If you are going to feel a PR is going too big, you don't need to finish it in a single PR. Keeping the test unchanged and submitting a partial fix is also a preferable way. Don't feel pressured by the unit of test. Thank you!

Ok I'll aim to get this merged as it is now, and then later look into the last issue with optimize=2 and docstrings.

Looks like some more compile flags need to be added to the flag validator:

# The CO_xxx symbols are defined here under the same names defined in
# code.h and used by compile.h, so that an editor search will find them here.
# However, they're not exported in __all__, because they don't really belong to
# this module.
CO_NESTED = 0x0010                      # nested_scopes
CO_GENERATOR_ALLOWED = 0                # generators (obsolete, was 0x1000)
CO_FUTURE_DIVISION = 0x20000            # division
CO_FUTURE_ABSOLUTE_IMPORT = 0x40000     # perform absolute imports by default
CO_FUTURE_WITH_STATEMENT = 0x80000      # with statement
CO_FUTURE_PRINT_FUNCTION = 0x100000     # print function
CO_FUTURE_UNICODE_LITERALS = 0x200000   # unicode string literals
CO_FUTURE_BARRY_AS_BDFL = 0x400000
CO_FUTURE_GENERATOR_STOP = 0x800000     # StopIteration becomes RuntimeError in generators
CO_FUTURE_ANNOTATIONS = 0x1000000       # annotations become strings at runtime

I'll look into it

Looks like there is an error in macos - pylib/Lib/test/test_socket.py. I'm not sure why though - and the test file runs fine on my own mac. Do you have any insight?

That might be just flaky

One question I have is how __future__ works. It looks like from compiler/codegen/src/compile.rs compile_future_features() that __future__ settings like annotations are only collected from __future__ import statements, meaning the flags, especially CO_FUTURE_ANNOTATIONS, while allowed, are meaningless. Is there some plan to pass flags into compile.rs?

That's new to me. How about adding a function to collect __future__ from parsed ast and pass the flag value to compile? This way must be easy in current RustPython compile design.
On the other hand, RustPython is premature and we are on the moving target. Ignoring the future and always enabling CO_FUTURE_ANNOTATIONS will be another way to do it.