The reason why I am doing this is that consuming could cause no implicit
copy errors. Instead __owned will just potentially cause an error if
someone assigns into sending.
rdar://131066640
Previously we would just not print sending. This causes problems since sending
implies a +1 parameter and by removing it we convert the parameter to a +0
parameter, breaking ABI. In this commit, I make it so that when we suppress
sending from argument parameters, we just replace it with consuming so that we
preserve ABI even for callers who do not support sending.
rdar://131066640
I am using this to better test out suppression statements. I am finding that
FileCheck runs into issues with some of the '#if' lines I am trying to match. I
am able to use this option with my asserts only test to uniquely identify a '#if
...' statement and thus have the pattern matching work. I needed this to get the
test in the next commit to pass testing.
These already worked... I am just adding the code coverage before I fix a
different issue in the next commit. This will make it clearer what I am actually
fixing in the next commit when one reads the tests.
The code here was assuming that if we already emitted a compiler guard for
non-Suppressable features, we could avoid doing it for suppressable
features. The problem with this is that compiler() does more than just check for
compiler versions... it also tells the compiler that parser errors in the if
block should be ignored when if evaluates to false.
rdar://129045783