Make $$ writable, but still magical

This commit makes $$ writable again, as it was in 5.6, while preserv-
ing the magical pid-fetching added recently (post-5.14.0) by com-
mit 0e219455.

It does this by following Aristotle Pagaltzis’ brilliant suggestion in
<20110609145148.GD8471@klangraum.plasmasturm.org>; namely, to store
the PID in magic when $$ is written to, so that get-magic can detect
whether a fork() has occurred and reset $$ accordingly.  This makes it
seem as though the fork() code sets $$ itself (which it used to before
0e219455), while even working when C code outside of perl’s control
calls fork().

This restores compatibility with DBIx::Connector and PPerl.

4 files changed, 50 insertions, 14 deletions

diff --git a/gv.c b/gv.c
index f8de97f813..9bb428d40b 100644
--- a/gv.c
+++ b/gv.c
@@ -1470,9 +1470,6 @@ Perl_gv_fetchpvn_flags(pTHX_ const char *nambeg, STRLEN full_len, I32 flags,
 #endif
 	    goto magicalize;
 
-	case '$':		/* $$ */
-	    SvREADONLY_on(GvSVn(gv));
-	    goto magicalize;
 	case '!':		/* $! */
 	    GvMULTI_on(gv);
 	    /* If %! has been used, automatically load Errno.pm. */
@@ -1544,6 +1541,7 @@ Perl_gv_fetchpvn_flags(pTHX_ const char *nambeg, STRLEN full_len, I32 flags,
 	case '>':		/* $> */
 	case '\\':		/* $\ */
 	case '/':		/* $/ */
+	case '$':		/* $$ */
 	case '\001':	/* $^A */
 	case '\003':	/* $^C */
 	case '\004':	/* $^D */
diff --git a/mg.c b/mg.c
index 86f1eb6e4d..1bdf5c4cda 100644
--- a/mg.c
+++ b/mg.c
@@ -1080,7 +1080,13 @@ Perl_magic_get(pTHX_ SV *sv, MAGIC *mg)
 	    sv_copypv(sv, PL_ors_sv);
 	break;
     case '$': /* $$ */
-	sv_setiv(sv, (IV)PerlProc_getpid());
+	{
+	    IV const pid = (IV)PerlProc_getpid();
+	    if (isGV(mg->mg_obj) || SvIV(mg->mg_obj) != pid)
+		/* never set manually, or at least not since last fork */
+		sv_setiv(sv, pid);
+	    /* else a value has been assigned manually, so do nothing */
+	}
 	break;
 
     case '!':
@@ -2881,6 +2887,17 @@ Perl_magic_set(pTHX_ SV *sv, MAGIC *mg)
     case ':':
 	PL_chopset = SvPV_force(sv,len);
 	break;
+    case '$': /* $$ */
+	/* Store the pid in mg->mg_obj so we can tell when a fork has
+	   occurred.  mg->mg_obj points to *$ by default, so clear it. */
+	if (isGV(mg->mg_obj)) {
+	    if (mg->mg_flags & MGf_REFCOUNTED) /* probably never true */
+		SvREFCNT_dec(mg->mg_obj);
+	    mg->mg_flags |= MGf_REFCOUNTED;
+	    mg->mg_obj = newSViv((IV)PerlProc_getpid());
+	}
+	else sv_setiv(mg->mg_obj, (IV)PerlProc_getpid());
+	break;
     case '0':
 	LOCK_DOLLARZERO_MUTEX;
 #ifdef HAS_SETPROCTITLE
diff --git a/pod/perldelta.pod b/pod/perldelta.pod
index 9f718c32ed..86841579b6 100644
--- a/pod/perldelta.pod
+++ b/pod/perldelta.pod
@@ -42,6 +42,14 @@ here, but most should go in the L</Performance Enhancements> section.
 The C<CORE::> prefix can now be used on keywords enabled by
 L<feature.pm|feature>, even outside the scope of C<use feature>.
 
+=head2 C<$$> can be assigned to
+
+C<$$> was made read-only in Perl 5.8.0.  But only sometimes: C<local $$>
+would make it writable again.  Some CPAN modules were using C<local $$> or
+XS code to bypass the read-only check, so there is no reason to keep C<$$>
+read-only.  (This change also allowed a bug to be fixed while maintaining
+backward compatibility.)
+
 =head1 Security
 
 XXX Any security-related notices go here.  In particular, any security
@@ -54,13 +62,6 @@ L</Selected Bug Fixes> section.
 
 [ List each incompatible change as a =head2 entry ]
 
-=head2 C<$$> no longer caches PID
-
-Previously, if one embeds Perl or uses XS and calls fork(3) from C, Perls
-notion of C<$$> could go out of sync with what getpid() returns. By always
-fetching the value of C<$$> via getpid(), this potential bug is eliminated.
-Code that depends on the caching behavior will break.
-
 =head1 Deprecations
 
 XXX Any deprecated features, syntax, modules etc. should be listed here.
@@ -987,6 +988,12 @@ fixed [RT #85026].
 
 =item *
 
+Previously, if one embeds Perl or uses XS and calls fork(3) from C, Perl's
+notion of C<$$> could go out of sync with what getpid() returns.  By always
+fetching the value of C<$$> via getpid(), this potential bug is eliminated.
+
+=item *
+
 Passing the same constant subroutine to both C<index> and C<formline> no
 longer causes one or the other to fail [RT #89218].
 
diff --git a/t/op/magic.t b/t/op/magic.t
index 31286877ef..585cc40cb5 100644
--- a/t/op/magic.t
+++ b/t/op/magic.t
@@ -12,7 +12,7 @@ BEGIN {
 use warnings;
 use Config;
 
-plan (tests => 87);
+plan (tests => 88);
 
 $Is_MSWin32  = $^O eq 'MSWin32';
 $Is_NetWare  = $^O eq 'NetWare';
@@ -169,8 +169,22 @@ eval { die "foo\n" };
 is $@, "foo\n";
 
 cmp_ok($$, '>', 0);
-eval { $$++ };
-like ($@, qr/^Modification of a read-only value attempted/);
+eval { $$ = 42 };
+is $$, 42, '$$ can be modified';
+SKIP: {
+    skip "no fork", 1 unless $Config{d_fork};
+    (my $kidpid = open my $fh, "-|") // skip "cannot fork: $!", 1;
+    if($kidpid) { # parent
+	my $kiddollars = <$fh>;
+	close $fh or die "cannot close pipe from kid proc: $!";
+	is $kiddollars, $kidpid, '$$ is reset on fork';
+    }
+    else { # child
+	print $$;
+	$::NO_ENDING = 1; # silence "Looks like you only ran..."
+	exit;
+    }
+}
 
 # $^X and $0
 {