This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Regexp optimizations
authorIlya Zakharevich <ilya@math.ohio-state.edu>
Tue, 4 Feb 1997 11:02:10 +0000 (06:02 -0500)
committerChip Salzenberg <chip@atlantic.net>
Mon, 10 Feb 1997 19:29:00 +0000 (07:29 +1200)
Subject: Re: Regexp optimizations

p5p-msgid: <199702051450.JAA13439@rio.atlantic.net>
private-msgid: <199702041102.GAA24805@monk.mps.ohio-state.edu>

regcomp.c
regexec.c

index d736c18..9e39afe 100644 (file)
--- a/regcomp.c
+++ b/regcomp.c
@@ -145,6 +145,13 @@ PMOP* pm;
     I32 minlen = 0;
     I32 sawplus = 0;
     I32 sawopen = 0;
     I32 minlen = 0;
     I32 sawplus = 0;
     I32 sawopen = 0;
+#define MAX_REPEAT_DEPTH 12
+    struct {
+       char *opcode;
+       I32 count;
+    } repeat_stack[MAX_REPEAT_DEPTH];
+    I32 repeat_depth = 0;
+    I32 repeat_count = 1;      /* We start unmultiplied. */
 
     if (exp == NULL)
        croak("NULL regexp argument");
 
     if (exp == NULL)
        croak("NULL regexp argument");
@@ -289,9 +296,9 @@ PMOP* pm;
                char *t;
 
                first = scan;
                char *t;
 
                first = scan;
-               while (OP(t = regnext(scan)) == CLOSE)
+               while ((t = regnext(scan)) && OP(t) == CLOSE)
                    scan = t;
                    scan = t;
-               minlen += *OPERAND(first);
+               minlen += *OPERAND(first) * repeat_count;
                if (curback - backish == len) {
                    sv_catpvn(longish, OPERAND(first)+1,
                        *OPERAND(first));
                if (curback - backish == len) {
                    sv_catpvn(longish, OPERAND(first)+1,
                        *OPERAND(first));
@@ -310,22 +317,57 @@ PMOP* pm;
                    curback += *OPERAND(first);
            }
            else if (strchr(varies,OP(scan))) {
                    curback += *OPERAND(first);
            }
            else if (strchr(varies,OP(scan))) {
-               curback = -30000;
+               int tcount;
+               char *next;
+
+               if (repeat_depth < MAX_REPEAT_DEPTH
+                   && ((OP(scan) == PLUS
+                        && (tcount = 1)
+                        && (next = NEXTOPER(scan)))
+                       || (regkind[(U8)OP(scan)] == CURLY
+                           && (tcount = ARG1(scan))
+                           && (next = NEXTOPER(scan)+4))))
+               {
+                   /* We treat (abc)+ as (abc)(abc)*. */
+
+                   /* Mark the place to return back. */
+                   repeat_stack[repeat_depth].opcode = regnext(scan);
+                   repeat_stack[repeat_depth].count = repeat_count;
+                   repeat_depth++;
+                   repeat_count *= tcount;
+
+                   /* Go deeper: */
+                   scan = next;
+                   continue;
+               }
+               else {
+                   curback = -30000;
+                   len = 0;
+                   if (SvCUR(longish) > SvCUR(longest)) {
+                       sv_setsv(longest,longish);
+                       backest = backish;
+                   }
+                   sv_setpvn(longish,"",0);
+               }
+           }
+           else if (strchr(simple,OP(scan))) {
+               curback++;
+               minlen += repeat_count;
                len = 0;
                if (SvCUR(longish) > SvCUR(longest)) {
                    sv_setsv(longest,longish);
                    backest = backish;
                }
                sv_setpvn(longish,"",0);
                len = 0;
                if (SvCUR(longish) > SvCUR(longest)) {
                    sv_setsv(longest,longish);
                    backest = backish;
                }
                sv_setpvn(longish,"",0);
-               if (OP(scan) == PLUS && strchr(simple,OP(NEXTOPER(scan))))
-                   minlen++;
-               else if (regkind[(U8)OP(scan)] == CURLY &&
-                 strchr(simple,OP(NEXTOPER(scan)+4)))
-                   minlen += ARG1(scan);
            }
            }
-           else if (strchr(simple,OP(scan))) {
-               curback++;
-               minlen++;
+           scan = regnext(scan);
+           if (!scan) {                /* Go up PLUS or CURLY. */
+               if (!repeat_depth--)
+                   croak("panic: re scan");
+               scan = repeat_stack[repeat_depth].opcode;
+               repeat_count = repeat_stack[repeat_depth].count;
+               /* Need to submit the longest string found: */
+               curback = -30000;
                len = 0;
                if (SvCUR(longish) > SvCUR(longest)) {
                    sv_setsv(longest,longish);
                len = 0;
                if (SvCUR(longish) > SvCUR(longest)) {
                    sv_setsv(longest,longish);
@@ -333,12 +375,11 @@ PMOP* pm;
                }
                sv_setpvn(longish,"",0);
            }
                }
                sv_setpvn(longish,"",0);
            }
-           scan = regnext(scan);
        }
 
        /* Prefer earlier on tie, unless we can tail match latter */
 
        }
 
        /* Prefer earlier on tie, unless we can tail match latter */
 
-       if (SvCUR(longish) + (regkind[(U8)OP(first)] == EOL)
+       if (SvCUR(longish) + (first && regkind[(U8)OP(first)] == EOL)
                > SvCUR(longest))
        {
            sv_setsv(longest,longish);
                > SvCUR(longest))
        {
            sv_setsv(longest,longish);
@@ -357,11 +398,12 @@ PMOP* pm;
            if (backest < 0)
                backest = -1;
            r->regback = backest;
            if (backest < 0)
                backest = -1;
            r->regback = backest;
-           if (SvCUR(longest) > !(sawstudy || regkind[(U8)OP(first)] == EOL))
+           if (SvCUR(longest) > !(sawstudy || 
+                                  (first && regkind[(U8)OP(first)] == EOL)))
                fbm_compile(r->regmust);
            (void)SvUPGRADE(r->regmust, SVt_PVBM);
            BmUSEFUL(r->regmust) = 100;
                fbm_compile(r->regmust);
            (void)SvUPGRADE(r->regmust, SVt_PVBM);
            BmUSEFUL(r->regmust) = 100;
-           if (regkind[(U8)OP(first)] == EOL && SvCUR(longish))
+           if (first && regkind[(U8)OP(first)] == EOL && SvCUR(longish))
                SvTAIL_on(r->regmust);
        }
        else {
                SvTAIL_on(r->regmust);
        }
        else {
index c55eb97..12908b2 100644 (file)
--- a/regexec.c
+++ b/regexec.c
@@ -290,7 +290,7 @@ I32 safebase;       /* no need to remember string in subbase */
                s++;
            }
        }
                s++;
            }
        }
-       else if (SvPOK(prog->regstart) == 3) {
+       else if (SvTYPE(prog->regstart) == SVt_PVBM) {
            /* We know what string it must start with. */
            while ((s = fbm_instr((unsigned char*)s,
              (unsigned char*)strend, prog->regstart)) != NULL)
            /* We know what string it must start with. */
            while ((s = fbm_instr((unsigned char*)s,
              (unsigned char*)strend, prog->regstart)) != NULL)
@@ -300,7 +300,7 @@ I32 safebase;       /* no need to remember string in subbase */
                s++;
            }
        }
                s++;
            }
        }
-       else {
+       else {                          /* Optimized fbm_instr: */
            c = SvPVX(prog->regstart);
            while ((s = ninstr(s, strend, c, c + SvCUR(prog->regstart))) != NULL)
            {
            c = SvPVX(prog->regstart);
            while ((s = ninstr(s, strend, c, c + SvCUR(prog->regstart))) != NULL)
            {