Junctions cause crash / infinite loop in &put and related routines and methods

[p6rt]

Migrated from rt.perl.org#132549 (status was 'resolved')

Searchable as RT132549$

[p6rt]

From @briandfoy

This comes from an answer to a Perl 6 question on Stackoverflow that
showed a different bug​:

  https://stackoverflow.com/q/45527881/2766176

With put() it does not and gives a strange error​:

  $ perl6 -e 'put any( 1, 3, 7 ) '
  This type cannot unbox to a native string​: P6opaque, Junction
  in block <unit> at -e line 1

  $ perl6 -e 'put( any( 1, 3, 7 ) )'
  This type cannot unbox to a native string​: P6opaque, Junction
  in block <unit> at -e line 1

With say it has no problem​:

  $ perl6 -e 'say any( 1, 3, 7 )'
  any(1, 3, 7)
  $ perl6 -e 'put any( 1, 3, 7 ).gist'
  any(1, 3, 7)

I'd expect anything that can .gist would be able to .Str.

This is Rakudo 2017.10 on macOS 10.13.

[p6rt]

From @zoffixznet

On Fri, 08 Dec 2017 08​:28​:32 -0800, comdog wrote​:

This comes from an answer to a Perl 6 question on Stackoverflow that
showed a different bug​:

https://stackoverflow.com/q/45527881/2766176

With put() it does not and gives a strange error​:

I guess jnthn++ gets a score point for predicting[^1] this would happen​:

  <jnthn> Hmmm...not too keen on the Junction.Str patch
  <jnthn> Anything that (quite reasonably) does nqp​::unbox_s($foo.Str) is now
  going to (quite rightly) explode
  <timotimo> clearly we have to build UNBOXABLE_STR :)
  <jnthn> No, it can just explode, and then I'll point people at this commit. :P

If put() were made to work here, I'd expect it to junct and be equivalent to `put 1`, `put 3`, `put 7` executed in random order, but the OP in that SO has an entirely different expectation.

I'd expect anything that can .gist would be able to .Str.

Side note on that​: only the opposite is true. A basic class would .gist to its name, while its .Str would be an empty string accompanied by a warning. Also, some things, like self-referential structures or non-lazy infinite iterables, would take an infinite time to .Str, while getting .gisted is no trouble.

[1] https://irclog.perlgeek.de/perl6-dev/2017-08-09#i_14992203

[p6rt]

The RT System itself - Status changed from 'new' to 'open'

[p6rt]

From @lizmat

On 8 Dec 2017, at 19​:21, Zoffix Znet via RT <perl6-bugs-followup@​perl.org> wrote​:

On Fri, 08 Dec 2017 08​:28​:32 -0800, comdog wrote​:

This comes from an answer to a Perl 6 question on Stackoverflow that
showed a different bug​:

https://stackoverflow.com/q/45527881/2766176

With put() it does not and gives a strange error​:

I guess jnthn++ gets a score point for predicting[^1] this would happen​:

<jnthn> Hmmm...not too keen on the Junction.Str patch
<jnthn> Anything that (quite reasonably) does nqp​::unbox_s($foo.Str) is now
going to (quite rightly) explode
<timotimo> clearly we have to build UNBOXABLE_STR :)
<jnthn> No, it can just explode, and then I'll point people at this commit. :P

Which would be me.

And as far as I recall atm, that was in response to making​:

  $ 6 'dd "foo" ~ any(1,3,5) ~ "bar"'
  any("foo1bar", "foo3bar", "foo5bar”)

work. If that shouldn’t work, or work differently, it can be ripped out / replaced. If that should work, then we need to look at fixing -put-.

If put() were made to work here, I'd expect it to junct and be equivalent to `put 1`, `put 3`, `put 7` executed in random order, but the OP in that SO has an entirely different expectation.

FWIW, I would also expect it to junct.

[p6rt]

From @zoffixznet

On Sat, 09 Dec 2017 13​:19​:23 -0800, elizabeth wrote​:

If that shouldn’t work, or work differently, it can be ripped
out / replaced. If that should work, then we need to look at fixing
-put-.

IMO it should work, considering you can't nqp​::unbox_n/nqp​::unbox_i Junctions either, yet we don't have Junction-related explosions with numeric stuff.

I also found another issue​: IO​::Handle.print/put infiniloop in dispatch when given a Junction.

FWIW, I would also expect it to junct.

I started a fix down that path, but it kinda looks gross and dirty. Similar thing would be needed
in src/core/io_operators.pm for subroutine versions. The Junctions go through the **@​foo slurpy candidates
and the explosions then happen inside optimizations, where we assume x.Str gives an unboxable Str.

Is there a better way? Here's my thing​:

2017.11.50 zoffix@​VirtualBox~/CPANPRC/rakudo (master)$ gd

Inline Patch

diff --git a/src/core/IO/Handle.pm b/src/core/IO/Handle.pm
index 55ff911..0303047 100644
--- a/src/core/IO/Handle.pm
+++ b/src/core/IO/Handle.pm
@@ -652,6 +652,11 @@ my class IO::Handle {
     multi method print(IO::Handle:D: **@list is raw --> True) { # is raw gives List, which is cheaper
         self.print(@list.join);
     }
+    multi method print(IO::Handle:D: Junction \j --> True) {
+        # junct the Junction. Without this candidate, we'd go through the
+        # **@list slurpy candidate and dispatch infiniloop.
+        -> Any \el { self.print: el }(j)
+    }
 
     proto method put(|) {*}
     multi method put(IO::Handle:D: Str:D \x --> True) {
@@ -662,6 +667,11 @@ my class IO::Handle {
     multi method put(IO::Handle:D: **@list is raw --> True) { # is raw gives List, which is cheaper
         self.put(@list.join);
     }
+    multi method put(IO::Handle:D: Junction \j --> True) {
+        # junct the Junction. Without this candidate, we'd go through the
+        # **@list slurpy candidate and dispatch infiniloop.
+        -> Any \el { self.put: el }(j)
+    }
 
     multi method say(IO::Handle:D: Str:D $x --> True) {
         $!decoder or die X::IO::BinaryMode.new(:trying<say>);

[p6rt]

From @zoffixznet

On Sat, 09 Dec 2017 15​:26​:25 -0800, cpan@​zoffix.com wrote​:

On Sat, 09 Dec 2017 13​:19​:23 -0800, elizabeth wrote​:

If that shouldn’t work, or work differently, it can be ripped
out / replaced. If that should work, then we need to look at fixing
-put-.

IMO it should work, considering you can't nqp​::unbox_n/nqp​::unbox_i
Junctions either, yet we don't have Junction-related explosions with
numeric stuff.

I also found another issue​: IO​::Handle.print/put infiniloop in
dispatch when given a Junction.

FWIW, I would also expect it to junct.

I started a fix down that path, but it kinda looks gross and dirty.
Similar thing would be needed
in src/core/io_operators.pm for subroutine versions. The Junctions go
through the **@​foo slurpy candidates
and the explosions then happen inside optimizations, where we assume
x.Str gives an unboxable Str.

Is there a better way? Here's my thing​:

2017.11.50 zoffix@​VirtualBox~/CPANPRC/rakudo (master)$ gd
diff --git a/src/core/IO/Handle.pm b/src/core/IO/Handle.pm
index 55ff911..0303047 100644
--- a/src/core/IO/Handle.pm
+++ b/src/core/IO/Handle.pm
@​@​ -652,6 +652,11 @​@​ my class IO​::Handle {
multi method print(IO​::Handle​:D​: **@​list is raw --> True) { # is
raw gives List, which is cheaper
self.print(@​list.join);
}
+ multi method print(IO​::Handle​:D​: Junction \j --> True) {
+ # junct the Junction. Without this candidate, we'd go through
the
+ # **@​list slurpy candidate and dispatch infiniloop.
+ -> Any \el { self.print​: el }(j)
+ }

proto method put(|) {*}
multi method put(IO​::Handle​:D​: Str​:D \x --> True) {
@​@​ -662,6 +667,11 @​@​ my class IO​::Handle {
multi method put(IO​::Handle​:D​: **@​list is raw --> True) { # is raw
gives List, which is cheaper
self.put(@​list.join);
}
+ multi method put(IO​::Handle​:D​: Junction \j --> True) {
+ # junct the Junction. Without this candidate, we'd go through
the
+ # **@​list slurpy candidate and dispatch infiniloop.
+ -> Any \el { self.put​: el }(j)
+ }

multi method say(IO​::Handle​:D​: Str​:D $x --> True) {
$!decoder or die X​::IO​::BinaryMode.new(​:trying<say>);

Thank you for the report. lizmat++ fixed this.

Fix​: rakudo/rakudo@8155c4b885
  rakudo/rakudo@9de4a60efe
  rakudo/rakudo@07616effd1
Test​: Raku/roast@97ceb93f40214cb7c

[p6rt]

@zoffixznet - Status changed from 'open' to 'resolved'