Implemented as per https://mail.python.org/pipermail/python-dev/2016-April/144135.html

Attached are modifications to the patch based on feedback from Nikita. It produces a larger patch tho. The changes are to combine return branches in compiler_make_closure & to combine code between compiler_function & compiler_lambda into compiler_default_arguments

At first look your patch looks good (I'll make more detailed review later), but you should wait until we have finished all work related to the conversion to wordcode.

May've been best to wait on posting a patch, but long weekend yesterday made time available

mkfu3 updates mkfu2 with wordcode. Includes fix to EXTENDED_ARG documentation

In general the patch LGTM, besides few minor comments.

But new implementation uses longer bytecode for annotations. Current code packs argument names in one constant tuple:

$ ./python -m dis
def f(x: int, y: str, z: float): pass
  1           0 LOAD_NAME                0 (int)
              2 LOAD_NAME                1 (str)
              4 LOAD_NAME                2 (float)
              6 LOAD_CONST               0 (('x', 'y', 'z'))
              8 LOAD_CONST               1 (<code object f at 0xb6fc1640, file "<stdin>", line 1>)
             10 LOAD_CONST               2 ('f')
             12 EXTENDED_ARG             4
             14 EXTENDED_ARG          1024
             16 MAKE_FUNCTION        262144
             18 STORE_NAME               3 (f)
             20 LOAD_CONST               3 (None)
             22 RETURN_VALUE

New code uses LOAD_CONST for every name:

$ ./python -m dis
def f(x: int, y: str, z: float): pass
  1           0 LOAD_CONST               0 ('x')
              2 LOAD_NAME                0 (int)
              4 LOAD_CONST               1 ('y')
              6 LOAD_NAME                1 (str)
              8 LOAD_CONST               2 ('z')
             10 LOAD_NAME                2 (float)
             12 BUILD_MAP                3
             14 LOAD_CONST               3 (<code object f at 0xb7035a20, file "<stdin>", line 1>)
             16 LOAD_CONST               4 ('f')
             18 MAKE_FUNCTION            4
             20 STORE_NAME               3 (f)
             22 LOAD_CONST               5 (None)
             24 RETURN_VALUE

With new opcode that creates a dict from values and a tuple of keys (bpo-27140) new code would be only one instruction longer.

Now with adding BUILD_CONST_KEY_MAP I think MAKE_FUNCTION can be more compact.

mkfu4 implements bpo-27140. It doesn't special case 1-tuples into BUILD_MAP 1

It may be easier to have BUILD_CONST_KEY_MAP 1 peepholed if it's really preferable to strength reduce

I'm also noticing that it could've been suggested to go to the extreme with BUILD_CONST_KEY_MAP where instead of relying on peepholer to constant fold tuple, we implement constant folding const keys entirely there. Just a random thought, not a suggestion or anything else

Added more comments on Rietveld.

Kind of amusing how visit_argannoation logic went full circle. Makes sense considering pre-mkfu patch the ABI was quite similar on that front

Yes, this is why I wanted first push the patch with BUILD_CONST_KEY_MAP.

Regenerated for review.

New changeset 8a0fe3481c91 by Serhiy Storchaka in branch 'default':
Issue bpo-27095: Simplified MAKE_FUNCTION and removed MAKE_CLOSURE opcodes.
https://hg.python.org/cpython/rev/8a0fe3481c91

Pushed with minor changes. Thank you for your contribution Demur!

You change to the documentation of EXTENDED_ARG isn't pushed. I would prefer to see it as a part of larger documentation patch in bpo-26647.

After pushing I found a bug (similar bug was existing) in compile.c. I'll open separate issue for this.