Re: [bug #36567] grep -i (case-insensitive) is broken with UTF8

[Paul Eggert]

On 06/01/2012 08:18 AM, Jim Meyering wrote:
> I've implemented the above, and have begun testing.

Thanks.  One thing I noticed when looking at the code is
that the mbtolower API causes the compiler to refuse to
put the 'size' local into a register.  The following patch
fixes this inefficiency.  This undoubtedly interacts with
what you're doing so I'm just putting it into this email
now as a heads-up.

>From 3cd025a7b2700799e74457bd9c6d6d6e9e345a19 Mon Sep 17 00:00:00 2001
From: Paul Eggert <address@hidden>
Date: Fri, 1 Jun 2012 08:59:50 -0700
Subject: [PATCH] grep: tune by allowing compiler to put size in register

* src/dfasearch.c (EGexecute):
* src/kwsearch.c (Fcompile, Fexecute):
Pass address of otherwise-unused variable to mbtolower, so that
the compiler can keep the often-used size variable in a register.
---
 src/dfasearch.c |    4 +++-
 src/kwsearch.c  |    8 +++++---
 2 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/src/dfasearch.c b/src/dfasearch.c
index bd09aa6..b7c0da7 100644
--- a/src/dfasearch.c
+++ b/src/dfasearch.c
@@ -223,10 +223,12 @@ EGexecute (char const *buf, size_t size, size_t 
*match_size,
         {
           /* mbtolower adds a NUL byte at the end.  That will provide
              space for the sentinel byte dfaexec may add.  */
-          char *case_buf = mbtolower (buf, &size);
+          size_t case_size = size;
+          char *case_buf = mbtolower (buf, &case_size);
           if (start_ptr)
             start_ptr = case_buf + (start_ptr - buf);
           buf = case_buf;
+          size = case_size;
         }
     }
 
diff --git a/src/kwsearch.c b/src/kwsearch.c
index f1a802e..829fe12 100644
--- a/src/kwsearch.c
+++ b/src/kwsearch.c
@@ -33,10 +33,10 @@ void
 Fcompile (char const *pattern, size_t size)
 {
   char const *err;
-  size_t psize = size;
   char const *pat = (match_icase && MB_CUR_MAX > 1
-                     ? mbtolower (pattern, &psize)
+                     ? mbtolower (pattern, &size)
                      : pattern);
+  size_t psize = size;
 
   kwsinit (&kwset);
 
@@ -87,10 +87,12 @@ Fexecute (char const *buf, size_t size, size_t *match_size,
     {
       if (match_icase)
         {
-          char *case_buf = mbtolower (buf, &size);
+          size_t case_size = size;
+          char *case_buf = mbtolower (buf, &case_size);
           if (start_ptr)
             start_ptr = case_buf + (start_ptr - buf);
           buf = case_buf;
+          size = case_size;
         }
     }
 
-- 
1.7.6.5