After printing our header, we use Perl's built-in function binmode to indicate that we are outputting binary data. This is important. On Unix systems, binmode does nothing (thus on these systems it can be omitted), but on Windows, MacOS, and other operating systems that do not use a single newline as an end-of-line character, it disables automatic end-of-line translation that may otherwise corrupt binary output.

Finally, we read and output the image data. Note that because it is a binary file, there are no standard line endings, so we must use read instead of <> used on text files.

For CGI scripts like random_image.cgi that determine the filename and/or extension dynamically, you can still accomplish this with redirection. For example, we could replace the line that sets $image in random_image.cgi (Example 13-1) with the following lines:

my( $image ) = $q->path_info =~ /(\w+\.\w+)$/;

unless ( defined $image and -e IMAGE_DIRECTORY . "/$image" ) {
    $image = random_file( IMAGE_DIRECTORY, '\\.(png|jpg|gif)$' );
    print $q->redirect( $q->script_name . "/$image" );
    exit;
}

The first time this script is accessed, there is no additional path information, so it fetches a new image from our random_file function and redirects to itself with the filename appended as path information. When this second request arrives, the script retrieves the filename from the path information and uses this if the filename matches our regular expression and it exists. If it isn't a valid filename, the script acts as if no path had been passed and generates a new filename.

Note that our filename regular expression, /(\w+\.\w+)$/, prevents any images in our image directory that have characters not matched by \w from being displayed, including images that contain hyphens. Depending on the filenames you are using, you may need to adjust this pattern.

In Example 13-1, we generated an Expires HTTP header in order to discourage caching. Unfortunately, not all browsers respect this header, so it is quite possible for a user to get a stale image instead of a dynamic one. Some browsers also try to determine whether a resource is generated dynamically by something such as a CGI script or whether it is static; these browsers seem to assume that images are static, especially if you append additional path information as we just discussed.

There is a way to force browsers not to cache images, but this requires that the tag for the image also be dynamically generated. In these circumstances, you can add a value that constantly changes, such as the time in seconds, to the URL:

my $time = time;
print $q->img( { -src => "/cgi/survey_graph.cgi/$time/survey.png" } );

By adding the time to the additional path information, the browser views each request (more than a second apart) as a new resource. However, this technique does fill the browser's cache with duplicate images, so use it sparingly, and always combine this with an Expires header for the sake of browsers that support it. Adding a value like this to the query string also works:

print $q->img( { -src => "/cgi/survey_graph.cgi/survey.png?random=$time" } );

If nothing else on the HTML page is dynamic, and you do not wish to convert it to a CGI script, then you can also accomplish this via a server-side include (see Chapter 6, "HTML Templates"):

<!--#config timefmt="%d%m%y%H%M%S" -->
<IMG SRC="/cgi/survey_graph.cgi/<!--#echo var="DATE_LOCAL"-->/survey.png">

Although this is a little hard to read and is syntactically invalid HTML, the SSI tag will be parsed by an SSI-enabled server and replaced with a number representing the current date and time before it is sent to the user.