Speed up single byte cases. See ChangeLog.

6 files changed, 45 insertions, 10 deletions

diff --git a/ChangeLog b/ChangeLog
index f9e4e1d1..da63c9d9 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,14 @@
+Sat Dec  4 21:44:38 2010  Arnold D. Robbins  <arnold@skeeve.com>
+
+	* node.c (init_btowc_cache): New function.
+	(btowc_cache): New array.
+	(str2wstr): Use is_valid_character in test instead of several isXXX
+	calls.
+	* awk.h [is_valid_character]: Macro to use btowc_cache.
+	* main.c (main): Call init_btowc_cache().
+	* io.c (rs1scan): Add call to is_valid_character when processing
+	characters byte by byte.
+
 Wed Dec  1 08:10:21 2010  Arnold D. Robbins  <arnold@skeeve.com>
 
 	* awk.h, awkgram.y, debug.c: Change CONTEXT to AWK_CONTEXT
diff --git a/TODO b/TODO
index 02dbcc28..9276532a 100644
--- a/TODO
+++ b/TODO
@@ -1,9 +1,7 @@
 FIX regular field splitting to use FPAT algorithm.
 
 #Revise plug-in code to check for special symbol per GNU coding standards.
-
-Paolo's code for checking for single char values in str2wstr().
-
+#Paolo's code for checking for single char values in str2wstr().
 #Fix os_close_on_exec to do read/modify/write of the fd flags.
 
 #Consider forcing [a-z] int abc...wxyz in regexes, no matter what the locale.
diff --git a/awk.h b/awk.h
index 993e5820..8a5c6eb8 100644
--- a/awk.h
+++ b/awk.h
@@ -1323,8 +1323,12 @@ extern const wchar_t *wstrstr(const wchar_t *haystack, size_t hs_len,
 extern const wchar_t *wcasestrstr(const wchar_t *haystack, size_t hs_len,
 		const wchar_t *needle, size_t needle_len);
 extern void free_wstr(NODE *n);
+extern wint_t btowc_cache[];
+extern void init_btowc_cache();
+#define is_valid_character(b)	(btowc_cache[b] != WEOF)
 #else
 #define free_wstr(NODE)	/* empty */
+#define is_valid_character(c)	(TRUE)
 #endif
 /* re.c */
 extern Regexp *make_regexp(const char *s, size_t len, int ignorecase, int dfa, int canfatal);
diff --git a/io.c b/io.c
index 54fa46bb..efad6af3 100644
--- a/io.c
+++ b/io.c
@@ -2684,13 +2684,19 @@ rs1scan(IOBUF *iop, struct recmatch *recm, SCANSTATE *state)
 	if (rs != '\n' && gawk_mb_cur_max > 1) {
 		int len = iop->dataend - bp;
 		int found = 0;
+
 		memset(&mbs, 0, sizeof(mbstate_t));
 		do {
 			if (*bp == rs)
 				found = 1;
-			mbclen = mbrlen(bp, len, &mbs);
-			if ((mbclen == 1) || (mbclen == (size_t) -1)
-					|| (mbclen == (size_t) -2) || (mbclen == 0)) {
+			if (is_valid_character(*bp))
+				mbclen = 1;
+			else
+				mbclen = mbrlen(bp, len, &mbs);
+			if (   (mbclen == 1)
+			    || (mbclen == (size_t) -1)
+			    || (mbclen == (size_t) -2)
+			    || (mbclen == 0)) {
 				/* We treat it as a singlebyte character.  */
 				mbclen = 1;
 			}
diff --git a/main.c b/main.c
index b09027cf..01561b47 100644
--- a/main.c
+++ b/main.c
@@ -293,6 +293,9 @@ main(int argc, char **argv)
 	 */
 	gawk_mb_cur_max = MB_CUR_MAX;
 	/* Without MBS_SUPPORT, gawk_mb_cur_max is 1. */
+
+	/* init the cache for checking bytes if they're characters */
+	init_btowc_cache();
 #endif
 
 	(void) bindtextdomain(PACKAGE, LOCALEDIR);
diff --git a/node.c b/node.c
index d181bff9..0dd5f52a 100644
--- a/node.c
+++ b/node.c
@@ -706,11 +706,9 @@ str2wstr(NODE *n, size_t **ptr)
 		 * 9/2010: Check the current byte; if it's a valid character,
 		 * then it doesn't start a multibyte sequence. This brings a
 		 * big speed up. Thanks to Ulrich Drepper for the tip.
+		 * 11/2010: Thanks to Paolo Bonzini for some even faster code.
 		 */
-		if (   isprint(*sp)
-		    || isgraph(*sp)
-		    || iscntrl(*sp)
-		    || *sp == '\0' ) {
+		if (is_valid_character(*sp)) {
 			count = 1;
 			wc = *sp;
 		} else
@@ -894,3 +892,18 @@ get_ieee_magic_val(const char *val)
 
 	return v;
 }
+
+#ifdef MBS_SUPPORT
+wint_t btowc_cache[256];
+
+/* init_btowc_cache --- initialize the cache */
+
+void init_btowc_cache()
+{
+	int i;
+
+	for (i = 0; i < 255; i++) {
+		btowc_cache[i] = btowc(i);
+	}
+}
+#endif