protect against broken repr in lib.pretty

[minrk]

for example:

from lxml import html
import urllib2
from IPython.display import display

term = 'cat'
data = html.parse("""http://en.wiktionary.org/w/index.php?title=%s&printable=yes""" % urllib2.quote(term.encode('utf-8')), parser=html.HTMLParser(encoding="utf-8"))
d = data.xpath(".//div[@id='mw-content-text']")[0]
display(d.getchildren())

[rkern]

I find it useful to include the type of the exception as well.

return "<repr(obj) failed: {0}: {1}>".format(type(e), e)

[rkern]

While I agree that reporting the exception message is useful, can we still at least include the usual <module.type at 0x.....> information? That will help locate the source of the problem.

[minrk]

type info / address added, along with some extra paranoia about bad objects

[rkern]

I like it. The docstring on _safe_getattr could be cleaned up, but otherwise LGTM.

[minrk]

cleaned up the docstring a bit, thanks.

[takluyver]

For the record, there are a couple of pathological cases which could still fail. Exceptions inheriting from BaseException directly won't be caught - that's kind of the design of the system, to allow KeyboardInterrupt and SystemExit to keep propagating when you'd catch any other exception. And on Python 2, old style classes from outside the exception hierarchy can still be raised. However, all of the exceptions Python itself uses are in the hierarchy, so that's only possible if something is called than does an explicit raise.

I don't think we need to worry about either of those cases, I just wanted to mention it in case anyone disagrees.

[takluyver]

Merging this.