Revert speed regression from 2008-06-16.

* m4/output.c (m4_shipout_int): Avoid obstack_printf in hot path.
* modules/m4.c (numb_obstack): Likewise.
* m4/input.c (MATCH): Use fewer conditionals, and factor
adjustment of S...
(match_input): ...here for smaller code size.
* m4/syntax.c (m4_reset_syntax, m4_set_quotes, m4_set_comment):
Supply trailing NUL to delimiters, to meet contract of faster
MATCH.

Signed-off-by: Eric Blake <ebb9@byu.net>

5 files changed, 71 insertions, 27 deletions

diff --git a/ChangeLog b/ChangeLog
index 7dc8277f..c838a8b8 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,15 @@
+2008-06-19  Eric Blake  <ebb9@byu.net>
+
+	Revert speed regression from 2008-06-16.
+	* m4/output.c (m4_shipout_int): Avoid obstack_printf in hot path.
+	* modules/m4.c (numb_obstack): Likewise.
+	* m4/input.c (MATCH): Use fewer conditionals, and factor
+	adjustment of S...
+	(match_input): ...here for smaller code size.
+	* m4/syntax.c (m4_reset_syntax, m4_set_quotes, m4_set_comment):
+	Supply trailing NUL to delimiters, to meet contract of faster
+	MATCH.
+
 2008-06-18  Eric Blake  <ebb9@byu.net>
 
 	Whitespace cleanup.
@@ -38,7 +50,7 @@
 	rather than hand-rolled equivalents.  Ensure that embedded NUL in
 	trace output does not truncate the trace string.
 	Memory impact: none.
-	Speed impact: none noticed.
+	Speed impact: noticeable penalty, from obstack_printf overhead.
 	* ltdl/m4/gnulib-cache.m4: Import obstack-printf-posix module.
 	* m4/macro.c (trace_format): Delete; use obstack_printf instead.
 	(trace_header, trace_pre, trace_post): All callers updated.
diff --git a/m4/input.c b/m4/input.c
index ea59b44a..4306613c 100644
--- a/m4/input.c
+++ b/m4/input.c
@@ -1353,10 +1353,11 @@ m4_skip_line (m4 *context, const m4_call_info *caller)
 
 
 /* If the string S of length LEN matches the next characters of the
-   input stream, return true.  If CONSUME is true and a match is
-   found, the input is discarded; otherwise any characters read are
-   pushed back again.  The function is used only when multicharacter
-   quotes or comment delimiters are used.
+   input stream, return true.  If CONSUME, the first byte has already
+   been matched.  If a match is found and CONSUME is true, the input
+   is discarded; otherwise any characters read are pushed back again.
+   The function is used only when multicharacter quotes or comment
+   delimiters are used.
 
    All strings herein should be unsigned.  Otherwise sign-extension
    of individual chars might break quotes with 8-bit chars in it.
@@ -1373,6 +1374,11 @@ match_input (m4 *context, const char *s, size_t len, bool consume)
   m4_obstack *st;
   bool result = false;
 
+  if (consume)
+    {
+      s++;
+      len--;
+    }
   assert (len);
   ch = peek_char (context, false);
   if (ch != to_uchar (*s))
@@ -1407,17 +1413,19 @@ match_input (m4 *context, const char *s, size_t len, bool consume)
   return result;
 }
 
-/* The macro MATCH() is used to match an unsigned char string S of
-  length LEN against the input.  The first character is handled
-  inline, for speed.  Hopefully, this will not hurt efficiency too
-  much when single character quotes and comment delimiters are used.
-  If CONSUME, then CH is the result of next_char, and a successful
-  match will discard the matched string.  Otherwise, CH is the result
-  of peek_char, and the input stream is effectively unchanged.  */
+/* The macro MATCH() is used to match a string S of length LEN against
+  the input.  The first character is handled inline for speed, and
+  S[LEN] must be safe to dereference (it is faster to do character
+  comparison prior to length checks).  This improves efficiency for
+  the common case of single character quotes and comment delimiters,
+  while being safe for disabled delimiters as well as longer
+  delimiters.  If CONSUME, then CH is the result of next_char, and a
+  successful match will discard the matched string.  Otherwise, CH is
+  the result of peek_char, and the input stream is effectively
+  unchanged.  */
 #define MATCH(C, ch, s, len, consume)					\
-  ((len) && to_uchar ((s)[0]) == (ch)					\
-   && ((len) == 1							\
-       || match_input (C, (s) + (consume), (len) - (consume), consume)))
+  (to_uchar ((s)[0]) == (ch)						\
+   && ((len) >> 1 ? match_input (C, s, len, consume) : (len)))
 
 /* While the current input character has the given SYNTAX, append it
    to OBS.  Take care not to pop input source unless the next source
diff --git a/m4/output.c b/m4/output.c
index f86f913b..7f6f9c12 100644
--- a/m4/output.c
+++ b/m4/output.c
@@ -575,7 +575,23 @@ m4_divert_text (m4 *context, m4_obstack *obs, const char *text, size_t length,
 void
 m4_shipout_int (m4_obstack *obs, int val)
 {
-  obstack_printf (obs, "%d", val);
+  /* Using obstack_printf (obs, "%d", val) has too much overhead.  */
+  unsigned int uval;
+  char buf[INT_BUFSIZE_BOUND (unsigned int)];
+  char *p = buf + INT_STRLEN_BOUND (unsigned int);
+
+  if (val < 0)
+    {
+      obstack_1grow (obs, '-');
+      uval = -(unsigned int) val;
+    }
+  else
+    uval = val;
+  *p = '\0';
+  do
+    *--p = '0' + uval % 10;
+  while (uval /= 10);
+  obstack_grow (obs, p, strlen (p));
 }
 
 /* Output the text S, of length LEN, to OBS.  If QUOTED, also output
diff --git a/m4/syntax.c b/m4/syntax.c
index 3bba0a8b..8dda94ea 100644
--- a/m4/syntax.c
+++ b/m4/syntax.c
@@ -21,6 +21,7 @@
 #include <config.h>
 
 #include "m4private.h"
+#include "xmemdup0.h"
 
 /* Define this to see runtime debug info.  Implied by DEBUG.  */
 /*#define DEBUG_SYNTAX */
@@ -393,13 +394,14 @@ m4_reset_syntax (m4_syntax_table *syntax)
   free (syntax->comm.str1);
   free (syntax->comm.str2);
 
-  syntax->quote.str1 = xmemdup (DEF_LQUOTE, 1);
+  /* The use of xmemdup0 is exploited by input.c.  */
+  syntax->quote.str1 = xmemdup0 (DEF_LQUOTE, 1);
   syntax->quote.len1 = 1;
-  syntax->quote.str2 = xmemdup (DEF_RQUOTE, 1);
+  syntax->quote.str2 = xmemdup0 (DEF_RQUOTE, 1);
   syntax->quote.len2 = 1;
-  syntax->comm.str1 = xmemdup (DEF_BCOMM, 1);
+  syntax->comm.str1 = xmemdup0 (DEF_BCOMM, 1);
   syntax->comm.len1 = 1;
-  syntax->comm.str2 = xmemdup (DEF_ECOMM, 1);
+  syntax->comm.str2 = xmemdup0 (DEF_ECOMM, 1);
   syntax->comm.len2 = 1;
 
   add_syntax_attribute (syntax, to_uchar (syntax->quote.str2[0]),
@@ -619,9 +621,10 @@ m4_set_quotes (m4_syntax_table *syntax, const char *lq, size_t lq_len,
 
   free (syntax->quote.str1);
   free (syntax->quote.str2);
-  syntax->quote.str1 = xmemdup (lq, lq_len);
+  /* The use of xmemdup0 is exploited by input.c.  */
+  syntax->quote.str1 = xmemdup0 (lq, lq_len);
   syntax->quote.len1 = lq_len;
-  syntax->quote.str2 = xmemdup (rq, rq_len);
+  syntax->quote.str2 = xmemdup0 (rq, rq_len);
   syntax->quote.len2 = rq_len;
 
   /* changequote overrides syntax_table, but be careful when it is
@@ -692,9 +695,10 @@ m4_set_comment (m4_syntax_table *syntax, const char *bc, size_t bc_len,
 
   free (syntax->comm.str1);
   free (syntax->comm.str2);
-  syntax->comm.str1 = xmemdup (bc, bc_len);
+  /* The use of xmemdup0 is exploited by input.c.  */
+  syntax->comm.str1 = xmemdup0 (bc, bc_len);
   syntax->comm.len1 = bc_len;
-  syntax->comm.str2 = xmemdup (ec, ec_len);
+  syntax->comm.str2 = xmemdup0 (ec, ec_len);
   syntax->comm.len2 = ec_len;
 
   /* changecom overrides syntax_table, but be careful when it is used
diff --git a/modules/m4.c b/modules/m4.c
index 1857d6f4..c46b3af3 100644
--- a/modules/m4.c
+++ b/modules/m4.c
@@ -1207,9 +1207,13 @@ numb_obstack (m4_obstack *obs, number value, int radix, int min)
       s++;
     }
   len = strlen (s);
-  if (min < len)
-    min = len;
-  obstack_printf (obs, "%.*d%s", min - len, 0, s);
+  if (len < min)
+    {
+      min -= len;
+      obstack_blank (obs, min);
+      memset ((char *) obstack_next_free (obs) - min, '0', min);
+    }
+  obstack_grow (obs, s, len);
 }