The description of __repr__ in section 3 of the Language Ref
says " This should normally look like a
valid Python expression that can be used to recreate an object with
the same value."  This isn't true, isn't a good idea, 
and often isn't possible anyway.  Rewrite this to emphasize 
that repr() is usually for debugging.

Assigning to GvR, since I assume the LangRef is his responsibility.
I can rewrite the text if the change's intent is approved.

Harrumph.  For starters it's true for strings, ints, longs and (as of 1.6) floats, plus lists, tuples and dicts recursively composed of these.  And it's a great idea.  I believe Guido meant what he wrote here:  "should" -- no bug.

I disagree with the "isn't a good idea" part.

While it's indeed not a good idea to use eval(repr(x)), it *is* a good idea to make repr(x) look like a syntactically correct expression that would recreate an object with the same value as x, given the appropriate environment (e.g. imported the class or factory function).

I hate non-standard object types whose repr() is indistinguishable from that of a similar standard object -- e.g. UserList makes this mistake, and xrange() used to pretend it was a tuple.

Nevertheless I'll try to think of something to add to the docs.

Checked something in as rev. 1.55.  Let me know what you think.

That's a bit better, so I won't re-open the bug on you.  (Noticed a small typo and fixed it.)