Switch to a much faster invalidation algorithm

_cached_vis is always calculated as a side effect of calculating any
other cached value, and so doubles as a flag for whether an item (symbol
or choice) has any cached values.

If _cached_vis is None for some item, it also indirectly means that no
other item can have any cached values that (actually) depend on the
item, because _cached_vis would have gotten calculated as a side effect
of calculating any such cached value. Therefore, it's safe to stop
recursive invalidation at an item that has _cached_vis None.

Huge speedup for the allyes/noconfig*.py scripts. allnoconfig_simpler.py
went from 2.671 seconds to 1.067 seconds.

The dependency selftests need to be updated too now that
_get_dependent() is gone. The kernel defconfig tests pass even if all
global invalidation is disabled at least (except for the few defconfigs
that don't set MODULES=y), and that's a huge invalidation test. Seems
pretty speedy too, even though there's some redundant work, so maybe
things could be simplified a bit...

0 files changed, 0 insertions, 0 deletions