implement mbtowc directly, not as a wrapper for mbrtowc

the interface contract for mbtowc admits a much faster implementation
than mbrtowc can achieve; wrapping mbrtowc with an extra call frame
only made the situation worse.

since the regex implementation uses mbtowc already, this change should
improve regex performance too. it may be possible to improve
performance in other places internally by switching from mbrtowc to
mbtowc.

src/multibyte/mbtowc.c

@@ -10,10 +10,44 @@
 #include <errno.h>
 
 #include "internal.h"
-
-int mbtowc(wchar_t *restrict wc, const char *restrict s, size_t n)
+#include <stdio.h>
+int mbtowc(wchar_t *restrict wc, const char *restrict src, size_t n)
 {
-	mbstate_t st = { 0 };
-	n = mbrtowc(wc, s, n, &st);
-	return n+2 ? n : -1;
+	unsigned c;
+	const unsigned char *s = (const void *)src;
+
+	if (!s) return 0;
+	if (!n) goto ilseq;
+	if (!wc) wc = (void *)&wc;
+
+	if (*s < 0x80) return !!(*wc = *s);
+	if (*s-SA > SB-SA) goto ilseq;
+	c = bittab[*s++-SA];
+
+	/* Avoid excessive checks against n: If shifting the state n-1
+	 * times does not clear the high bit, then the value of n is
+	 * insufficient to read a character */
+	if (n<4 && ((c<<(6*n-6)) & (1U<<31))) goto ilseq;
+
+	if (OOB(c,*s)) goto ilseq;
+	c = c<<6 | *s++-0x80;
+	if (!(c&(1U<<31))) {
+		*wc = c;
+		return 2;
+	}
+
+	if (*s-0x80u >= 0x40) goto ilseq;
+	c = c<<6 | *s++-0x80;
+	if (!(c&(1U<<31))) {
+		*wc = c;
+		return 3;
+	}
+
+	if (*s-0x80u >= 0x40) goto ilseq;
+	*wc = c<<6 | *s++-0x80;
+	return 4;
+
+ilseq:
+	errno = EILSEQ;
+	return -1;
 }