Reverted back to using a counter 'count' to keep track of precision.

Using our own counter, opposed to enumerates, allows us to skip increment::

 if d[ unit ] == 0: continue # skip 0's

The continue moves us back to the top of the loop, and which bypasses count += 1.  This allows us to have a datetime/timedelta that has a 0 value for a unit, and it will not effect precision fufillment.

For example, review this test::

 ONE_YEAR_FOUR_HOURS_DELTA = timedelta( 365, 14400, 0 )

 def test_zero_day_is_skipped_display_hour():
     _result = human( ONE_YEAR_FOUR_HOURS_DELTA, precision = 2 )
     assert 'year' in _result
     # day is 0 so it is skipped, so we should show hours ...
     assert 'hour' in _result

This test will fail using enumerate or using a slice on the units[:precision] tuple.

ago.py

@@ -25,11 +25,14 @@ def human(dt, precision=2, past_tense='{} ago', future_tense='in {}'):
 
     d = delta2dict( delta )
     hlist = [] 
+    count = 0
     units = ( 'year', 'day', 'hour', 'minute', 'second', 'microsecond' )
-    for unit in units[:precision]: # strip units to precision
+    for unit in units:
+        if count >= precision: break # met precision
         if d[ unit ] == 0: continue # skip 0's
         s = '' if d[ unit ] == 1 else 's' # handle plurals
         hlist.append( '%s %s%s' % ( d[unit], unit, s ) )
+        count += 1
     human_delta = ', '.join( hlist )
     return the_tense.format(human_delta)