bpo-29469: Remove unnecessary peephole optimizer

[methane]

Conversions like not a is b -> a is not b are implemented
in AST optimizer in 7ea143a.
This commit removes them from peephole optimizer.

https://bugs.python.org/issue29469

[serhiy-storchaka]

See also Raymond's comment on the tracker.

[methane]

I saw it but I don't understand it yet.
Is it easy enough to do in this pull request?

[serhiy-storchaka]

Are CONST_STACK_* macros and lastn_const_start() and fold_tuple_on_constants() still needed?

[methane]

I'm reading it now.

[methane]

I did it, but importlib.h and importlib_external.h has changed.
I'm checking what happened.

[serhiy-storchaka]

You have removed the optimization for for i in [1, 2]. It can be implemented at AST level.

[methane]

Removing BUILD_TUPLE folding makes regression on MAKE_FUNCTION:

before:

 459         204 LOAD_CONST              96 ((None, None))
             206 LOAD_CONST              39 (<code object _spec_from_module at 0x7f7d61834420, file "_bootstrap.py", line 459>)
             208 LOAD_CONST              40 ('_spec_from_module')
             210 MAKE_FUNCTION            1
             212 STORE_NAME              22 (_spec_from_module)

after:

 459         204 LOAD_CONST               1 (None)
             206 LOAD_CONST               1 (None)
             208 BUILD_TUPLE              2
             210 LOAD_CONST              39 (<code object _spec_from_module at 0x7f5f36a2c5d0, file "_bootstrap.py", line 459>)
             212 LOAD_CONST              40 ('_spec_from_module')
             214 MAKE_FUNCTION            1
             216 STORE_NAME              22 (_spec_from_module)

[serhiy-storchaka]

Oh, right. And this can't be moved to the AST level. Thus tuple constants folding should be kept. It can be moved to the compiler lever, but this is a separate complex issue.

[methane]

Maybe, I need to re-implement fold_tuple_on_constants() to export consts from
codes. But I have no time for it today.