"I was thinking that writing a file out to disc and then reading it back in was a messy workaround ..."

I agree and had the same thought myself. There is a way to do this with in-memory filehandles; however, there were two reasons why I didn't pursue this.

1. I didn't know if you'd want to reuse the generated image. I see from your "new, additional information" this isn't the case; or, at least, I assume it's not.

   By the way, the "unlink $png_file;" was intended as optional for you and as housekeeping for me; I omitted to specifically state that.

2. I recalled that you were stuck with a fairly old version of Perl from your ISP. See the "Technical note" in "Opening a filehandle into an in-memory scalar" which may possibly still be a "hurdle" for you.

As well as a filename, and a GD::Image object when that bug gets fixed, PDF::API2::image() can also take a filehandle. The following code produces exactly the same PDF as my previous code.

#!/usr/bin/env perl

use strict;
use warnings;
use autodie;

use PDF::API2;
use GD::Tiler 'tile';

my $pdf_file = 'tiled_dogs.pdf';
my @dog_jpegs = qw{husky.jpg labrador.jpg retriever.jpg};

my $tiled_dogs = tile(
    Images => [@dog_jpegs],
    Center          => 1,
    ImagesPerRow    => 3,
);

my $pdf = PDF::API2::->new();
my $page = $pdf->page();

open my $png_fh, '<:raw', \$tiled_dogs;
my $canine_triptych = $pdf->image($png_fh);
close $png_fh;

$page->object($canine_triptych, 100, 650, 400);
$pdf->save($pdf_file);
[download]

Assuming you don't want to keep the generated image, and you're not confronted with technical difficulties, this would be a better solution. I would expect this would also be much faster; but do Benchmark if speed is important to you.

Update: Two minor tweaks to the code: removed parts left over from the previous script that were no longer needed here.

Thanks Ken - seeing a working example is extremely helpful.

My first thought is that I would need to use a pipe but then realised that there doesn't seem to be anything to pipe to...

You are right, I am using Perl v5.16 (still not moved toa VPS). I will try it without writing out and deleting a file some time but it is not a very high priority as the thing now works and will only get used a couple of times each month. Speed is not at all important - it could take half an hour, and it would still be acceptable.

In the actual code, I still create a GD::Image so that I can read the width

No need. PDF::API2 has methods to get image size. + Using PDF::API2::image_whatever directly is safe, BTW.

use strict;
use warnings;
use feature 'say';
use open IO => ':raw';
use GD;
use PDF::API2;

my $gd_obj = GD::Image-> new( scalar qx/convert rose: png:-/ );
my $pdf = PDF::API2-> new;
my $pdf_image = $pdf-> image_gd( $gd_obj );

say $pdf_image-> width;
say $pdf_image-> height;
[download]

Ah!
That's very helpful - thanks