There's a bit of history I don't understand and couldn't find the appropriate trail for.

The original error message from the 2.6 era:

$ python2.6 -c 0[0]
Traceback (most recent call last):
  File "<string>", line 1, in <module>
TypeError: 'int' object is unsubscriptable


This was changed in https://github.com/python/cpython/commit/f5fd5239819c5dfb1d7a33484be49dc705544d02 for clarity

As seen in 2.7.1

$ ./python2.7.1 -c 0[0]
Traceback (most recent call last):
  File "<string>", line 1, in <module>
TypeError: 'int' object is not subscriptable


It was later changed in the 2.7 era here: 
https://github.com/python/cpython/commit/7d1483cbadbe48620964348a2039690624e7dc8e


To this error message (as demonstrated with 2.7.12):

$ python2.7 -c 0[0]
Traceback (most recent call last):
  File "<string>", line 1, in <module>
TypeError: 'int' object has no attribute '__getitem__'


However, this patch never made it into the 3.x branch:

$ python3.6 -c 0[0]
Traceback (most recent call last):
  File "<string>", line 1, in <module>
TypeError: 'int' object is not subscriptable


Should this patch be (have been) cherry picked into master?

I'm not sure that this is an enhancement. It makes an error message inconsistent with other error messages.

>>> class I(int): pass
... 
>>> I()[0]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: 'I' object does not support indexing
>>> 0[0] = 0
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: 'int' object does not support item assignment
>>> del 0[0]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: 'int' object does not support item deletion
>>> ''['']                                                                                                                                                                                                                                   
Traceback (most recent call last):                                                                                                                                                                                                           
  File "<stdin>", line 1, in <module>                                                                                                                                                                                                        
TypeError: string indices must be integers, not str                                                                                                                                                                                          
>>> iter(0)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: 'int' object is not iterable
>>> reversed(0)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: argument to reversed() must be a sequence
>>> next(0)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: int object is not an iterator
>>> {}[[]]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: unhashable type: 'list'

And actually this message is misleading. An attribute '__getitem__' is not looked up on objects.

>>> class I(int): pass
... 
>>> x = I()
>>> x.__getitem__ = lambda *args: None
>>> I()[0:]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: 'I' object has no attribute '__getitem__'
>>> x.__getitem__
<function <lambda> at 0x7f11b62dd648>

All I'm really looking for is consistency -- should the change to 2.7 be reverted instead?

I think it should. But let to hear Raymond's arguments.

Shouldn't you wait for Raymond's arguments before outright closing the PR?

I made a new PR which instead *reverts* the python2.7 patch to restore consistency

I would like to leave this as-is.  Consistency between error message wording is one of our least important considerations.  The __getitem__ message is somewhat useful -- it helps learners know which magic method is being called and what they would need to add to a class to make it work.  This was the typical wording for old-style classes.  For example, calling len() on an old-style class without __len__ gave this message, "AttributeError: A instance has no attribute '__len__'".

In addition, we are getting very close to end-of-life for Python2.7.  Accordingly, any changes at this point should be limited to things that really matter.  It will be too easy to introduce a new problem that we won't have an opportunity to fix later.

I recommend closing this and shifting attention to things that matter.

I think the main concern here is ease in portability coupled with the incorrectness of the current message (pointed out in https://bugs.python.org/issue31550#msg302738)

For instance it was consistent in 2.7.1, but not later on in the 2.7.x tree.  *And* it's different than python3 (for no good reason?)

The message is misleading as well, it's not looking for __getitem__ on the instance but on the type.

> it's different than python3 (for no good reason?)
Python 3 has new style classes which were always different.

> main concern here is ease in portability  
We've long stated that there should never be dependency on the exact wording of an error message.  We're allowed to change that.

> For instance it was consistent in 2.7.1
To the extent we care about this, it is more important to be consistent with later releases than earlier releases.

Since I don't see any evidence that there is a real world problem, I would like to see this left alone.  At this point, in Python 2.7's life we need to have a strong aversion to any changes at all.  Not only has that ship sailed, its voyage is almost over.

If the reason of this change was making the error message more consistent with other error messages in classic classes (at the cost of larger dereference with Python 3), it LGTM.

Marked as closed for the reasons listed.

Thank you for the suggestion, but we're going to decline.