Currently a list of constants is replaced with constant tuple in `x in [1, 2]` and `for x in [1, 2]`. The resulted AST is the same as for `x in (1, 2)` and `for x in (1, 2)`.

The proposed simple PR extends this optimization to lists containing non-constants. `x in [a, b]` will be changed into `x in (a, b)` and `for x in [a, b]` will be changed into `for x in (a, b)`. Since creating a tuple is faster than creating a list the latter form is a tiny bit faster.

$ ./python -m timeit -s 'a, b = 1, 2' -- 'for x in [a, b]: pass'
5000000 loops, best of 5: 93.6 nsec per loop

$ ./python -m timeit -s 'a, b = 1, 2' -- 'for x in (a, b): pass'
5000000 loops, best of 5: 74.3 nsec per loop

./python -m timeit -s 'a, b = 1, 2' -- '1 in [a, b]'
5000000 loops, best of 5: 58.9 nsec per loop

$ ./python -m timeit -s 'a, b = 1, 2' -- '1 in (a, b)'
10000000 loops, best of 5: 39.3 nsec per loop

An obvious optimisation, if you ask me. PR looks good to me superficially, but I don't know the AST code well.

The direct inspiration of this optimization was your note that similar optimization was implemented in Python.

New changeset 3f7e9aa2ef215917b9f1521441f67f4ecd33a1bc by Serhiy Storchaka in branch 'master':
bpo-32925: Optimized iterating and containing test for literal lists (GH-5842)
https://github.com/python/cpython/commit/3f7e9aa2ef215917b9f1521441f67f4ecd33a1bc