parser outside of web tree. whats the trick?

If I have php installed as an executable outside the web tree I can't seem to get php scripts, called as a cgi, to work correctly. When run from the command line this script.... --------------------- #!/usr/local/bin/php <html><body> <?php echo "hello world\n"; ?> </body></html> --------------------- Produces.... --------------------- X-Powered-By: PHP/4.0.4pl1 Content-type: text/html <html><body> hello world </body></html> --------------------- Just as expected. However, when called from a browser it ends up getting.... --------------------- #!/usr/local/bin/php <html><body> hello world </body></html> --------------------- when we view the source with a browser. If we telnet to port 80 and do a get for the url in question we find that the above is all that we get, no headers of any kind. If we add a -q the #! line we get the behavior we expect from the command line, the headers are supresed, but the browser ends up getting "No input file specified." sans headers or any html tags. Why am I seeing this behavior? I would expect the script the produce the exact same output whether it is called via the command line or as a cgi outside the web tree. I had thought that perhaps it was a system dependent bug, however, I am seeing this same behavior on several machines, AIX 4.2 running 4.06, FreeBSD 4.3 running 4.0.4.pl1, and I've found similar reports in the bug db for HP/UX etc. So... this leads me to believe I must be missing something somewhere.... anyone have any thoughts? How do I supress the printing of the #! line? It has to be in col 1 row 1, if it wasn't for that I could manually enter the headers at least and continue on my way, Where did the headers go? if I use the -q option to purposely turn off headers, I could provide my own, but if I have that option it tells me that there is no input file specified if its not invoked from the command line.