Section 3: Techniques: "How do I..."

This section comprises programming hints and tips for a number of popular
tasks. Also included are a number of common questions to which the answer
is "you can't", with the reasons why.

3.1: Can I get information about who is visiting?

*sigh*
Many people keep mailing me questions or suggested hacks to get
visitor information, particularly email addresses.   It seems they
won't take "NO" for an answer.

The bottom line is that whatever information is available to _you_
is _equally_ available to every spammer on the net.   Therefore when
a browser bug _does_ permit personal data to be collected, it gets
reported and fixed very quickly (one short-lived Netscape release
reportedly had such a bug).

You can get some limited information from the environment variables
passed to you by the browser.   Relatively few of these are guaranteed
to be available, and some may be misleading.   For particular types
of information, see below.   For full details, see NCSA's reference pages.

[Table of Contents] [Index]

3.2: Can I get the email of visitors?

Why do you want to do this?

The best information available is the REMOTE_ADDR and REMOTE_HOST,
which tell you nothing about the user.   Techniques such as "finger@"
are not reliable, are widely disliked, and generally serve only to
introduce long delays in your CGI.   Better - as well as more polite -
just to ask your users to fill in a form.

BTW: the "From:" header line (HTTP_FROM variable) is usually only set
by robots, since human visitors to your webpage will not normally want
their addresses collected without permission, and browsers respect this.

[Table of Contents] [Index]

3.3: "But I saw some.kool.site display my email address..."

Some sites will play party tricks, which can get *some users* email
addresses.   Possible tell-tale signs of this are inordinate delays
loading a page (fingering @REMOTE_HOST - doesn't often work but
probably can't be detected from the webpage), or a submit button that
appears to do nothing at all (a mailto: link - works quite well but
trivially detectable).   As a "snoop" party trick that's fine, but
if you find someone abusing these facilities (eg they send you
junkmail), alert their service provider!

[Table of Contents] [Index]

3.4: Can I verify the email addresses people enter in my Form?

Unfortunately people will sometimes enter an incorrect or invalid
email address in your Form.   Worse, they may enter a valid but
incorrect email address that will deliver to someone who doesn't
want your mail.

Proposed regexps to match email addresses are sometimes posted.
Most of these will fail against perfectly valid email addresses,
like "S=N.OTHER/OU1=X12345A/RECIPNUM=1/MTA-BASIC@attmail.com"
(which is what your address looks like if you are connected to
the Internet via X400 - and if you think that example is too easy,
check the ones at the end of Eli the Bearded's Email Addressing FAQ).

Probably the most complete parser and checker available for download
is Tom Christiansen's, at
http://www.perl.com/CPAN/authors/Tom_Christiansen/scripts/ckaddr.gz
Of course, this still says nothing about deliverability.

A frequently-suggested hack that doesn't work is to use
SMTP EXPN or VRFY commands.   Modern versions of sendmail permit
administrators to disable these commands, and many sites take
advantage of this facility to protect their users' privacy.

Probably the best way to verify an email address is to send mail to
it, asking the user to respond.   Include a clause like "if you have
received this mail in error, please accept our apologies..."

[Table of Contents] [Index]

3.5: Subject: How can I get the hostname of the remote user?

You can't. Well, not always.

IF it is available, you'll find it in the REMOTE_HOST environment
variable.  However, this will more often than not contain the numerical
IP address rather than the IP name of the remote host. Remember that
not all IP addresses have a hostname associated with them; this is the
case of most IP addresses assigned to dialup users, for example. Your
web server may also not perform a reverse lookup on incoming
connections, in which case REMOTE_HOST will contain the IP address even
if it has a corresponding IP name. In the second case, you can do a
reverse lookup yourself in your script, but this is expensive and
should probably be avoided unless absolutely necessary.

Even if you do manage to obtain a hostname, you should be aware that it
may not correspond to the hostname the user is accessing your page
from. It may instead be that of an intervening proxy host.

The short answer is therefore that there is no reliable way of finding
out what the remote user's hostname is.

[Table of Contents] [Index]

3.6: Can I get browser details and return different pages?

Why do you want to do this?

Well-written HTML will display correctly in any browser, so the correct
answer to this question is to design a template for your output in good
HTML, and make sure your output is correct.

If you insist on a different answer, you can use the HTTP_USER_AGENT
environment variable.  This requires care, and can lead to unexpected
results.   For example, checking for "Mozilla" and serving a frameset
to it ensures that you *also* serve the frameset to early (Non-Frame)
Netscapes, me-too browsers (notably MicroSoft) and others who have
chosen to lie to you about their browser.

Note also that not every User Agent is a browser.   Your page may be
read by a user agent you've never heard of, and then displayed by
100 different browsers.   Or retrieved by different browsers from
a cache.   Another reason to write good HTML, and not try to
devise a clever or koool substitute.

[Table of Contents] [Index]

3.7: Can I trace where a user has come from/is going to?

HTTP_REFERER might or might not tell you anything.   By all means
use it to collect partial statistics if you participate in (say)
an advertising banner scheme.   But it is not always set, and may
be meaningless (eg if a user has accessed your page from a bookmark,
and the browser is too dumb to cope with this).

You cannot trace outgoing links at all.   If you really must try,
point all the external links to your HTTPD and use its redirection
facility (which gives you generally-reliable logs).   This is much
less inefficient than using a CGI script.

BTW: don't even think about asking Javascript to send you information
on some event: it's a violation of privacy which Netscape fixed as
soon as complaints about its abuse started coming in.   If it works
with *your* browser, you should upgrade!

[Table of Contents] [Index]

3.8: Can I launch a long process and return a page before it's finished?

[UNIX]
You have to fork/spawn the long-running process.
The important thing to remember is to close all its file descriptors;
otherwise nothing will be returned to the browser until it's finished.
The standard trick to accomplish this is redirection to/from /dev/null:

        exec ("long_process < /dev/null > /dev/null 2>&1 &")
        print HTML page as usual

(don't take "exec" as literal in anything but a shell script - in
C, Perl, etc use fork+exec or system() :-)

[Table of Contents] [Index]

3.9: Can I launch a long process which the user interacts with?

This does not fit well with the basic mechanics of the Web, in which
each transaction comprises a single request and response.
If your processing can be done on the Client machine, you can use
a clientside application; for example a Java applet.

For processing on the server, one trick that works well for Clients
running an X server (and far more efficient than a JAVA solution) is:
  if ( fork() ) {
    print HTML page explaining what's going on and advising about xhost
  } else {
    exec ("xterm -display THEIR_DISPLAY -title MY_APP -e MY_PROG ARGS
        < /dev/null > /dev/null 2>&1 &") ;
  }
NOTE: THEIR_DISPLAY is not necessarily the same as REMOTE_HOST or REMOTE_ADDR.
You have to ask users to supply their display (set REMOTE_HOST as default).

[ Question: Is there a JAVA alternative to xterm yet, for platforms
  which support JAVA but not X ? ]

[Table of Contents] [Index]

3.10: Can I password-protect my pages?

Yes.   Use your HTTPD's authentication, just as you would a basic HTML page.
Now you'll have the identity of every visitor in REMOTE_USER.

[Table of Contents] [Index]

3.11: Can I do HTTP authentication using CGI?

It depends on which version of the question you asked.

Yes, you can use CGI to trigger the browser's standard Username/Password
dialogue.   Send a response code 401, together with a "WWW-authenticate"
header including details of the the authentication scheme and realm:
e.g. (in a non-NPH script)

	Status: 401 Unauthorized to access the document
	WWW-authenticate: Basic realm="foobar"
	Content-type: text/plain

	Unauthorised to access this document

The use you can make of this is server-dependent, and harder,
since most servers expect to deal with authentication before ever
reaching the CGI (eg through .www_acl or .htaccess).
Thus it cannot usefully replace the standard login sequence, although
it can be applied to other situations, such as re-validating a user -
e.g after a certain timeout period or if the same person may need to
login under more than one userid.

What you can never get in CGI is the credentials returned by the user.
The HTTPD takes care of this, and simply sets REMOTE_USER to the
username if the correct password was entered.

For a much longer discussion of this question (with code extracts),
see my discussion at http://www.webthing.com/tutorials/login.html

[Table of Contents] [Index]

3.12: Can I identify users/sessions without password protection?

The most usual (but browser-dependent) way to do this is to set a cookie.
If you do this, you are accepting that not all users will have a 'session'.

An alternative is to pass a session ID in every GET URL, and in hidden
fields of POST requests.   This can be a big overhead unless _every_ page
requires CGI in any case.

Another alternative is the Hyper-G[1] solution of encoding a session-id in
the URLs of pages returned:
	http://hyper-g.server/session_id/real/path/to/page
This has the drawback of making the URLs very confusing, and causes any
bookmarked pages to generate old session_ids.

Note that a session ID based solely on REMOTE_HOST (or REMOTE_ADDR)
will NOT work, as multiple users may access your pages concurrently
from the same machine.

[1] Actually I don't think that's been true of Hyper-G since sometime in '96.

[Table of Contents] [Index]

3.13: Can I redirect users to another page?

For permanent and simple redirection, use the HTTPD configuration file:
it's much more efficient than doing it yourself.   Some servers enable
you to do this using a file in your own directory (eg Apache) whereas
others use a single configuration file (eg CERN).

For more complicated cases (eg process form inputs and conditionally
redirect the user), use the "Location:" response header.
If the redirection is itself a CGI script,  it is easy to URLencode
parameters to it in a GET request, but don't forget to escape the URL!

[Table of Contents] [Index]

3.14: Can I run a CGI script without returning a new page to the browser?

Yes, but think carefully first:  How are your readers going to know
that their "submit" has succeeded?   They may hit 'submit' many times!

The correct solution according to the HTTP specification is to
return HTTP status code 204.   As an NPH script, this would be:

	#!/bin/sh
	# do processing (or launch it as background job)
	echo "HTTP/1.0 204 No Change"
	echo

(as non-NPH, you'd simply replace HTTP/1.0 with the Status: CGI header).

Alan J Flavell has pointed out that this will fail with certain
popular browsers, and suggests a workaround to accommodate them:

[ May 1998 update[1]: I'm deleting Alan's suggestion, because the problem
  is mainly of historical interest, and the workaround is no longer
  recommended.  See his page for a a detailed survey and recommendations.
]

His survey is at
http://ppewww.ph.gla.ac.uk/%7Eflavell/status204/results.html

[1] With apologies to Alan for having left it in so long.

[Table of Contents] [Index]

3.15: Can I write output to a different Netscape frame?

Yep.   The fact you're using CGI makes no difference: use
"target=" in your links as usual.   Alternatively, the script
can print a "Window-target:" header.   Read Netscape's pages
for detail: these answer all the questions about things like
"getting rid of" or "breaking out of" frames, too.

[Table of Contents] [Index]

3.16: Can I write output to several frames at once?

A single CGI script can only ever print to one frame.

However, this limitation may be overcome by using more than one script.
The first script (the URL of the "submit" button) prints a frameset,
typically to a "_parent" or "_top" target.   The sources for one or
more of the frames thus generated may also be CGI scripts, to which
you can easily pass parameters (eg encoded in URLs with method GET).
This hack is definitely not recommended.   If you find yourself wanting
to update several frames from a single user event, it probably means
you should review the design of your application at a higher level.

Warnings:
 1. Don't forget to escape your URLs.
 2. This technique results in your server being hit by multiple 
    concurrent CGI requests.   You'll need LOTS of memory, especially
    if you use a memory-hog like Perl.   It can be a good recipe
    for bringing a server to its knees.

Javascript is often a valid alternative here, but note just how silly
it can (and often does) look in a different browser.

[Table of Contents] [Index]

3.17: Can I use a CGI script to generate both text and inline images?

Not directly.   One script generates one response to one request.

If you want to generate a dynamic page including dynamic images
(say, a report including graphs, all of which depend on user input)
then your primary script will print the usual
   <img src="[script-to-generate-image]" alt="[what you asked for]">
and, just as in the multiple frames case, you can pass data to the
image-generating program encoded in a GET URL.   Of course, the same
caveats apply: see above.

[Table of Contents] [Index]

3.18: How can I use Caches to make CGI scripts faster and more Net-friendly?

This is currently beyond the scope of this FAQ (whose author urgently
needs to improve his own applications in this regard).   However,
there is an excellent introduction to net-friendly webpages, including
CGI pages, at http://vancouver-webpages.com/CacheNow/

A sample cacheing perl/cgi script by Andrew Daviel is available at
http://vancouver-webpages.com/proxy/log-tail.pl

[Table of Contents] [Index]

3.19: How can I avoid users hitting "submit" twice?

You can't.   You just have to deal with it when they do.

You can avoid re-processing a submission by embedding a unique ID in your
Form each time it is displayed.   When you process the form, you enter
the ID in a database.  Or, if it's already there, you don't repeat the
processing.

You probably want to expire your database entries after a little time:
an hour should be fine in a typical situation.

If you're already using cookies (e.g. a shoppingcart), an alternative is
to use the cookie as a unique identifier.   This means you also have to
handle the situation where a user deliberately "goes round twice" and
submits the same form with different contents.

If your script may take some time to process, you should also consider
running it as a background job, and returning an immediate
acknowledgement to the user (see above if your "immediate" response
gets delayed until processing is complete in any case).

[Table of Contents] [Index]

3.20: How can I stop my CGI script reading and writing files as "nobody"?

CGI scripts are run by the HTTPD, and therefore by the UID of the HTTPD
process, which is (by convention) usually a special user "nobody".

There are two basic ways to run a script under your own userid:
(1) The direct approach: use a setuid program.
(2) The double-server approach: have your CGI script communicate
    with a second process (e.g. a daemon) running under your userid,
    which is responsible for the actual file management.

The direct approach is usually faster, but the client-server architecture
may help with other problems, such as maintaining integrity of a database.

When running a compiled CGI program (e.g. C, C++), you can make it
setuid by simply setting the setuid bit:
e.g. "chmod 4755 myprog.cgi"

For security reasons, this is not possible with scripting languages
(eg Perl, Tcl, shell).   A workaround is to run them from a setuid
program, such as cgiwrap.

In most cases where you'd want to use the client-server approach,
the server is a finished product (such as an SQL server) with its
own CGI interface.
A lightweight alternative to this is Don Libes' "expect" package.

Note that any program running under your userid has access to all your
files, and could do serious damage if hacked.   Take care!

[Table of Contents] [Index]

3.21: How can I prevent my CGI results being cached by the browser?

Firstly, we need to debunk a myth.  People asking this question usually
add that they tried "Pragma: no-cache".  Whilst this is not actively
wrong, there is no requirement on browsers to take any notice of it,
and most of them don't.

The "Pragma: no-cache" header (now superseded by HTTP/1.1 Cache-Control)
is a directive to proxies.  The browser sends it with an HTTP request
to indicate that it wants the request to be dealt with by the original
server and will not accept a proxy's cached document (e.g. when you
use a reload button).  The server may send it to tell a proxy not to
cache the document.

Having said all that, a practical hack to get round cacheing is
to use a different URL for your CGI script each time it's called.
This can easily be accomplished by adding a unique identifier such
as current time in the QUERY_STRING or PATH_INFO.  The browser will
see a different URL, but the script can just ignore it.  Note that
this can be very inefficient, and should be avoided where possible.

[Table of Contents] [Index]

3.22: How can I control the default filename when downloading a file via CGI?

	(from a newsgroup post by Matthew Healy)

One option, assuming you aren't already using the PATH_INFO
environment variable, is just to call your CGI script with extra
path information.

For example, suppose the URL to your script is actually

http://server.com/scriptname?name1=value1&name2=value2

Instead, try calling it as

http://server.com/scriptname/filename.ext?name1=value1&name2=value2

And probably the browser will assign the name given in the last chunk
as the suggested filename for downloading.

This works because the http server looks for the program file to run,
then passes any extra path to the program as PATH_INFO variable; the
browser cannot tell where the SCRIPT_NAME part ends and the PATH_INFO
part begins.

This can also be very useful if you want one script to generate more
than one filename -- the script can check the PATH_INFO value and
alter its response accordingly...

[Table of Contents] [Index]