patch by Isaac Dunham. matched closely (maybe not exact) to glibc's
idea of what _BSD_SOURCE should make visible.

it's ok to overlap with integer slot 3 on 32-bit because only slots
0-2 are used on process-local barriers.

these are cruft from the original code which used an explicit string
length rather than null termination. i blindly converted all the
checks to null terminator checks, without noticing that in several
cases, the subsequent switch statement would automatically handle the
null byte correctly.

i don't understand why this has to be conditional on being in BRE
mode, but enabling this code unconditionally breaks a huge number of
ERE test cases.

i've been trying out openmp and it seems like it won't be much use
without this...

we do not bother making h_errno thread-local since the only interfaces
that use it are inherently non-thread-safe. but still use the
potentially-thread-local ABI to access it just to avoid lock-in.

written to avoid multiple conditional jumps and avoid ugly repetitive
lines in the header file.

this one is for program(s|ers) who haven't heard of uint16_t and
uint32_t (which are obviously the correct types for use in such
situations, as they're the argument/return types for ntohs/htons and
ntohl/htonl).

there's no sense in using a powerful lock in exit, because it will
never be unlocked. a thread that arrives at exit while exit is already
in progress just needs to hang forever. use the pause syscall for this
because it's cheap and easy and universally available.

this is all junk, but some programs use it.

the non-prototype declaration of basename in string.h is an ugly
compromise to avoid breaking 2 types of broken software:

1. programs which assume basename is declared in string.h and thus
would suffer from dangerous pointer-truncation if an implicit
declaration were used.

2. programs which include string.h with _GNU_SOURCE defined but then
declare their own prototype for basename using the incorrect GNU
signature for the function (which would clash with a correct
prototype).

however, since C++ does not have non-prototype declarations and
interprets them as prototypes for a function with no arguments, we
must omit it when compiling C++ code. thankfully, all known broken
apps that suffer from the above issues are written in C, not C++.

backported fix from freebsd:
http://svnweb.FreeBSD.org/base?view=revision&revision=233973

1. * in BRE is not special at the beginning of the regex or a
subexpression. this broke ncurses' build scripts.

2. \\( in BRE is a literal \ followed by a literal (, not a literal \
followed by a subexpression opener.

3. the ^ in \\(^ in BRE is a literal ^ only at the beginning of the
entire BRE. POSIX allows treating it as an anchor at the beginning of
a subexpression, but TRE's code for checking if it was at the
beginning of a subexpression was wrong, and fixing it for the sake of
supporting a non-portable usage was too much trouble when just
removing this non-portable behavior was much easier.

this patch also moved lots of the ugly logic for empty atom checking
out of the default/literal case and into new cases for the relevant
characters. this should make parsing faster and make the code smaller.
if nothing else it's a lot more readable/logical.

at some point i'd like to revisit and overhaul lots of this code...

updated nextafter* to use FORCE_EVAL, it can be used in many other
places in the math code to improve readability.

apparently initializing a variable is not "using" it but assigning to
it is "using" it. i don't really like this fix, but it's better than
trying to make a bigger cleanup just before a release, and it should
work fine (tested against nsz's math tests).