Add a PriorityQueue that works with unhashable items? about boltons HOT 4 CLOSED

object() < object() raises an exception in Python3, but this works and is fast:

class CmpFalse(object): __eq__ = __lt__ = __gt__ = lambda s,o: False
import heapq
h = []
heapq.heappush(h, (0, CmpFalse(), ['x']))
heapq.heappush(h, (0, CmpFalse(), {1:2}))
print(heapq.heappop(h))
print(heapq.heappop(h))

That's likely too simple to warrant adding to boltons.queueutils.

So, http://boltons.readthedocs.io/en/latest/queueutils.html should mention that task must be hashable and then this issue is finished.

from boltons.

Would float('nan') behave the same as your CmpFalse()?
Given that:

    >>> float('nan')==float('nan')
    False
    >>> float('nan')<float('nan')
    False
    >>> float('nan')>float('nan')
    False

from boltons.

heapq won't work either, because if I store tuples (-priority, item), I can get comparisons between dicts if priorities match.

Just a note: As a workaround, it's possible put something that will always compare to False in between:

(-priority, object(), item)

The overhead should be manageable in many cases; even unpacking a 3-tuple vs. a 2-tuple in a single assignment seems to take almost the same (very short) time.

from boltons.

Docs updated, but I would like to think some more about this. I'm not sure there's a way to update priority without a uniqueness constraint, thus one of the original features would be defeated. That said, it has come up in discussion that someone may need a priority queue for non-unique items, but I haven't had any need over the years :)

I'm definitely fine with adding more simple queues, as long as their use case is clear! :)

from boltons.