From andrew@cleverdomain.org

Ordinarily, Perl changes a construct like while(<>) or
while(readline($handle)) into while(defined($_ = ...)). But if
&CORE​::GLOBAL​::readline is defined when compiling these constructs, $_ won't
be assigned and the defined check won't be added.

This is because when CORE​::GLOBAL​::readline is overriden, S_scan_inputsymbol
in toke.c will generate an OP_ENTERSUB instead of an OP_READLINE, and so
Perl_newLOOPOP will not recognize it as something to transform.

This is fairly low-impact since it only affects people who use
CORE​::GLOBAL​::readline, but it definitely stands in contradiction to the docs.

(This affects every perl since at least 2002 up to blead).

From mobrule@gmail.com

Perl's behavior that treats the expressions

  while (<HANDLE>)
  while (readline(<HANDLE>))

as equivalent to

  while (defined($_=readline(HANDLE)))

breaks when &CORE​::GLOBAL​::readline is defined. This makes it
difficult to override the builtin readline function with any legacy
code.

Demo​:

BEGIN {
   *CORE​::GLOBAL​::readline = sub { readline(shift || *ARGV) }
};
($_,$bar) = ("foo\n", "bar\n");
open X, '<', \$bar;
while (<X>) {
   if ($_ eq "bar\n") {
  print "PASS​: <X> is the same as ($_=<X>).\n";
   } elsif ($_ eq "foo\n") {
  print "FAIL​: <X> is not the same as ($_=<X>).\n";
   }
}

This script fails as written but passes if you comment out the BEGIN block.

Another demo​:

Compare​:

    $ perl -MO=Deparse -e 'print while <>'
    print $_ while (defined($_ = <ARGV>));   <---- $_ is set implicitly
    -e syntax OK

with​:

    $ perl -MO=Deparse -e 'BEGIN{*CORE​::GLOBAL​::readline=sub{}}' \
             -e 'print while <>'
    sub BEGIN {
        *CORE​::GLOBAL​::readline = sub {
        }
        ;
    }
    print $_ while readline(ARGV);          <----- $_ is not set !
    -e syntax OK

See also​: http​://stackoverflow.com/questions/5081767/.

If this isn't a bug, then it is at least underdocumented.

@jkeenan - Status changed from 'new' to 'open'