This is because we currently JIT in process resulting in weirdness around
swift-frontend wanting to link against the host stdlib and the exec swift code
wanting to link against the target stdlib. For now, we work around this by just
saying in that case we will force swift-frontend to use the just built runtime
so we are consistent. The only potential problem is that a bug in the just built
runtime may now cause these tests to fail. But overall, that would suggest a
problem we may want to catch in it of itself, so the work around I think makes
sense until JITing is done in a separate process (which I think is the right fix).
rdar://78768013
First, on x86-64 Linux does not reserve lower 2 Gb of the address space,
and a space saving trick in the runtime did not work properly, confusing
pointers and failure generation numbers together.
Second, we ignored protocol conformances inside the executable (only
considered shared libraries).
Swift SVN r25972
Most tests were using %swift or similar substitutions, which did not
include the target triple and SDK. The driver was defaulting to the
host OS. Thus, we could not run the tests when the standard library was
not built for OS X.
Swift SVN r24504
