merge

[irc/quakenet/snircd-patchqueue.git] / match.patch

diff -r 2b442008e1b7 ircd/match.c
--- a/ircd/match.c	Sun Jan 25 21:58:36 2009 +0000
+++ b/ircd/match.c	Sun Jan 25 21:59:17 2009 +0000
@@ -27,6 +27,9 @@
 #include "ircd_string.h"
 #include "ircd_snprintf.h"
 
+#define likely(x) __builtin_expect((x),1)
+#define unlikely(x) __builtin_expect((x),0)
+
 /*
  * mmatch()
  *
@@ -62,53 +65,79 @@
  *
  * @param[in] old_mask One wildcard mask.
  * @param[in] new_mask Another wildcard mask.
+ * @param[in] case_sensitive Indicate case sensitivity
  * @return Zero if \a old_mask is a superset of \a new_mask, non-zero otherwise.
  */
-int mmatch(const char *old_mask, const char *new_mask)
+int _mmatch(const char *old_mask, const char *new_mask, int rfcmatch);
+
+int mmatch(const char *old_mask, const char *new_mask) {
+  return _mmatch(old_mask, new_mask, 1);
+}
+
+__inline__ int _mmatch(const char *old_mask, const char *new_mask, int rfcmatch)
 {
   const char *m = old_mask;
   const char *n = new_mask;
-  const char *ma = m;
-  const char *na = n;
-  int wild = 0;
-  int mq = 0, nq = 0;
+  /* Note that ma / na never point to a character escaped by a backslash. */
+  const char *ma = NULL; // Remembered m for backtracking.
+  const char *na = NULL;
+  int mq = 0, nq = 0; // Is *m / *n escaped?
+  int match;
+
+  if ( m[0] == '*' && m[1] == '\0' ) {
+    return 0;
+  } else if ( n[0] == '*' && n[1] == '\0' ) {
+    return 1;
+  }
 
   while (1)
   {
-    if (*m == '*')
+    if (unlikely(*m == '*'))
     {
+      /* Optimization: Skip redundant *'s */
       while (*m == '*')
         m++;
-      wild = 1;
+      /* And remember this position for backtracking. */
       ma = m;
       na = n;
     }
 
-    if (!*m)
+    if (unlikely(!*m))
     {
       if (!*n)
         return 0;
+      /* This construct speeds up matches of patterns ending with a *
+       * followed by any number of ?. The tricky part is figuring
+       * out whether or not that * was escaped. */
       for (m--; (m > old_mask) && (*m == '?'); m--)
-        ;
-      if ((*m == '*') && (m > old_mask) && (m[-1] != '\\'))
-        return 0;
-      if (!wild)
+        ; /* Skip trailing ?'s */
+      if (*m == '*') {
+        if ((--m >= old_mask) && (*m != '\\'))
+          return 0;
+        /* Now if there's an odd number of backslashes, the for loop
+         * breaks out and we backtrack. */
+        if (!rfcmatch) /* In rfc, backslashes can't be escaped. */
+          for(--m; (m >= old_mask) && (*m == '\\'); m--)
+            if ((--m >= old_mask) && (*m != '\\'))
+              return 0;
+      }
+      if (!ma)
         return 1;
       m = ma;
-
-      /* Added to `mmatch' : Because '\?' and '\*' now is one character: */
-      if ((*na == '\\') && ((na[1] == '*') || (na[1] == '?')))
+      /* skip one escaped character */
+      if (*na == '\\' && (!rfcmatch || na[1] == '*' || na[1] == '?'))
         ++na;
-
       n = ++na;
     }
-    else if (!*n)
+
+    if (unlikely(!*n))
     {
-      while (*m == '*')
+      while (unlikely(*m == '*'))
         m++;
       return (*m != 0);
     }
-    if ((*m == '\\') && ((m[1] == '*') || (m[1] == '?')))
+
+    if (unlikely(*m == '\\' && (!rfcmatch || m[1] == '*' || m[1] == '?')))
     {
       m++;
       mq = 1;
@@ -116,8 +145,7 @@
     else
       mq = 0;
 
-    /* Added to `mmatch' : Because '\?' and '\*' now is one character: */
-    if ((*n == '\\') && ((n[1] == '*') || (n[1] == '?')))
+    if (unlikely(*n == '\\' && (!rfcmatch || n[1] == '*' || n[1] == '?')))
     {
       n++;
       nq = 1;
@@ -126,24 +154,24 @@
       nq = 0;
 
 /*
- * This `if' has been changed compared to match() to do the following:
- * Match when:
- *   old (m)         new (n)         boolean expression
- *    *               any             (*m == '*' && !mq) ||
- *    ?               any except '*'  (*m == '?' && !mq && (*n != '*' || nq)) ||
- * any except * or ?  same as m       (!((*m == '*' || *m == '?') && !mq) &&
- *                                      ToLower(*m) == ToLower(*n) &&
- *                                        !((mq && !nq) || (!mq && nq)))
- *
- * Here `any' also includes \* and \? !
- *
- * After reworking the boolean expressions, we get:
- * (Optimized to use boolean short-circuits, with most frequently occurring
- *  cases upfront (which took 2 hours!)).
+ * There was fancy short-circuit logic here. It got killed. Fuck 2 hours.
+ * It was probably slower than the branches here now. Nobody will notice
+ * in any case.    -- BP
  */
-    if ((*m == '*' && !mq) ||
-        ((!mq || nq) && ToLower(*m) == ToLower(*n)) ||
-        (*m == '?' && !mq && (*n != '*' || nq)))
+    if (unlikely(mq))   {          /* m is quoted, match the exact same, or the
+                          * same character if quoting is irrelevant.  */
+      match = (*m == *n && (nq ||
+                (*n != '*' && *n != '?' && ToUpper(*n) == ToLower(*n))));
+    } else if (unlikely(*m == '?')) { /* m is '?', match anything but unquoted '*' */
+      match = (*n != '*' || nq);
+    } else if (unlikely(*m == '*')) { /* m is '*', match.                          */
+      match=1;
+    } else                /* m is neither quoted nor special           */
+    {
+      match = (ToLower(*m) == ToLower(*n));
+    }
+
+    if (unlikely(match))
     {
       if (*m)
         m++;
@@ -151,15 +179,12 @@
         n++;
     }
     else
-    {
-      if (!wild)
+      if (unlikely(!ma))
         return 1;
       m = ma;
-
-      /* Added to `mmatch' : Because '\?' and '\*' now is one character: */
-      if ((*na == '\\') && ((na[1] == '*') || (na[1] == '?')))
+      /* skip one escaped character */
+      if (unlikely(*na == '\\' && (!rfcmatch || na[1] == '*' || na[1] == '?')))
         ++na;
-
       n = ++na;
     }
   }
@@ -186,8 +211,15 @@
  * @param[in] name String to check against \a mask.
  * @return Zero if \a mask matches \a name, non-zero if no match.
  */
+int _match(const char *mask, const char *name, int rfcmatch);
+
 int match(const char *mask, const char *name)
 {
+  return _match(mask, name, 1);
+}
+
+int _match(const char *mask, const char *name, int rfcmatch)
+{
   const char *m = mask, *n = name;
   const char *m_tmp = mask, *n_tmp = name;
   int star_p;
@@ -205,10 +237,14 @@
       return 1;
     break;
   case '\\':
-    m++;
     /* allow escaping to force capitalization */
-    if (*m++ != *n++)
-      goto backtrack;
+    if (!rfcmatch) {
+      m++;
+      if (*m++ != *n++)
+        goto backtrack;
+    } else {
+      goto fallthrough;
+    }
     break;
   case '*': case '?':
     for (star_p = 0; ; m++) {
@@ -234,6 +270,7 @@
     }
     /* and fall through */
   default:
+  fallthrough:
     if (!*n)
       return *m != '\0';
     if (ToLower(*m) != ToLower(*n))