Re^2: unexpected modify hash in a distance with grep { $_ } (inconsistent behaviour of aliasing)

[LanX]

It seems to be a general issue that accessing an alias is sometimes triggering autovivification.

Didn't know that. Smells like bug.

 
#!/usr/bin/perl 

use strict;
use warnings;
use Data::Dumper;

my $h = {
          'a' => 1,
          'b' => 2,
          'c' => 3,
        };

warn "x: ",$_  for $$h{x};              # autovivifiaction via alias

warn Dumper [ keys %$h ];



tst($$h{y});                            # no autovivification

warn Dumper [ keys %$h ];

sub tst {
    warn "y: ",$_[0] ;
}

C:/Perl_524/bin\perl.exe  d:/tmp/pm/grep_autovivify.pl 
Use of uninitialized value $_ in warn at d:/tmp/pm/grep_autovivify.pl 
+line 13.
x:  at d:/tmp/pm/grep_autovivify.pl line 13.
$VAR1 = [
          'c',
          'x',
          'b',
          'a'
        ];
Use of uninitialized value $_[0] in warn at d:/tmp/pm/grep_autovivify.
+pl line 24.
y:  at d:/tmp/pm/grep_autovivify.pl line 24.
$VAR1 = [
          'c',
          'x',
          'b',
          'a'
        ];

[dave_the_m]

grep (and for) provide lvalue context for their args, since the expectation is that when aliased to $_, the $_ variable (and thus the arg) could be modified by the code in the code block. This lvalue context is what causes the autovivification action to be compiled in. The 'x' element of %$h is created at run-time when the arg list for grep is being assembled, but before grep itself is invoked (i.e. before the code block is called for the first time).

However, function call arguments which are hash subscripts are special-cased. Although function args are technically always in lvalue context (e.g. the sub about to be called might do sub f { $_[0]++ }, who knows!), perl tries to avoid auto-vivification by deferring hash lookup. Instead, if you do foo($h{$k}) then rather than looking up the hash value and passing it as an arg, perl creates a special proxy object (PVLV) which holds pointers to $h and the key and passes that instead. For example:

$ perl -MDevel::Peek -e'sub f { Dump $_[0] } f($h{akey})'
SV = PVLV(0xd28a78) at 0xcb2658
  REFCNT = 1
  FLAGS = (GMG,SMG)
  IV = 0
  NV = 0
  PV = 0
  MAGIC = 0xce9828
    MG_VIRTUAL = &PL_vtbl_defelem
    MG_TYPE = PERL_MAGIC_defelem(y)
    MG_FLAGS = 0x02
      REFCOUNTED
    MG_OBJ = 0xcb2820
    SV = PV(0xcb2ed8) at 0xcb2820
      REFCNT = 1
      FLAGS = (POK,pPOK)
      PV = 0xce97f8 "akey"\0
      CUR = 4
      LEN = 10
  TYPE = y
  TARGOFF = 0
  TARGLEN = 1
  TARG = 0xce1a80
  FLAGS = 0
  SV = PVHV(0xcb8218) at 0xce1a80
    REFCNT = 2
    FLAGS = (SHAREKEYS)
    ARRAY = 0x0
    KEYS = 0
    FILL = 0
    MAX = 7

If this PVLV values is assigned to then it autovivifies the hash. If it is just read, it doesn't. It behaves kinda like a tied variable. Personally I don't like like this behaviour much, but it was added in 5.004, before my time as a p5 porter.

Dave.

[LanX]

> Personally I don't like like this behaviour much, but it was added in 5.004, before my time as a p5 porter

Which of both?

I personally find the behaviour of the sub arg aliasing far less surprising.

The other behaviour rather looks like sacrificing logic for performance or easier code logic.

I suppose changing that might be too expensive now, but most code doesn't try to assign to $_ in a loop.

If we are going to keep this behaviour, we'll need to document it properly.

Something like loop aliasing always triggers autovivification

[dave_the_m]

To clarify:

I am happy with the fact that 'for', 'grep' etc evaluate their args in lvalue context and thus autovivify, and I don't wish it to change. I find that this behaviour makes the rules for autovivification logically simple(r) and consistent.

I am unhappy with the deferring mechanism in hash lookups when used as an arg to a function call. It's a tricksy special-case behaviour and is hard to explain (above, I had to start talking about PVLVs and showing the output of Devel::Peek). It adds performance + memory overhead for such function calls, and is too clever for its own good. But I'm not proposing that it be be changed, since it was intentional behaviour added over 20 years ago.

Dave.

[Anonymous Monk]

Here is the link do bug in 2016: https://github.com/Perl/perl5/issues/15583

[Anonymous Monk]

Re^3: unexpected modify hash in a distance with grep { $_ }