[ruby-dev:8638] Re: slow gsub

まつもと ゆきひろです

In message "[ruby-dev:8637] Re: slow gsub"
    on 99/12/13, Yukihiro Matsumoto <[e-mail]> writes:

||mbclen から mbclen2 に変わって gsub がとても遅くなりました.
||単なるマクロから関数になってるわけで, しかたないのかなあ?
|
|mbclenで十分なときにはmbclenを使う方向で考えてみましょう。

真の原因はstartposを文字単位で調整する処理にありました。
これを別関数に分離してみたら、それほど遅くないみたいです。

Index: regex.h
===================================================================
RCS file: /home/cvs/ruby/regex.h,v
retrieving revision 1.4
diff -u -1 -r1.4 regex.h
--- regex.h	1999/12/06 09:03:58	1.4
+++ regex.h	1999/12/13 09:47:22
@@ -30,2 +30,3 @@
 #ifdef RUBY
+# define re_adjust_startpos ruby_re_adjust_startpos
 # define re_compile_fastmap ruby_re_compile_fastmap
@@ -188,2 +189,3 @@
 /* Is this really advertised?  */
+extern int re_adjust_startpos (struct re_pattern_buffer *, const char*, int, int, int);
 extern void re_compile_fastmap (struct re_pattern_buffer *);
@@ -208,2 +210,3 @@
 /* Is this really advertised? */
+extern int re_adjust_startpos ();
 extern void re_compile_fastmap ();
Index: regex.c
===================================================================
RCS file: /home/cvs/ruby/regex.c,v
retrieving revision 1.10
diff -u -1 -r1.10 regex.c
--- regex.c	1999/12/06 09:03:56	1.10
+++ regex.c	1999/12/13 09:47:31
@@ -2924,17 +2924,5 @@
 
-
-/* Using the compiled pattern in BUFP->buffer, first tries to match
-   STRING, starting first at index STARTPOS, then at STARTPOS + 1, and
-   so on.  RANGE is the number of places to try before giving up.  If
-   RANGE is negative, it searches backwards, i.e., the starting
-   positions tried are STARTPOS, STARTPOS - 1, etc.  STRING is of SIZE.
-   In REGS, return the indices of STRING that matched the entire
-   BUFP->buffer and its contained subexpressions.
-
-   The value returned is the position in the strings at which the match
-   was found, or -1 if no match was found, or -2 if error (such as
-   failure stack overflow).  */
-
+/* adjust startpos value to the position between characters. */
 int
-re_search(bufp, string, size, startpos, range, regs)
+re_adjust_startpos(bufp, string, size, startpos, range)
      struct re_pattern_buffer *bufp;
@@ -2942,13 +2930,5 @@
      int size, startpos, range;
-     struct re_registers *regs;
 {
-  register char *fastmap = bufp->fastmap;
-  int val, anchor = 0;
-
-  /* Check for out-of-range starting position.  */
-  if (startpos < 0  ||  startpos > size)
-    return -1;
-
   /* Update the fastmap now if not correct already.  */
-  if (fastmap && !bufp->fastmap_accurate) {
+  if (!bufp->fastmap_accurate) {
     re_compile_fastmap(bufp);
@@ -2982,2 +2962,37 @@
   }
+  return startpos;
+}
+
+
+/* Using the compiled pattern in BUFP->buffer, first tries to match
+   STRING, starting first at index STARTPOS, then at STARTPOS + 1, and
+   so on.  RANGE is the number of places to try before giving up.  If
+   RANGE is negative, it searches backwards, i.e., the starting
+   positions tried are STARTPOS, STARTPOS - 1, etc.  STRING is of SIZE.
+   In REGS, return the indices of STRING that matched the entire
+   BUFP->buffer and its contained subexpressions.
+
+   The value returned is the position in the strings at which the match
+   was found, or -1 if no match was found, or -2 if error (such as
+   failure stack overflow).  */
+
+int
+re_search(bufp, string, size, startpos, range, regs)
+     struct re_pattern_buffer *bufp;
+     const char *string;
+     int size, startpos, range;
+     struct re_registers *regs;
+{
+  register char *fastmap = bufp->fastmap;
+  int val, anchor = 0;
+
+  /* Check for out-of-range starting position.  */
+  if (startpos < 0  ||  startpos > size)
+    return -1;
+
+  /* Update the fastmap now if not correct already.  */
+  if (fastmap && !bufp->fastmap_accurate) {
+    re_compile_fastmap(bufp);
+  }
+
 
Index: re.h
===================================================================
RCS file: /home/cvs/ruby/re.h,v
retrieving revision 1.3
diff -u -1 -r1.3 re.h
--- re.h	1999/11/04 08:39:37	1.3
+++ re.h	1999/12/13 09:47:31
@@ -34,2 +34,3 @@
 VALUE rb_reg_regsub _((VALUE, VALUE, struct re_registers *));
+int rb_reg_adjust_startpos _((VALUE, VALUE, int, int));
 
Index: re.c
===================================================================
RCS file: /home/cvs/ruby/re.c,v
retrieving revision 1.7
diff -u -1 -r1.7 re.c
--- re.c	1999/12/01 09:24:13	1.7
+++ re.c	1999/12/13 09:47:34
@@ -510,2 +510,28 @@
 int
+rb_reg_adjust_startpos(reg, str, pos, reverse)
+    VALUE reg, str;
+    int pos, reverse;
+{
+    int range;
+
+    if (may_need_recompile)
+	rb_reg_prepare_re(reg);
+
+    if (FL_TEST(reg, KCODE_FIXED))
+	kcode_set_option(reg);
+    else if (reg_kcode != curr_kcode)
+	kcode_reset_option();
+
+    if (reverse) {
+	range = -pos;
+    }
+    else {
+	range = RSTRING(str)->len - pos;
+    }
+    return re_adjust_startpos(RREGEXP(reg)->ptr,
+			      RSTRING(str)->ptr, RSTRING(str)->len,
+			      pos, range);
+}
+
+int
 rb_reg_search(reg, str, pos, reverse)
Index: string.c
===================================================================
RCS file: /home/cvs/ruby/string.c,v
retrieving revision 1.11
diff -u -1 -r1.11 string.c
--- string.c	1999/12/07 09:23:27	1.11
+++ string.c	1999/12/13 09:47:37
@@ -587,2 +590,3 @@
       case T_REGEXP:
+	pos = rb_reg_adjust_startpos(sub, str, pos, 0);
 	pos = rb_reg_search(sub, str, pos, 0);
@@ -637,2 +641,3 @@
 	if (RREGEXP(sub)->len) {
+	    pos = rb_reg_adjust_startpos(sub, str, pos, 1);
 	    pos = rb_reg_search(sub, str, pos, 1);