Re: CGI/1.1 draft
Sun, 6 Mar 1994 16:51:20 --100

From: (Ari Luotonen)

> I like the idea of a data root directory environment variable. At
> user request I have already implemented this as a non-standard feature
> in GN.

Then what would I do? cern_httpd doesn't have a single document
root, and there is no reason why it should. I would like to see
DOCUMENT_ROOT as a server-specific feature rather than dictated
by the spec.

-- Cheers, Ari --

Seems to me the concept of server-specific features is contrary to the
idea of CGI itself. I perceive CGI to be a (de-facto) standard to
write portable server scripts. If you put useful functionality off as
'server-specific' people would use it and their scripts would not be
CGI-conformant, thus lessening the value of CGI itself. I can thus find
arguments in favor of NOT allowing server-specific stuff at all.

Portability to me is the ability to take a script from one machine,
move it to another and it works without any modifications. A weak spot
in the current CGI is finding the location of the scripts local data
and the document tree in a portable way. Thus my suggestions of

As to DOCUMENT_ROOT and the virtual/physical mapping: I agree with Ari
there _need_ not be any direct mapping from URL to file system. However,
I guess most(all?) http server uses the filesystem as the basic mechanism
to do the URL->file mapping, modified by alias-mechanisms etc as needed.
Even if we can't have a 100% solution, a 90% solution would make life a
lot easier. The script author will just have to put into his
documentation what URL's the script accesses as files and the server
administrator will have to make sure they have a direct file mapping.
Baard Haafjeld | When you give a wolf a poodle cut, you
Norwegian Telecom Research | don't get a show dog but a pissed wolf.
SMTP-mail: | -Robert Asprin