regex: copy back from Gnulib

Copy regex-related files back from Gnulib, to fix a problem with
static checking of regex calls noted by Martin Sebor.  This merges the
following changes:

* New macro __attribute_nonnull__ in misc/sys/cdefs.h, for use later
when copying other files back from Gnulib.

* Use __GNULIB_CDEFS instead of __GLIBC__ when deciding
whether to include bits/wordsize.h etc.

* Avoid duplicate entries in epsilon closure table.

* New regex.h macro _REGEX_NELTS to let regexec say that its pmatch
arg should contain nmatch elts.  Use that for regexec, instead of
__attr_access (which is incorrect).

* New regex.h macro _Attr_access_ which is like __attr_access except
portable to non-glibc platforms.

* Add some DEBUG_ASSERTs to pacify gcc -fanalyzer and to catch
recently-fixed performance bugs if they recur.

* Add Gnulib-specific stuff to port the dynarray- and lock-using parts
of regex code to non-glibc platforms.

* Fix glibc bug 11053.

* Avoid some undefined behavior when popping an empty fail stack.

1 files changed, 14 insertions, 4 deletions

diff --git a/include/intprops.h b/include/intprops.h
index 2b6e5e9..3fe64e8 100644
--- a/include/intprops.h
+++ b/include/intprops.h
@@ -132,7 +132,8 @@
    operators might not yield numerically correct answers due to
    arithmetic overflow.  They do not rely on undefined or
    implementation-defined behavior.  Their implementations are simple
-   and straightforward, but they are a bit harder to use than the
+   and straightforward, but they are harder to use and may be less
+   efficient than the INT_<op>_WRAPV, INT_<op>_OK, and
    INT_<op>_OVERFLOW macros described below.
 
    Example usage:
@@ -157,6 +158,9 @@
    must have minimum value MIN and maximum MAX.  Unsigned types should
    use a zero MIN of the proper type.
 
+   Because all arguments are subject to integer promotions, these
+   macros typically do not work on types narrower than 'int'.
+
    These macros are tuned for constant MIN and MAX.  For commutative
    operations such as A + B, they are also tuned for constant B.  */
 
@@ -338,9 +342,15 @@
    arguments should not have side effects.
 
    The WRAPV macros are not constant expressions.  They support only
-   +, binary -, and *.  Because the WRAPV macros convert the result,
-   they report overflow in different circumstances than the OVERFLOW
-   macros do.
+   +, binary -, and *.
+
+   Because the WRAPV macros convert the result, they report overflow
+   in different circumstances than the OVERFLOW macros do.  For
+   example, in the typical case with 16-bit 'short' and 32-bit 'int',
+   if A, B and R are all of type 'short' then INT_ADD_OVERFLOW (A, B)
+   returns false because the addition cannot overflow after A and B
+   are converted to 'int', whereas INT_ADD_WRAPV (A, B, &R) returns
+   true or false depending on whether the sum fits into 'short'.
 
    These macros are tuned for their last input argument being a constant.