Rather than bodging the test to make it more robust, fix the functions
in the Threading layer to behave the same as they do on other platforms,
i.e. to guarantee that they always wait *at least* the amount of time
you asked for.
rdar://100236038
We were inadvertently rounding down, which was fine almost 100% of the
time, but on Windows where the timer resolution is only 1ms, it made
the test _very slightly_ flaky because we were asking to wait for
9ms rather than 10ms, then checking we'd waited at least 10ms.
We should explicitly round up, for that reason.
rdar://100236038
This repairs the Windows x86 SDK build after #59287. Use `intptr_t`
rather than the explicitly sized integer as this is always guaranteed to
be the size of the pointer width (irrespective of LP64/LP32 or
LLP64/LLP32). Since this impacts only the once predicate, this
shouldn't impact any assumptions around structure sizes.
There's no guarantee that e.g. pthread_key_t is an integral type. It could
be some kind of struct, or some other thing that isn't valid as a template
argument.
rdar://90776105
Removing thread_get_main() means we don't need a static initializer on
Darwin, which means we can delete Darwin.cpp as well. We can also delete
Nothreads.cpp while we're about it, because there's nothing in that file.
rdar://90776105
Declaring _RTL_SRWLOCK ourselves causes clashes with <windows.h>.
Rather than doing that, declare an equivalent struct, and some overloads.
rdar://90776105
We can't safely include <windows.h> because it defines a large number
of macros, some of which clash with things in the Swift source tree,
and others of which clash with things in the LLVM source tree. Sadly
we *also* can't just include the Windows headers we need, because they
pull in some of the problematic macros.
In this instance, the best thing seems to be to grab the definitions
for the types and functions we are going to use and put them in their
own header file. If we define them correctly, then #including
<windows.h> before or after this header won't have any adverse effects.
rdar://90776105
Moved all the threading code to one place. Added explicit support for
Darwin, Linux, Pthreads, C11 threads and Win32 threads, including new
implementations of Once for Linux, Pthreads, C11 and Win32.
rdar://90776105
There's no guarantee that e.g. pthread_key_t is an integral type. It could
be some kind of struct, or some other thing that isn't valid as a template
argument.
rdar://90776105
Removing thread_get_main() means we don't need a static initializer on
Darwin, which means we can delete Darwin.cpp as well. We can also delete
Nothreads.cpp while we're about it, because there's nothing in that file.
rdar://90776105
Declaring _RTL_SRWLOCK ourselves causes clashes with <windows.h>.
Rather than doing that, declare an equivalent struct, and some overloads.
rdar://90776105
We can't safely include <windows.h> because it defines a large number
of macros, some of which clash with things in the Swift source tree,
and others of which clash with things in the LLVM source tree. Sadly
we *also* can't just include the Windows headers we need, because they
pull in some of the problematic macros.
In this instance, the best thing seems to be to grab the definitions
for the types and functions we are going to use and put them in their
own header file. If we define them correctly, then #including
<windows.h> before or after this header won't have any adverse effects.
rdar://90776105
Moved all the threading code to one place. Added explicit support for
Darwin, Linux, Pthreads, C11 threads and Win32 threads, including new
implementations of Once for Linux, Pthreads, C11 and Win32.
rdar://90776105
