file actions are not yet implemented, but everything else should be
mostly complete and roughly correct.

also modify wcsncpy to use the same loop logic

the observed symptom was that the code was incorrectly rounding up
1.0625 to 1.063 despite the rounding mode being round-to-nearest with
ties broken by rounding to even last place. however, the code was just
not right in many respects, and i'm surprised it worked as well as it
did. this time i tested the values that end up in the variables round,
small, and the expression round+small, and all look good.

the new approach relies on the fact that the only ways to create
sigset_t objects without invoking UB are to use the sig*set()
functions, or from the masks returned by sigprocmask, sigaction, etc.
or in the ucontext_t argument to a signal handler. thus, as long as
sigfillset and sigaddset avoid adding the "protected" signals, there
is no way the application will ever obtain a sigset_t including these
bits, and thus no need to add the overhead of checking/clearing them
when sigprocmask or sigaction is called.

note that the old code actually *failed* to remove the bits from
sa_mask when sigaction was called.

the new implementations are also significantly smaller, simpler, and
faster due to ignoring the useless "GNU HURD signals" 65-1024, which
are not used and, if there's any sanity in the world, never will be
used.

these should be tweaked according to testing. offhand i know 1000 is
too low and 5000 is likely to be sufficiently high. consider trying to
add futexes to file locking, too...

the previous implementation had at least 2 problems:

1. the case where additional threads reached the barrier before the
first wave was finished leaving the barrier was untested and seemed
not to be working.

2. threads leaving the barrier continued to access memory within the
barrier object after other threads had successfully returned from
pthread_barrier_wait. this could lead to memory corruption or crashes
if the barrier object had automatic storage in one of the waiting
threads and went out of scope before all threads finished returning,
or if one thread unmapped the memory in which the barrier object
lived.

the new implementation avoids both problems by making the barrier
state essentially local to the first thread which enters the barrier
wait, and forces that thread to be the last to return.

the previous fix was incorrect, as it would prevent f->close(f) from
being called if fflush(f) failed. i believe this was the original
motivation for using | rather than ||. so now let's just use a second
statement to constrain the order of function calls, and to back to
using |.

pcc turned up this bug by calling f->close(f) before fflush(f),
resulting in lost output and error on flush.

with this patch, musl compiles and mostly works with pcc 1.0.0. a few
tests are still failing and i'm uncertain whether they are due to
portability problems in musl, or bugs in pcc, but i suspect the
latter.

this slightly cuts down on the degree musl "fights with" gcc, but more
importantly, it fixes a critical bug when gcc inlines a variadic
function and optimizes out the variadic arguments due to noticing that
they were "not used" (by __builtin_va_arg).

we leave the old code in place if __GNUC__ >= 3 is false; it seems
like it might be necessary at least for tinycc support and perhaps if
anyone ever gets around to fixing gcc 2.95.3 enough to make it work..

the old versions worked, but conflicted with programs which declared
their own prototypes and generated warnings with some versions of gcc.

Smoothsort is an adaptive variant of heapsort. This version was
written by Valentin Ochs (apo) specifically for inclusion in musl. I
worked with him to get it working in O(1) memory usage even with giant
array element widths, and to optimize it heavily for size and speed.
It's still roughly 4 times as large as the old heap sort
implementation, but roughly 20 times faster given an almost-sorted
array of 1M elements (20 being the base-2 log of 1M), i.e. it really
does reduce O(n log n) to O(n) in the mostly-sorted case. It's still
somewhat slower than glibc's Introsort for random input, but now
considerably faster than glibc when the input is already sorted, or
mostly sorted.

strictly speaking this and a few other ops should be factored into
asm.h or the file should just be renamed to asm.h, but whatever. clean
it up someday.

1. failed match of literal chars from the format string would always
return matching failure rather than input failure at eof, leading to
infinite loops in some programs.

2. unread of eof would wrongly adjust the character counts reported by
%n, yielding an off-by-one error.

this is explicitly allowed by POSIX

some functions that should have been testing whether pthread_self()
had been called and initialized the thread pointer were instead
testing whether pthread_create() had been called and actually made the
program "threaded". while it's unlikely any mismatch would occur in
real-world problems, this could have introduced subtle bugs. now, we
store the address of the main thread's thread descriptor in the libc
structure and use its presence as a flag that the thread register is
initialized. note that after fork, the calling thread (not necessarily
the original main thread) is the new main thread.

the linux documentation for dup2 says it can fail with EBUSY due to a
race condition with open and dup in the kernel. shield applications
(and the rest of libc) from this nonsense by looping until it succeeds