Generate cache for builtins

[alexandrasp]

• Create a set with all attributes available in the most recent python (3.12.2) builtins.

• Use this cache in Symtab.py in declare_builtin functions.

Fix to #5591

[da-woods]

In terms of tests:

I don't think we have any tests for the previous version of this feature. It's probably the sort of thing that needs to be tested by the content of the C code, which is always a little fiddly.

The way to do that is test_assert_c_code_has or test_fail_if_c_code_has.

[scoder]

Regarding tests, I propose the following:

Add a constant KNOWN_PYTHON_BUILTINS_VERSION next to the set that holds the exact sys.version_info tuple value of the CPython version that you took the list from.

Add a test that compares the version to sys.version_info and skips the test for older Python releases. For that version and newer, it compares the set of builtins to the system list and fails if they differ.

That way, we notice changes in the list of builtins through test failures and can adapt in time.

EDIT: when skipping the test (on older versions), it would still be nice to print any differences in the skip message, so that it's easy to look through build logs for the list of supported/expected builtins of a specific Python version.

[alexandrasp]

Regarding tests, I propose the following:

Add a constant KNOWN_PYTHON_BUILTINS_VERSION next to the set that holds the exact sys.version_info tuple value of the CPython version that you took the list from.

done!

Add a test that compares the version to sys.version_info and skips the test for older Python releases. For that version and newer, it compares the set of builtins to the system list and fails if they differ.

done! But, I added the test in TestBuiltin.py.

That way, we notice changes in the list of builtins through test failures and can adapt in time.

EDIT: when skipping the test (on older versions), it would still be nice to print any differences in the skip message, so that it's easy to look through build logs for the list of supported/expected builtins of a specific Python version.

I got the difference between the builtin available in the runtime env and the fixed version and printed the list of possible missing builtins if there are any.

[alexandrasp]

After checking the errors reported by the CI, I am wondering if the environment used to run the tests could have any additional features, or even if I should use anything different than dir() to obtain __builtins__ attributes.

For example:

ubuntu-20.04, 3.13-dev, c,cpp, --limited-api, --no-file, true, -limited_api

ubuntu-20.04, 3.12, c,cpp, --limited-api, --no-file, true, -limited_api

windows-2019, cpp, 3.12

Skipping seems to be fine in the case of old versions:

windows-2019, c, 3.7

ubuntu-20.04, c, 3.9

[scoder]

I just noticed that we didn't keep uncachable_builtins up to date, which holds the list of builtins that change across Python releases and thus cannot be cached globally. I'll update it.

[scoder]

Looks like the list (and the test) includes module attributes that are not builtins.