sub inplace_uc {
   our $text; local *text = \$_[0];
   $text =~ tr/a-z/A-Z/;
}
[download]

Update: Added missing "$text =~".

I was enquiring as to what the orginal poster wanted to accomplish and why he thought he needed the typeglob syntax. That's a different question than how it works. You definately know how it works. We have a great intellectual question, but maybe we should be answering a different question? I'm just probing a bit to see if there is some more fundamental application issue here.

I looked at this inplace_uc code and I did add "$text =~" in front of the tr, which seemed appropriate.

Here, there is really no performance to be gained that I can see. The inplace_uc() call passes the whole string on the stack and there is some stuff to get the address of it. If we are doing an "inplace" replacement, why not just pass the addresss of the string instead of the whole string? My inplace_uc2() uses the address of the string in a very straightforward way.

As a perfence of style, I would always expect an inplace modification to to take place on an address or in Perl lingo a referece to something. Ok, mileage varies as they say.

#!/usr/bin/perl -w
use strict;

my $lc_msg = 'message';
inplace_uc ($lc_msg);
print "$lc_msg\n";

sub inplace_uc 
{
   our $text; 
   local *text = \$_[0];  # equivalent to:
                          #  local *text = \shift;
   $text =~ tr/a-z/A-Z/;  # added $text =~
}

my $lc_msg_y = 'another_message';
inplace_uc2 (\$lc_msg_y);
print "$lc_msg_y\n";

sub inplace_uc2
{
     my $str_ref = shift;
     $$str_ref =~ tr/a-z/A-Z/;
}
# prints:
#MESSAGE
#ANOTHER_MESSAGE
[download]

Here, there is really no performance to be gained that I can see.

The difference is in the interface, not the performance.

The inplace_uc() call passes the whole string on the stack

No, just a "C" pointer.