stop $foo =~ /(bar)/g skipping copy

Normally in the presence of captures, a successful regex execution
makes a copy of the matched string, so that $1 et al give the right
value even if the original string is changed; i.e.

    $foo =~ /(123)/g;
    $foo = "bar";
    is("$1", "123");

Until now that test would fail, because perl used to skip the copy for
the scalar /(...)/g case (but not the C<$&; //g> case). This was to
avoid a huge slowdown in code like the following:

    $x = 'x' x 1_000_000;
    1 while $x =~ /(.)/g;

which would otherwise end up copying a 1Mb string a million times.

Now that (with the last commit but one) we copy only the required
substring of the original string (a 1-byte substring in the above
example), we can remove this fast-but-incorrect hack.

diff --git a/pp_hot.c b/pp_hot.c
index 91958ac..6530ae5 100644 (file)
--- a/pp_hot.c
+++ b/pp_hot.c
@@ -1319,12 +1319,7 @@ PP(pp_match)
            }
        }
     }
-    /* XXX: comment out !global get safe $1 vars after a
-       match, BUT be aware that this leads to dramatic slowdowns on
-       /g matches against large strings.  So far a solution to this problem
-       appears to be quite tricky.
-       Test for the unsafe vars are TODO for now. */
-    if (       (!global && RX_NPARENS(rx))
+    if (       RX_NPARENS(rx)
             || PL_sawampersand
             || SvTEMP(TARG)
             || SvAMAGIC(TARG)

diff --git a/t/re/pat_advanced.t b/t/re/pat_advanced.t
index 6692e1c..05cc191 100644 (file)
--- a/t/re/pat_advanced.t
+++ b/t/re/pat_advanced.t
@@ -1660,7 +1660,6 @@ $x='123';
 print ">$1<\n";
 EOP
 
-        local $::TODO = 'RT #86042';
         fresh_perl_is(<<'EOP', ">abc<\n", {}, 'no mention of $&');
 my $x; 
 ($x='abc')=~/(abc)/g; 

diff --git a/t/re/pat_psycho.t b/t/re/pat_psycho.t
index 0880242..0433760 100644 (file)
--- a/t/re/pat_psycho.t
+++ b/t/re/pat_psycho.t
@@ -25,7 +25,7 @@ BEGIN {
 
 
 skip_all('$PERL_SKIP_PSYCHO_TEST set') if $ENV{PERL_SKIP_PSYCHO_TEST};
-plan tests => 11;  # Update this when adding/deleting tests.
+plan tests => 15;  # Update this when adding/deleting tests.
 
 run_tests() unless caller;
 
@@ -160,6 +160,49 @@ sub run_tests {
         }
         ok($ok, $msg);
     }
+
+
+    {
+       # these bits of test code used to run quadratically. If we break
+       # anything, they'll start to take minutes to run, rather than
+       # seconds. We don't actually measure times or set alarms, since
+       # that tends to be very fragile and prone to false positives.
+       # Instead, just hope that if someone is messing with
+       # performance-related code, they'll re-run the test suite and
+       # notice it suddenly takes a lot longer.
+
+       my $x;
+
+       $x = 'x' x 1_000_000;
+       1 while $x =~ /(.)/g;
+       pass "ascii =~ /(.)/";
+
+       {
+           local ${^UTF8CACHE} = 1; # defeat debugging
+           $x = "\x{100}" x 1_000_000;
+           1 while $x =~ /(.)/g;
+           pass "utf8 =~ /(.)/";
+       }
+
+       # run these in separate processes, since they set $&
+
+        fresh_perl_is(<<'EOF', "ok\n", {}, 'ascii =~ /(.)/, mention $&');
+$&;
+$x = 'x' x 1_000_000;
+1 while $x =~ /(.)/g;
+print "ok\n";
+EOF
+
+        fresh_perl_is(<<'EOF', "ok\n", {}, 'utf8 =~ /(.)/, mention $&');
+$&;
+local ${^UTF8CACHE} = 1; # defeat debugging
+$x = "\x{100}" x 1_000_000;
+1 while $x =~ /(.)/g;
+print "ok\n";
+EOF
+
+
+    }
 } # End of sub run_tests
 
 1;