[blshkv]

Summary

Fixes #1089

Gtk.accelerator_parse() returns a gi._gi.ResultTuple, a subclass of tuple that has __hash__ hardcoded to 0. Plain Python tuples use content-based hashing.

This breaks set membership checks in two tests:

all_default_accelerators = {
    Gtk.accelerator_parse(accel)   # ResultTuple, hash=0
    ...
}
assert (key, mods) in all_default_accelerators  # plain tuple, hash≠0 → always False

Even though ResultTuple == plain_tuple returns True, Python set lookup first checks the hash bucket. Since hash(plain_tuple) != 0 in general, the element is never found.

Fix

Wrap Gtk.accelerator_parse() with tuple() in both affected set/list comprehensions so all objects share the same hash implementation:

all_default_accelerators = {
    tuple(Gtk.accelerator_parse(accel))
    ...
}

Affected tests

• test_non_empty_default_keybinding_accels_are_distinct
• test_duplicate_accels_not_possible_to_set

[mattrose]

Given that the same tests pass in my test environment where I try to match the conditions described in the issue I'm hesitant to pull in this fix. If you could describe the environment and conditions in which the tests fail so that I can ensure the tests fix this, I would appreciate it. Maybe it has a newer version of Gtk that behaves differently?

[mattrose]

Found an environment where the issue could be reproduced. Testing fix now