There is a POSIX definition for sched_setscheduler(), but Linux does
not implement that; instead of changing the process's scheduler, it
only affects one thread.  This has caused some confusion among
application developers and C library developers.

While glibc implements Linux semantics, Musl has made their
sched_setscheduler() function an always-failing no-op, causing the
error message "sched_setscheduler failed: Function not implemented".

 http://git.musl-libc.org/cgit/musl/commit/src/sched/sched_setscheduler.c?id=1e21e78bf7a5c24c217446d8760be7b7188711c2

Instead of relying on the C library which may be unreliable here, we
now roll our own system call wrapper.

Closes #218

Juse in cas glibc gets a wrapper for the system call which would then
conflict with ours.

This attribute shall be used only for IsInside() to make this safe
against a race condition described in #188:

> There is no requirement on the implementation that the ID of the
> created thread be available before the newly created thread starts
> executing.

http://pubs.opengroup.org/onlinepubs/009695399/functions/pthread_create.html):

This means that on some pthread implementations (e.g. Haiku), the
assert(thread.IsInside()) could fail.

Closes #188

This method is only used inside assert().

Use the "==" operator instead of pthread_equal().

This allows us to eliminate two boolean flags which are there to avoid
race conditions, and made the thing so fragile that I got tons of
(correct) thread sanitizer warnings.

Only _WIN32 is defined by the compiler, and WIN32 is not standardized
and may be missing.

Closes #169

This eliminates some overhead, because the compiler doesn't need to
consider these functions throwing.

Make sure this deprecated API is not used.

We need purely RAII-style, and this is std::lock_guard.

pthread_cond_timedwait() in PosixCond.hxx:timed_wait(PosixMutex...) returns
EINVAL, if ts.tv_nsec >= 1E9. In this case, it returns to early.

Find attached a patch which fixes this. I chose a compare-subtraction method
to keep ts.tv_nsec below 1E9.

Another option would be
ts.tv_sec += ts.tv_nsec / 1000000000;
ts.tv_nsec %= 1000000000;

But I guess this takes more time on some ARM processors, which don't support
hardware division.

Reduce header clutter.

Apparently all other C libraries are not compatible with "constexpr".
Those which are not will get a performance penalty, but at least they
work at all.

This only applies to linux systems.  Here, sched_setscheduler() is
called to get realtime scheduling.  With this patch, the return value
of this function is now checked and a warning / error message is
generated if it fails.

Apparently all other C libraries are not compatible with "constexpr".
Those which are not will get a performance penalty, but at least they
work at all.

Caused a build failure with uClibc because snprintf() was not
available.

NetBSD's pthread_setname_np() prototype is incompatible with the rest
of the world, and it requires to pass the string argument as a
non-const pointer.  Instead of working around this misdesign, I hereby
disable the feature on NetBSD.

Add macro HAVE_THREAD_NAME which is set when any method to set the
thread name is available.  Use that macro in FormatThreadName()
instead of just checking for HAVE_PTHREAD_SETNAME_NP.

The "::" to explicitly refer to the global namespace appeared like a
good idea in C++, but it breaks with C libraries that implement
standard functions using macros (e.g. musl).

On NetBSD, PTHREAD_MUTEX_INITIALIZER and PTHREAD_COND_INITIALIZER are
not compatible with C++11 "constexpr" (see Mantis ticket 0004110).  As
a workaround, don't ues "constexpr", and use the functions
pthread_mutex_init(), pthread_mutex_destroy(), pthread_cond_init() and
pthread_cond_destroy() instead.  This adds some runtime overhead, but
is portable to POSIX implementations that have awkward initializer
macros.