Have .add(value) return the index for value

[riggs]

It would be very useful for the .add methods on both the SortedList & SortedSet classes to return the new index at which the added item exists in the logical array.

[grantjenks]

Can you describe your use case in more detail?

The .add method doesn't compute the new index at which the item was added. Adding this functionality would decrease the performance of the current method.

Returning that index would be pretty similar to:

def add(self, val):
    self._original_add(val)
    return self.index(val)

You might simply monkey-patch the add method yourself to suit your needs. I could also imagine others wanting the added value returned or the list/set itself.

[riggs]

The use case is timeseries-sorted data. I have multiple data streams, each
containing multiple data points, and I want to combine them simply without
storing the timestamp with each piece of data.

Also, looking at the code for your library, I don't think this would impact
performance. The call to bisect.insort is effectively a call to
bisect.bisect followed by list.insert. All that would be required would be
making these two calls directly rather than via insort and doing a bit of
math to adjust the index for which 'bin' it's in.

[grantjenks]

This seems like a pretty ideal case for subclassing and getting the behavior you want:

from sortedcontainers import SortedList as OriginalSortedList

class SortedList(OriginalSortedList):
    def add(self, val):
        super(SortedList, self).add(val)
        return self.index(val)

The insort function is also about 20% faster than bisect followed by insert:

>>> from timeit import timeit
>>> def test(command, setup):
...     times = [timeit(command, setup=setup, number=1) for rpt in xrange(99)]
...     times.sort()
...     avg = sum(times) / len(times)
...     return 'median:', times[49], 'average:', avg
>>> test('insort(l, 42)', 'from bisect import insort; l = list(xrange(100))')
('median:', 9.5367431640625e-07, 'average:', 1.2715657552083333e-06)
>>> test('pos = bisect(l, 42); l.insert(pos, 42)', 'from bisect import bisect; l = list(xrange(100))')
('median:', 1.1920928955078125e-06, 'average:', 1.625581221147017e-06)

Also, in most cases it should be just "a bit of math to adjust the index" if the _index cache is valid. But in case it's not, iterating the lengths of the segments is required. This can be a slow process.