evaluate -Wconditional-uninitialized

[nickdesaulniers]

This came up in an internal discussion, but we recently had a boogaloo because of some uninitialized value. It looks like clang could have caught that with -Wconditional-uninitialized, but:

1. -Wconditional-uninitialized is not part of the -Wuninitialized group (surprising)
2. The kernel doesn't enable -Wconditional-uninitialized.

I can't recall if we ever evaluated the use of this flag on the kernel or not. I'm curious if we can turn it on or if it has false positives?

Example: https://godbolt.org/z/4G7G5MjGa

[nathanchance]

I recall looking into -Wconditional-uninitialized at one point in the past couple of years but from what I can tell, it suffers from many of the same pitfalls that -Wmaybe-uninitialized (GCC's version) has, which caused it to get disabled. It was quite noisy and most of the instances I saw appeared to be false positives.

[nickdesaulniers]

It definitely caught an internal bug that -Wsometimes-uninitialized did not. I'm guessing it might have a measurable impact on compile times, but I did not measure.

[nickdesaulniers]

I tested it quickly, and it seemed like this pattern had false positives:

int foo;
for (...)
  foo = 7;
use(foo);

Would flag in cases where it's not obvious that the loop body is ever executed.