I'm now working on a site I can only access by ftp and browser.

Besides the comments you've gotten about what a rotten hosting site this is, and the answers to why your script is not behaving correctly (no headers!) here is a trick I use when I'm force to work under such rotten hosting conditions:

I write a simple shell wrapper script that executes my CGI and writes the stdout/stderr to the browser like so:

#!/bin/sh
echo "Content-type: text/html"
echo
echo
echo
echo "<pre>
(
    /path/to/my/script/here.pl
) 2>&1 
echo "</pre>"
echo
[download]

of course substituting the path accordingly. Now I can see what I'm not seeing in the logs since I don't have access to those.

I can empathise with your situation of "ftp access only" since I've inherited sites like that. Still I've worked as quickly as possible to move those sites to a new home or I've "un-inherited" them as quickly as possible.

I figure that any hosting folks that don't trust me with shell aren't worthy of my trust. :-/

Also, I try and set up a test environment that matches as closely as practical the environment I'm going to be running in and test before I publish.