m4 regex usage [was: Multi-Line Definitions]

Eric Blake <[email protected]> Sat, 29 Sep 2007 18:01:20 -0600

Newsgroups	gmane.comp.sysutils.autoconf.patches,gmane.comp.gnu.m4.general,gmane.comp.gnu.m4.patches
Message-ID	<[email protected]>

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

[adding m4-patches; this branch of the thread can drop other lists]

According to Eric Blake on 9/29/2007 1:31 PM:
> Here's something a bit more telling.  With the attached patch, and in the
> coreutils directory,
> 
> $ M4_TRACE_FILE=~/m4.trace M4=~/m4/src/m4 autoconf

I tweaked my tracer patch a bit to distinguish between patsubst and regexp.

$ sort <m4.trace | uniq -c |sort -n -k1,1 |tail -n 15
...
   1214 p:
   1596
...

So half of the empty lines in my trace actually did a multi-line regex.
But 1214 of them did a patsubst(string, [], []), and m4 wasted time
compiling the empty regex every one of those times!  Applying this to m4,
to add some benefit to autoconf < 2.62 vs. m4 > 1.4.10 (watch for a
followup to autoconf that avoids the empty regex to begin with).

2007-09-29  Eric Blake  <[email protected]>

	Optimize for Autoconf usage pattern.
	* src/builtin.c (m4_regexp, m4_patsubst): Handle empty regex
	faster.

- --
Don't work too hard, make some time for fun as well!

Eric Blake             [email protected]
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFG/udP84KuGfSFAYARAmHzAJwJO8+zwXssS/qlIEfotONpp/epRgCfQgQ3
Rjq/NWvO4ha9S+o3gpv9gdg=
=BSOA
-----END PGP SIGNATURE-----

m4.patch286 (text/plain, 3.2 KB)

From aa46ced67010190918295b965f5e2879dcd9a30c Mon Sep 17 00:00:00 2001
From: Eric Blake <[email protected]>
Date: Sat, 29 Sep 2007 17:48:29 -0600
Subject: [PATCH] Optimize for Autoconf usage pattern.

* src/builtin.c (m4_regexp, m4_patsubst): Handle empty regex
faster.

Signed-off-by: Eric Blake <[email protected]>
---
 ChangeLog     |    6 ++++++
 src/builtin.c |   36 +++++++++++++++++++++++++++---------
 2 files changed, 33 insertions(+), 9 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index 0cea5b5..f29b557 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,9 @@
+2007-09-29  Eric Blake  <[email protected]>
+
+	Optimize for Autoconf usage pattern.
+	* src/builtin.c (m4_regexp, m4_patsubst): Handle empty regex
+	faster.
+
 2007-09-24  Eric Blake  <[email protected]>
 
 	Create .gitignore alongside .cvsignore.
diff --git a/src/builtin.c b/src/builtin.c
index dee2276..65f4585 100644
--- a/src/builtin.c
+++ b/src/builtin.c
@@ -1968,8 +1968,19 @@ m4_regexp (struct obstack *obs, int argc, token_data **argv)
       return;
     }
 
-  victim = TOKEN_DATA_TEXT (argv[1]);
-  regexp = TOKEN_DATA_TEXT (argv[2]);
+  victim = ARG (1);
+  regexp = ARG (2);
+  repl = ARG (3);
+
+  if (!*regexp)
+    {
+      /* The empty regex matches everything!  */
+      if (argc == 3)
+	shipout_int (obs, 0);
+      else
+	obstack_grow (obs, repl, strlen (repl));
+      return;
+    }
 
   init_pattern_buffer (&buf, &regs);
   msg = re_compile_pattern (regexp, strlen (regexp), &buf);
@@ -1993,10 +2004,7 @@ m4_regexp (struct obstack *obs, int argc, token_data **argv)
   else if (argc == 3)
     shipout_int (obs, startpos);
   else if (startpos >= 0)
-    {
-      repl = TOKEN_DATA_TEXT (argv[3]);
-      substitute (obs, victim, repl, &regs);
-    }
+    substitute (obs, victim, repl, &regs);
 
   free_pattern_buffer (&buf, &regs);
 }
@@ -2013,6 +2021,7 @@ m4_patsubst (struct obstack *obs, int argc, token_data **argv)
 {
   const char *victim;		/* first argument */
   const char *regexp;		/* regular expression */
+  const char *repl;
 
   struct re_pattern_buffer buf;	/* compiled regular expression */
   struct re_registers regs;	/* for subexpression matches */
@@ -2029,7 +2038,17 @@ m4_patsubst (struct obstack *obs, int argc, token_data **argv)
       return;
     }
 
-  regexp = TOKEN_DATA_TEXT (argv[2]);
+  victim = ARG (1);
+  regexp = ARG (2);
+  repl = ARG (3);
+
+  /* The empty regex matches everywhere, but if there is no
+     replacement, we need not waste time with it.  */
+  if (!*regexp && !*repl)
+    {
+      obstack_grow (obs, victim, strlen (victim));
+      return;
+    }
 
   init_pattern_buffer (&buf, &regs);
   msg = re_compile_pattern (regexp, strlen (regexp), &buf);
@@ -2042,7 +2061,6 @@ m4_patsubst (struct obstack *obs, int argc, token_data **argv)
       return;
     }
 
-  victim = TOKEN_DATA_TEXT (argv[1]);
   length = strlen (victim);
 
   offset = 0;
@@ -2073,7 +2091,7 @@ m4_patsubst (struct obstack *obs, int argc, token_data **argv)
 
       /* Handle the part of the string that was covered by the match.  */
 
-      substitute (obs, victim, ARG (3), &regs);
+      substitute (obs, victim, repl, &regs);
 
       /* Update the offset to the end of the match.  If the regexp
 	 matched a null string, advance offset one more, to avoid
-- 
1.5.3.2