Python Cookbook

Sorting by Item or by Attribute

Credit: Matthew Wood

Problem

You need to sort a list of (x, y) coordinates by item y, or, more generally, sort a list of objects by any attribute or item.

Solution

You might first think of something like the following class, based on the simple but slow approach of passing a comparison function to the sort method:

class Sorter:
    # Notice how _ _compare is dependent on self._ _whichindex
    def _ _compare(self, x, y):
        return cmp(x[self._ _whichindex], y[self._ _whichindex])

    # Pass the sort function the _ _compare function defined above
    def _ _call_ _(self, data, whichindex = None):
        if whichindex is None :
            data.sort(  )
        else :
            self._ _whichindex = whichindex
            data.sort(self._ _compare)
        return data               # handy for inlining, and low-cost

The trick is to use a bound method that accesses instance state as the comparison function to determine which item (or attribute, but this version deals only with items) to fetch. Unfortunately, this makes the approach nonreentrant and not thread-safe.

Thanks to the faster, more robust, and more flexible DSU idiom, it’s not hard to make a more general version that allows attributes as well as items, guarantees a stable sort, is both reentrant and thread-safe, and is speedier to boot:

class Sorter: def _helper(self, data, aux, inplace): aux.sort( ) result = [data[i] for junk, i in aux] if inplace: data[:] = result return result def byItem(self, data, itemindex=None, inplace=1): if itemindex is None: if inplace: data.sort( ) result = ...