Bleadperl 8752206e27 breaks a test in MLEHMANN/Coro-5.25.tar.gz #10841
Comments
From @andkOMG, this is 27 days late. Example fail report: http://www.cpantesters.org/cpan/report/b78773ba-eddc-11df-8f7c-368b0c7e6507 Cutting & pasting from there: Can't locate object method "TIESCALAR" via package "Coro::Handle::FH" at /home/sand/.cpan/build/Coro-5.25-XSKMQd/blib/lib/Coro/Handle.pm line 65. git bisect: commit 8752206 [perl #77496] tied gets scalars and globs confused perl -V: Summary of my perl5 (revision 5 version 13 subversion 6) configuration: Characteristics of this binary (from libperl): -- |
From @cpansproutOn Sun Nov 21 12:58:44 2010, andreas.koenig.7os6VVqR@franz.ak.mind.de wrote:
See my recent message to ticket #79302. This would require 27 lines in Coro/Handle.pm to change (everything with Maybe those changes to (un)tie(d) should be reverted. What do Jesse |
|
The RT System itself - Status changed from 'new' to 'open' |
From schmorp@schmorp.deOn Sun, Nov 21, 2010 at 01:17:28PM -0800, Father Chrysostomos via RT <perlbug-followup@perl.org> wrote:
The docs I have didn't seem to indicate what Coro::Handle does wrong, or 27 lines sounds extensive - since this is an incompatible change in perl,
The code seems to match available documentation. If the code is buggy -- |
From @cpansproutOn Sun Nov 21 18:14:49 2010, schmorp@schmorp.de wrote:
Sorry, I was not clear enough. If tie $self is changed to tie *$self and The problem is that in older perls there is no way to see whether a I’m hoping you will consider it a bug, too. See perlfunc: tied VARIABLE That’s not very clear, but it sounds to me that whatever variable is |
From @cpansproutOn Sun Nov 21 18:38:20 2010, sprout wrote:
whatever variable is *named* is what will be looked at. |
From schmorp@schmorp.deOn Sun, Nov 21, 2010 at 06:38:20PM -0800, Father Chrysostomos via RT <perlbug-followup@perl.org> wrote:
Well, whether that is a bug or not doesn't really have anything to do with The important question for me would be how much code breaks, and whether Maybe I am a burned child, but it seems every major perl release breaks -- |
From @cpansproutOn Sun Nov 21 18:50:56 2010, schmorp@schmorp.de wrote:
How about I revert it and add a ‘Use of tie on a handle without * is |
From @cpansproutOn Sun Nov 21 22:25:19 2010, sprout wrote:
How about this patch (plus a perldiag entry)? These three commits will Since this is a run-time check, I could not make it warn once per op. So |
From @cpansproutInline Patchdiff --git a/gv.h b/gv.h
index c61f2e6..be371f1 100644
--- a/gv.h
+++ b/gv.h
@@ -138,6 +138,9 @@ Return the SV from the GV.
#define GVf_IMPORTED_HV 0x40
#define GVf_IMPORTED_CV 0x80
+/* Temporary flag for the tie deprecation warnings. */
+#define GVf_TIEWARNED 0x100
+
#define GvINTRO(gv) (GvFLAGS(gv) & GVf_INTRO)
#define GvINTRO_on(gv) (GvFLAGS(gv) |= GVf_INTRO)
#define GvINTRO_off(gv) (GvFLAGS(gv) &= ~GVf_INTRO)
diff --git a/pp_sys.c b/pp_sys.c
index e068ec6..d27bde6 100644
--- a/pp_sys.c
+++ b/pp_sys.c
@@ -827,6 +827,10 @@ PP(pp_tie)
case SVt_PVGV:
case SVt_PVLV:
if (isGV_with_GP(varsv)) {
+ if (SvFAKE(varsv) && !(GvFLAGS(varsv) & GVf_TIEWARNED)) {
+ deprecate("tie on a handle without *");
+ GvFLAGS(varsv) |= GVf_TIEWARNED;
+ }
methname = "TIEHANDLE";
how = PERL_MAGIC_tiedscalar;
/* For tied filehandles, we apply tiedscalar magic to the IO
@@ -903,8 +907,14 @@ PP(pp_untie)
const char how = (SvTYPE(sv) == SVt_PVHV || SvTYPE(sv) == SVt_PVAV)
? PERL_MAGIC_tied : PERL_MAGIC_tiedscalar;
- if (isGV_with_GP(sv) && !(sv = MUTABLE_SV(GvIOp(sv))))
+ if (isGV_with_GP(sv)) {
+ if (SvFAKE(sv) && !(GvFLAGS(sv) & GVf_TIEWARNED)) {
+ deprecate("untie on a handle without *");
+ GvFLAGS(sv) |= GVf_TIEWARNED;
+ }
+ if (!(sv = MUTABLE_SV(GvIOp(sv))))
RETPUSHYES;
+ }
if ((mg = SvTIED_mg(sv, how))) {
SV * const obj = SvRV(SvTIED_obj(sv, mg));
@@ -941,8 +951,14 @@ PP(pp_tied)
const char how = (SvTYPE(sv) == SVt_PVHV || SvTYPE(sv) == SVt_PVAV)
? PERL_MAGIC_tied : PERL_MAGIC_tiedscalar;
- if (isGV_with_GP(sv) && !(sv = MUTABLE_SV(GvIOp(sv))))
+ if (isGV_with_GP(sv)) {
+ if (SvFAKE(sv) && !(GvFLAGS(sv) & GVf_TIEWARNED)) {
+ deprecate("tied on a handle without *");
+ GvFLAGS(sv) |= GVf_TIEWARNED;
+ }
+ if (!(sv = MUTABLE_SV(GvIOp(sv))))
RETPUSHUNDEF;
+ }
if ((mg = SvTIED_mg(sv, how))) {
SV *osv = SvTIED_obj(sv, mg);
diff --git a/t/op/gmagic.t b/t/op/gmagic.t
index bc8a926..850f50d 100644
--- a/t/op/gmagic.t
+++ b/t/op/gmagic.t
@@ -10,10 +10,11 @@ print "1..24\n";
my $t = 1;
tie my $c => 'Tie::Monitor';
+my $tied_to;
sub ok {
my($ok, $got, $exp, $rexp, $wexp) = @_;
- my($rgot, $wgot) = (tied $c)->init(0);
+ my($rgot, $wgot) = ($tied_to || tied $c)->init(0);
print $ok ? "ok $t\n" : "# expected $exp, got $got\nnot ok $t\n";
++$t;
if ($rexp == $rgot && $wexp == $wgot) {
@@ -56,9 +57,11 @@ ok_string($s, '0', 1, 1);
# Assignment should not ignore magic when the last thing assigned
# was a glob
+$tied_to = tied $c;
$c = *strat;
$s = $c;
ok_string $s, *strat, 1, 1;
+$tied_to = undef;
# A plain *foo should not call get-magic on *foo.
# This method of scalar-tying an immutable glob relies on details of the
diff --git a/t/op/gv.t b/t/op/gv.t
index 862a0cf..e9fde9d 100644
--- a/t/op/gv.t
+++ b/t/op/gv.t
@@ -825,6 +825,7 @@ pass('Can assign strings to typeglobs');
tie my $a, "thrext";
() = "$a"; # do a fetch; now $a holds a glob
eval { *$a = sub{} };
+ eval { $a = undef }; # workaround for untie($handle) bug
untie $a;
eval { $a = "bar" };
::is $a, "bar",
diff --git a/t/op/tie.t b/t/op/tie.t
index 6e52a6e..b68102e 100644
--- a/t/op/tie.t
+++ b/t/op/tie.t
@@ -939,3 +939,26 @@ sub IO::File::TIEARRAY {
fileno FOO; tie @a, "FOO"
EXPECT
Can't locate object method "TIEARRAY" via package "FOO" at - line 5.
+########
+
+# Deprecation warnings for tie $handle
+
+use warnings 'deprecated';
+$SIG{__WARN__} = sub { $w = shift };
+$handle = *foo;
+eval { tie $handle, "" };
+print $w =~ /^Use of tie on a handle without \* is deprecated/
+ ? "ok tie\n" : "$w\n";
+$handle = *bar;
+tied $handle;
+print $w =~ /^Use of tied on a handle without \* is deprecated/
+ ? "ok tied\n" : "$w\n";
+$handle = *baz;
+untie $handle;
+print $w =~ /^Use of untie on a handle without \* is deprecated/
+ ? "ok untie\n" : "$w\n";
+
+EXPECT
+ok tie
+ok tied
+ok untie |
From @andk
> +print $w =~ /^Use of tie on a handle without \* is deprecated/ The problem with any such approach is that it is completely new to me In the past it was completely idiomatic perl to write foo(*STDOUT); sub foo { So this style is going to be verboten? I was wondering why only Marc Lehmann and Abe Timmerman have triggered I don't mind the change if it helps fixing otherwise unfixable bugs but Or what am I missing? -- |
From @LeontOn Mon, Nov 29, 2010 at 10:51 PM, Andreas J. Koenig
It is completely new to me that the argument to tie may be written
print isn't polymorphic the way tie is, it doesn't need any extra
I have used it one or two modules, but I have always used the sigiled
Where are the docs that say it is allowed to omit them? When I see Leon |
From @TuxOn Mon, 29 Nov 2010 23:47:22 +0100, Leon Timmermans <fawaka@gmail.com>
http://search.cpan.org/dist/Text-OutputFilter/ is of my hand, ant ties
-- |
From @demerphqOn 29 November 2010 23:47, Leon Timmermans <fawaka@gmail.com> wrote:
I agree. And perltie clearly shows a * sigil on the tie example for filehandles. http://perldoc.perl.org/perltie.html#Tying-FileHandles <quote> 1. tie(*FOO,'Shout'); yves -- |
From @andk
> On Mon, Nov 29, 2010 at 10:51 PM, Andreas J. Koenig
> It is completely new to me that the argument to tie may be written Good to hear different experiences.
> print isn't polymorphic the way tie is, it doesn't need any extra While I understand the argument I find this isn't documented.
> I have used it one or two modules, but I have always used the sigiled Please, just read the ticket. The modules that use it without the star MLEHMANN/Coro-5.25.tar.gz Not three especially unexperienced people.
> Where are the docs that say it is allowed to omit them? The word VARIABLE in documentations does not specify which sigil has to > When I see Fine with me but I still think the documentation ought to make this -- |
From @demerphqOn 30 November 2010 07:08, Andreas J. Koenig
No. But ones with a very long history. Maybe this is a hold over of
I am not convinced this is a sound argument. Where _else_ do we
As I said elsewhere perltie doesnt show any use of tieing a filehandle Sure it should be *better* documented, but I think one *can* argue cheers, -- |
From @ikegamiOn Tue, Nov 30, 2010 at 4:22 AM, demerphq <demerphq@gmail.com> wrote:
Maybe, but not very convincingly. It's not like you can change tie(*FH, ...) to my $fh = FH; # XXX strict error. someone could naturally assume one has to do one the following, since it's my $fh = *FH; my $fh = \*FH; And given that the first currently works... I'm not saying it shouldn't be changed. I'm saying it's a stretch to say - Eric |
From @demerphqOn 30 November 2010 20:49, Eric Brine <ikegami@adaelis.com> wrote:
Umm... I dont know about that. That would tie the variable $fh to some If it works, its an accident. This is a container/value problem. Tie acts on a container. Not on a
Again no, this is wrong. for the same reasons i stated above. (And it
Well, in perltie the only tied filehandle is written as tie *HANDLE, ... That is, WITH the * sigil. So, while you might be right in saying that tie(*$fh, ...) is not Certainly it is better documented than tie FH, ....; Of which i can find no example at all. cheers, -- |
From @LeontOn Tue, Nov 30, 2010 at 8:49 PM, Eric Brine <ikegami@adaelis.com> wrote:
It's ordinary syntax for a dereference of a globref. That's perfectly Leon |
From @cpansproutOn Sun Nov 28 12:15:12 2010, sprout wrote:
Now done as: |
|
@cpansprout - Status changed from 'open' to 'resolved' |
Migrated from rt.perl.org#79528 (status was 'resolved')
Searchable as RT79528$
The text was updated successfully, but these errors were encountered: