add(obj, width)
             self.buffer_width += width
             self._break_outer_groups()
         else:
             self.output.write(obj)
             self.output_width += width
 
     def breakable(self, sep=' '):
         """
         Add a breakable separator to the output.  This does not mean that it
         will automatically break here.  If no breaking on this position takes
         place the `sep` is inserted which default to one space.
         """
         width = len(sep)
         group = self.group_stack[-1]
         if group.want_break:
             self.flush()
             self.output.write(self.newline)
             self.output.write(' ' * self.indentation)
             self.output_width = self.indentation
             self.buffer_width = 0
         else:
             self.buffer.append(Breakable(sep, width, self))
             self.buffer_width += width
             self._break_outer_groups()
             
     def break_(self):
         """
         Explicitly insert a newline into the output, maintaining correct indentation.
         """
         self.flush()
         self.output.write(self.newline)
         self.output.write(' ' * self.indentation)
         self.output_width = self.indentation
         self.buffer_width = 0
         
 
     def begin_group(self, indent=0, open=''):
         """
         Begin a group.  If you want support for python < 2.5 which doesn't has
         the with statement this is the preferred way:
 
             p.begin_group(1, '{')
             ...
             p.end_group(1, '}')
 
         The python 2.5 expression would be this:
 
             with p.group(1, '{', '}'):
                 ...
 
         The first parameter specifies the indentation for the next line (usually
         the width of the opening text), the second the opening text.  All
         parameters are optional.
         """
         if open:
             self.text(open)
         group = Group(self.group_stack[-1].depth + 1)
         self.group_stack.append(group)
         self.group_queue.enq(group)
         self.indentation += indent
 
     def end_group(self, dedent=0, close=''):
         """End a group. See `begin_group` for more details."""
         self.indentation -= dedent
         group = self.group_stack.pop()
         if not group.breakables:
             self.group_queue.remove(group)
         if close:
             self.text(close)
 
     def flush(self):
         """Flush data that is left in the buffer."""
         for data in self.buffer:
             self.output_width += data.output(self.output, self.output_width)
         self.buffer.clear()
         self.buffer_width = 0
 
 
 def _get_mro(obj_class):
     """ Get a reasonable method resolution order of a class and its superclasses
     for both old-style and new-style classes.
     """
     if not hasattr(obj_class, '__mro__'):
         # Old-style class. Mix in object to make a fake new-style class.
         try:
             obj_class = type(obj_class.__name__, (obj_class, object), {})
         except TypeError:
             # Old-style extension type that does not descend from object.
             # FIXME: try to construct a more thorough MRO.
             mro = [obj_class]
         else:
             mro = obj_class.__mro__[1:-1]
     else:
         mro = obj_class.__mro__
     return mro
 
 
 class RepresentationPrinter(PrettyPrinter):
     """
     Special pretty printer that has a `pretty` method that calls the pretty
     printer for a python object.
 
     This class stores processing data on `self` so you must *never* use
     this class in a threaded environment.  Always lock it or reinstanciate
     it.
 
     Instances also have a verbose flag callbacks can access to control their
     output.  For example the default instance repr prints all attributes and
     methods that are not prefixed by an underscore if the printer is in
     verbose mode.
     """
 
     def __init__(self, output, verbose=False, max_width=79, newline='\n',
         singleton_pprinters=None, type_pprinters=None, deferred_pprinters=None):
 
         PrettyPrinter.__init__(self, output, max_width, newline)
         self.verbose = verbose
         self.stack = []
         if singleton_pprinters is None:
             singleton_pprinters = _singleton_pprinters.copy()
         self.singleton_pprinters = singleton_pprinters
         if type_pprinters is None:
             type_pprinters = _type_pprinters.copy()
         self.type_pprinters = type_pprinters
         if deferred_pprinters is None:
             deferred_pprinters = _deferred_type_pprinters.copy()
         self.deferred_pprinters = deferred_pprinters
 
     def pretty(self, obj):
         """Pretty print the given object."""
         obj_id = id(obj)
         cycle = obj_id in self.stack
         self.stack.append(obj_id)
         self.begin_group()
         try:
-            obj_class = getattr(obj, '__class__', None) or type(obj)
+            obj_class = _safe_getattr(obj, '__class__', None) or type(obj)
             # First try to find registered singleton printers for the type.
             try:
                 printer = self.singleton_pprinters[obj_id]
             except (TypeError, KeyError):
                 pass
             else:
                 return printer(obj, self, cycle)
             # Next walk the mro and check for either:
             #   1) a registered printer
             #   2) a _repr_pretty_ method
             for cls in _get_mro(obj_class):
                 if cls in self.type_pprinters:
                     # printer registered in self.type_pprinters
                     return self.type_pprinters[cls](obj, self, cycle)
                 else:
                     # deferred printer
                     printer = self._in_deferred_types(cls)
                     if printer is not None:
                         return printer(obj, self, cycle)
                     else:
                         # Finally look for special method names.
                         # Some objects automatically create any requested
                         # attribute. Try to ignore most of them by checking for
                         # callability.
                         if '_repr_pretty_' in cls.__dict__:
                             meth = cls._repr_pretty_
                             if callable(meth):
                                 return meth(obj, self, cycle)
             return _default_pprint(obj, self, cycle)
         finally:
             self.end_group()
             self.stack.pop()
 
     def _in_deferred_types(self, cls):
         """
         Check if the given class is specified in the deferred type registry.
 
         Returns the printer from the registry if it exists, and None if the
         class is not in the registry. Successful matches will be moved to the
         regular type registry for future use.
         """
-        mod = getattr(cls, '__module__', None)
-        name = getattr(cls, '__name__', None)
+        mod = _safe_getattr(cls, '__module__', None)
+        name = _safe_getattr(cls, '__name__', None)
         key = (mod, name)
         printer = None
         if key in self.deferred_pprinters:
             # Move the printer over to the regular registry.
             printer = self.deferred_pprinters.pop(key)
             self.type_pprinters[cls] = printer
         return printer
 
 
 class Printable(object):
 
     def output(self, stream, output_width):
         return output_width
 
 
 class Text(Printable):
 
     def __init__(self):
         self.objs = []
         self.width = 0
 
     def output(self, stream, output_width):
         for obj in self.objs:
             stream.write(obj)
         return output_width + self.width
 
     def add(self, obj, width):
         self.objs.append(obj)
         self.width += width
 
 
 class Breakable(Printable):
 
     def __init__(self, seq, width, pretty):
         self.obj = seq
         self.width = width
         self.pretty = pretty
         self.indentation = pretty.indentation
         self.group = pretty.group_stack[-1]
         self.group.breakables.append(self)
 
     def output(self, stream, output_width):
         self.group.breakables.popleft()
         if self.group.want_break:
             stream.write(self.pretty.newline)
             stream.write(' ' * self.indentation)
             return self.indentation
         if not self.group.breakables:
             self.pretty.group_queue.remove(self.group)
         stream.write(self.obj)
         return output_width + self.width
 
 
 class Group(Printable):
 
     def __init__(self, depth):
         self.depth = depth
         self.breakables = deque()
         self.want_break = False
 
 
 class GroupQueue(object):
 
     def __init__(self, *groups):
         self.queue = []
         for group in groups:
             self.enq(group)
 
     def enq(self, group):
         depth = group.depth
         while depth > len(self.queue) - 1:
             self.queue.append([])
         self.queue[depth].append(group)
 
     def deq(self):
         for stack in self.queue:
             for idx, group in enumerate(reversed(stack)):
                 if group.breakables:
                     del stack[idx]
                     group.want_break = True
                     return group
             for group in stack:
                 group.want_break = True
             del stack[:]
 
     def remove(self, group):
         try:
             self.queue[group.depth].remove(group)
         except ValueError:
             pass
 
 try:
     _baseclass_reprs = (object.__repr__, types.InstanceType.__repr__)
 except AttributeError:  # Python 3
     _baseclass_reprs = (object.__repr__,)
 
 
 def _default_pprint(obj, p, cycle):
     """
     The default print function.  Used if an object does not provide one and
     it's none of the builtin objects.
     """
-    klass = getattr(obj, '__class__', None) or type(obj)
-    if getattr(klass, '__repr__', None) not in _baseclass_reprs:
+    klass = _safe_getattr(obj, '__class__', None) or type(obj)
+    if _safe_getattr(klass, '__repr__', None) not in _baseclass_reprs:
         # A user-provided repr. Find newlines and replace them with p.break_()
-        output = repr(obj)
+        output = _safe_repr(obj)
         for idx,output_line in enumerate(output.splitlines()):
             if idx:
                 p.break_()
             p.text(output_line)
         return
     p.begin_group(1, '<')
     p.pretty(klass)
     p.text(' at 0x%x' % id(obj))
     if cycle:
         p.text(' ...')
     elif p.verbose:
         first = True
         for key in dir(obj):
             if not key.startswith('_'):
                 try:
                     value = getattr(obj, key)
                 except AttributeError:
                     continue
                 if isinstance(value, types.MethodType):
                     continue
                 if not first:
                     p.text(',')
                 p.breakable()
                 p.text(key)
                 p.text('=')
                 step = len(key) + 1
                 p.indentation += step
                 p.pretty(value)
                 p.indentation -= step
                 first = False
     p.end_group(1, '>')
 
 
 def _seq_pprinter_factory(start, end, basetype):
     """
     Factory that returns a pprint function useful for sequences.  Used by
     the default pprint for tuples, dicts, and lists.
     """
     def inner(obj, p, cycle):
         typ = type(obj)
         if basetype is not None and typ is not basetype and typ.__repr__ != basetype.__repr__:
             # If the subclass provides its own repr, use it instead.
             return p.text(typ.__repr__(obj))
 
         if cycle:
             return p.text(start + '...' + end)
         step = len(start)
         p.begin_group(step, start)
         for idx, x in enumerate(obj):
             if idx:
                 p.text(',')
                 p.breakable()
             p.pretty(x)
         if len(obj) == 1 and type(obj) is tuple:
             # Special case for 1-item tuples.
             p.text(',')
         p.end_group(step, end)
     return inner
 
 
 def _set_pprinter_factory(start, end, basetype):
     """
     Factory that returns a pprint function useful for sets and frozensets.
     """
     def inner(obj, p, cycle):
         typ = type(obj)
         if basetype is not None and typ is not basetype and typ.__repr__ != basetype.__repr__:
             # If the subclass provides its own repr, use it instead.
             return p.text(typ.__repr__(obj))
 
         if cycle:
             return p.text(start + '...' + end)
         if len(obj) == 0:
             # Special case.
             p.text(basetype.__name__ + '()')
         else:
             step = len(start)
             p.begin_group(step, start)
             # Like dictionary keys, we will try to sort the items.
             items = list(obj)
             try:
                 items.sort()
             except Exception:
                 # Sometimes the items don't sort.
                 pass
             for idx, x in enumerate(items):
                 if idx:
                     p.text(',')
                     p.breakable()
                 p.pretty(x)
             p.end_group(step, end)
     return inner
 
 
 def _dict_pprinter_factory(start, end, basetype=None):
     """
     Factory that returns a pprint function used by the default pprint of
     dicts and dict proxies.
     """
     def inner(obj, p, cycle):
         typ = type(obj)
         if basetype is not None and typ is not basetype and typ.__repr__ != basetype.__repr__:
             # If the subclass provides its own repr, use it instead.
             return p.text(typ.__repr__(obj))
 
         if cycle:
             return p.text('{...}')
         p.begin_group(1, start)
         keys = obj.keys()
         try:
             keys.sort()
         except Exception as e:
             # Sometimes the keys don't sort.
             pass
         for idx, key in enumerate(keys):
             if idx:
                 p.text(',')
                 p.breakable()
             p.pretty(key)
             p.text(': ')
             p.pretty(obj[key])
         p.end_group(1, end)
     return inner
 
 
 def _super_pprint(obj, p, cycle):
     """The pprint for the super type."""
     p.begin_group(8, '<super: ')
     p.pretty(obj.__self_class__)
     p.text(',')
     p.breakable()
     p.pretty(obj.__self__)
     p.end_group(8, '>')
 
 
 def _re_pattern_pprint(obj, p, cycle):
     """The pprint function for regular expression patterns."""
     p.text('re.compile(')
     pattern = repr(obj.pattern)
     if pattern[:1] in 'uU':
         pattern = pattern[1:]
         prefix = 'ur'
     else:
         prefix = 'r'
     pattern = prefix + pattern.replace('\\\\', '\\')
     p.text(pattern)
     if obj.flags:
         p.text(',')
         p.breakable()
         done_one = False
         for flag in ('TEMPLATE', 'IGNORECASE', 'LOCALE', 'MULTILINE', 'DOTALL',
             'UNICODE', 'VERBOSE', 'DEBUG'):
             if obj.flags & getattr(re, flag):
                 if done_one:
                     p.text('|')
                 p.text('re.' + flag)
                 done_one = True
     p.text(')')
 
 
 def _type_pprint(obj, p, cycle):
     """The pprint for classes and types."""
-    mod = getattr(obj, '__module__', None)
+    mod = _safe_getattr(obj, '__module__', None)
     if mod is None:
         # Heap allocated types might not have the module attribute,
         # and others may set it to None.
         return p.text(obj.__name__)
 
     if mod in ('__builtin__', 'builtins', 'exceptions'):
         name = obj.__name__
     else:
         name = mod + '.' + obj.__name__
     p.text(name)
 
 
 def _repr_pprint(obj, p, cycle):
     """A pprint that just redirects to the normal repr function."""
-    p.text(repr(obj))
+    p.text(_safe_repr(obj))
 
 
 def _function_pprint(obj, p, cycle):
     """Base pprint for all functions and builtin functions."""
     if obj.__module__ in ('__builtin__', 'builtins', 'exceptions') or not obj.__module__:
         name = obj.__name__
     else:
         name = obj.__module__ + '.' + obj.__name__
     p.text('<function %s>' % name)
 
 
 def _exception_pprint(obj, p, cycle):
     """Base pprint for all exceptions."""
     if obj.__class__.__module__ in ('exceptions', 'builtins'):
         name = obj.__class__.__name__
     else:
         name = '%s.%s' % (
             obj.__class__.__module__,
             obj.__class__.__name__
         )
     step = len(name) + 1
     p.begin_group(step, name + '(')
     for idx, arg in enumerate(getattr(obj, 'args', ())):
         if idx:
             p.text(',')
             p.breakable()
         p.pretty(arg)
     p.end_group(step, ')')
 
 
 #: the exception base
 try:
     _exception_base = BaseException
 except NameError:
     _exception_base = Exception
 
 
 #: printers for builtin types
 _type_pprinters = {
     int:                        _repr_pprint,
     float:                      _repr_pprint,
     str:                        _repr_pprint,
     tuple:                      _seq_pprinter_factory('(', ')', tuple),
     list:                       _seq_pprinter_factory('[', ']', list),
     dict:                       _dict_pprinter_factory('{', '}', dict),
     
     set:                        _set_pprinter_factory('{', '}', set),
     frozenset:                  _set_pprinter_factory('frozenset({', '})', frozenset),
     super:                      _super_pprint,
     _re_pattern_type:           _re_pattern_pprint,
     type:                       _type_pprint,
     types.FunctionType:         _function_pprint,
     types.BuiltinFunctionType:  _function_pprint,
     types.MethodType:           _repr_pprint,
     
     datetime.datetime:          _repr_pprint,
     datetime.timedelta:         _repr_pprint,
     _exception_base:            _exception_pprint
 }
 
 try:
     _type_pprinters[types.DictProxyType] = _dict_pprinter_factory('<dictproxy {', '}>')
     _type_pprinters[types.ClassType] = _type_pprint
     _type_pprinters[types.SliceType] = _repr_pprint
 except AttributeError: # Python 3
     _type_pprinters[slice] = _repr_pprint
     
 try:
     _type_pprinters[xrange] = _repr_pprint
     _type_pprinters[long] = _repr_pprint
     _type_pprinters[unicode] = _repr_pprint
 except NameError:
     _type_pprinters[range] = _repr_pprint
     _type_pprinters[bytes] = _repr_pprint
 
 #: printers for types specified by name
 _deferred_type_pprinters = {
 }
 
 def for_type(typ, func):
     """
     Add a pretty printer for a given type.
     """
     oldfunc = _type_pprinters.get(typ, None)
     if func is not None:
         # To support easy restoration of old pprinters, we need to ignore Nones.
         _type_pprinters[typ] = func
     return oldfunc
 
 def for_type_by_name(type_module, type_name, func):
     """
     Add a pretty printer for a type specified by the module and name of a type
     rather than the type object itself.
     """
     key = (type_module, type_name)
     oldfunc = _deferred_type_pprinters.get(key, None)
     if func is not None:
         # To support easy restoration of old pprinters, we need to ignore Nones.
         _deferred_type_pprinters[key] = func
     return oldfunc
 
 
 #: printers for the default singletons
 _singleton_pprinters = dict.fromkeys(map(id, [None, True, False, Ellipsis,
                                       NotImplemented]), _repr_pprint)
 
 
 if __name__ == '__main__':
     from random import randrange
     class Foo(object):
         def __init__(self):
             self.foo = 1
             self.bar = re.compile(r'\s+')
             self.blub = dict.fromkeys(range(30), randrange(1, 40))
             self.hehe = 23424.234234
             self.list = ["blub", "blah", self]
 
         def get_foo(self):
             print("foo")
 
     pprint(Foo(), verbose=True)

IPython/lib/tests/test_pretty.py

@@ -1,153 +1,184 b''
 """Tests for IPython.lib.pretty.
 """
 #-----------------------------------------------------------------------------
 # Copyright (c) 2011, the IPython Development Team.
 #
 # Distributed under the terms of the Modified BSD License.
 #
 # The full license is in the file COPYING.txt, distributed with this software.
 #-----------------------------------------------------------------------------
 
 #-----------------------------------------------------------------------------
 # Imports
 #-----------------------------------------------------------------------------
 from __future__ import print_function
 
 # Third-party imports
 import nose.tools as nt
 
 # Our own imports
 from IPython.lib import pretty
 from IPython.testing.decorators import skip_without
 
 #-----------------------------------------------------------------------------
 # Classes and functions
 #-----------------------------------------------------------------------------
 
 class MyList(object):
     def __init__(self, content):
         self.content = content
     def _repr_pretty_(self, p, cycle):
         if cycle:
             p.text("MyList(...)")
         else:
             with p.group(3, "MyList(", ")"):
                 for (i, child) in enumerate(self.content):
                     if i:
                         p.text(",")
                         p.breakable()
                     else:
                         p.breakable("")
                     p.pretty(child)
 
 
 class MyDict(dict):
     def _repr_pretty_(self, p, cycle):
         p.text("MyDict(...)")
 
 
 class Dummy1(object):
     def _repr_pretty_(self, p, cycle):
         p.text("Dummy1(...)")
 
 class Dummy2(Dummy1):
     _repr_pretty_ = None
 
 class NoModule(object):
     pass
 
 NoModule.__module__ = None
 
 class Breaking(object):
     def _repr_pretty_(self, p, cycle):
         with p.group(4,"TG: ",":"):
             p.text("Breaking(")
             p.break_()
             p.text(")")
 
 class BreakingRepr(object):
     def __repr__(self):
         return "Breaking(\n)"
 
 class BreakingReprParent(object):
     def _repr_pretty_(self, p, cycle):
         with p.group(4,"TG: ",":"):
             p.pretty(BreakingRepr())
 
+class BadRepr(object):
+    
+    def __repr__(self):
+        return 1/0
 
 
 def test_indentation():
     """Test correct indentation in groups"""
     count = 40
     gotoutput = pretty.pretty(MyList(range(count)))
     expectedoutput = "MyList(\n" + ",\n".join("   %d" % i for i in range(count)) + ")"
 
     nt.assert_equal(gotoutput, expectedoutput)
 
 
 def test_dispatch():
     """
     Test correct dispatching: The _repr_pretty_ method for MyDict
     must be found before the registered printer for dict.
     """
     gotoutput = pretty.pretty(MyDict())
     expectedoutput = "MyDict(...)"
 
     nt.assert_equal(gotoutput, expectedoutput)
 
 
 def test_callability_checking():
     """
     Test that the _repr_pretty_ method is tested for callability and skipped if
     not.
     """
     gotoutput = pretty.pretty(Dummy2())
     expectedoutput = "Dummy1(...)"
 
     nt.assert_equal(gotoutput, expectedoutput)
 
 
 def test_sets():
     """
     Test that set and frozenset use Python 3 formatting.
     """
     objects = [set(), frozenset(), set([1]), frozenset([1]), set([1, 2]),
         frozenset([1, 2]), set([-1, -2, -3])]
     expected = ['set()', 'frozenset()', '{1}', 'frozenset({1})', '{1, 2}',
         'frozenset({1, 2})', '{-3, -2, -1}']
     for obj, expected_output in zip(objects, expected):
         got_output = pretty.pretty(obj)
         yield nt.assert_equal, got_output, expected_output
 
 
 @skip_without('xxlimited')
 def test_pprint_heap_allocated_type():
     """
     Test that pprint works for heap allocated types.
     """
     import xxlimited
     output = pretty.pretty(xxlimited.Null)
     nt.assert_equal(output, 'xxlimited.Null')
 
 def test_pprint_nomod():
     """
     Test that pprint works for classes with no __module__.
     """
     output = pretty.pretty(NoModule)
     nt.assert_equal(output, 'NoModule')
     
 def test_pprint_break():
     """
     Test that p.break_ produces expected output
     """
     output = pretty.pretty(Breaking())
     expected = "TG: Breaking(\n    ):"
     nt.assert_equal(output, expected)
 
 def test_pprint_break_repr():
     """
     Test that p.break_ is used in repr
     """
     output = pretty.pretty(BreakingReprParent())
     expected = "TG: Breaking(\n    ):"
-    nt.assert_equal(output, expected)
  No newline at end of file
+    nt.assert_equal(output, expected)
+
+def test_bad_repr():
+    """Don't raise, even when repr fails"""
+    output = pretty.pretty(BadRepr())
+    nt.assert_in("failed", output)
+    nt.assert_in("at 0x", output)
+    nt.assert_in("test_pretty", output)
+
+class BadException(Exception):
+    def __str__(self):
+        return -1
+
+class ReallyBadRepr(object):
+    __module__ = 1
+    @property
+    def __class__(self):
+        raise ValueError("I am horrible")
+    
+    def __repr__(self):
+        raise BadException()
+
+def test_really_bad_repr():
+    output = pretty.pretty(ReallyBadRepr())
+    nt.assert_in("failed", output)
+    nt.assert_in("BadException: unknown", output)
+    nt.assert_in("unknown type", output)
+    
  No newline at end of file