The Perl Advent Calendar needs more articles for 2022. Submit your idea today!


CGI::Test - CGI regression test framework


 # In some t/script.t regression test, for instance
 use CGI::Test;                 # exports ok()

 my $ct = CGI::Test->new(
    -base_url   => "http://some.server:1234/cgi-bin",
    -cgi_dir    => "/path/to/cgi-bin",

 my $page = $ct->GET("http://some.server:1234/cgi-bin/script?arg=1");
 ok 1, $page->content_type =~ m|text/html\b|;

 my $form = $page->forms->[0];
 ok 2, $form->action eq "/cgi-bin/some_target";

 my $menu = $form->menu_by_name("months");
 ok 3, $menu->is_selected("January");
 ok 4, !$menu->is_selected("March");
 ok 5, $menu->multiple;

 my $send = $form->submit_by_name("send_form");
 ok 6, defined $send;

 # Now interact with the CGI

 $menu->select("March");        # "click" on the March label
 my $answer = $send->press;     # "click" on the send button
 ok 7, $answer->is_ok;          # and make sure we don't get an HTTP error


The CGI::Test module provides a CGI regression test framework which allows you to run your CGI programs offline, i.e. outside a web server, and interact with them programmatically, without the need to type data and click from a web browser.

If you're using the CGI module, you may be familiar with its offline testing mode. However, this mode is appropriate for simple things, and there is no support for conducting a full session with a stateful script. CGI::Test fills this gap by providing the necessary infrastructure to run CGI scripts, then parse the output to construct objects that can be queried, and on which you can interact to "play" with the script's control widgets, finally submitting data back. And so on...

Note that the CGI scripts you can test with CGI::Test need not be implemented in Perl at all. As far as this framework is concerned, CGI scripts are executables that are run on a CGI-like environment and which produce an output.

To use the CGI::Test framework, you need to configure a CGI::Test object to act like a web server, by providing the URL base where CGI scripts lie on this pseudo-server, and which physical directory corresponds to that URL base.

From then on, you may issue GET and POST requests giving an URL, and the pseudo-server returns a CGI::Test::Page object representing the outcome of the request. This page may be an error, plain text, some binary data, or an HTML page (see CGI::Test::Page for details).

The latter (an HTML page) can contain one or more CGI forms (identified by <FORM> tags), which are described by instances of CGI::Test::Form objects (see CGI::Test::Form for details).

Forms can be queried to see whether they contain a particular type of widget (menu, text area, button, etc...), of a particular name (that's the CGI parameter name). Once found, one may interact with a widget as the user would from a browser. Widgets are described by polymorphic objects which conform to the CGI::Test::Form::Widget type. The specific interaction that is offered depends on the dynamic type of the object (see CGI::Test::Form::Widget for details).

An interaction with a form ends by a submission of the form data to the server, and getting a reply back. This is done by pressing a submit button, and the press() routine returns a new page. Naturally, no server is contacted at all within the CGI::Test framework, and the CGI script is ran through a proper call to one of the GET/POST method on the CGI::Test object.

Finally, since CGI::Test is meant to be used from regression test scripts, it exports a single ok() routine which merely prints the messages expected by Test::Harness. This is the only functional routine in this module, all other accesses being made through a CGI::Test object.


Procedural Interface

There is only one such routine:

ok num, boolean [, comment]

Prints the ok or not ok message for Test::Harness depending on whether boolean is respectively true or false. An optional comment string may be supplied as well and will be printed after a '#' sign:

    ok 1, 2+2 == 4, "trivial arithmetic";

will print:

    ok 1 # trivial arithmetic

since the test trivially succeeds.

Creation Interface

The creation routine new() takes the following mandatory parameters:

-base_url => URL of the cgi-bin directory

Defines the URL domain which is handled by CGI::Test. This is the URL of the cgi-bin directory.

Note that there is no need to have something actually running on the specified host or port, and the server name can be any host name, whether it exists or not. For instance, if you say:

    -base_url => ""

you simply declare that the CGI::Test object will know how to handle a GET request for, say:

and it will do so internally, without contacting on port 70...

-cgi_dir => path to the cgi-bin directoru

Defines the physical path corresponding to the cgi-bin directory defined by the -base_url parameter.

For instance, given the settings:

    -base_url => "",
    -cgi_dir  => "/home/ram/cgi/test"

then requesting

will actually run


Those things are really easier to understand via examples than via formal descriptions, aren't they?

The following optional arguments may also be provided:

-cgi_env => HASH ref

Defines additional environment variables that must be set, or changes hardwirted defaults. Some variables like CONTENT_TYPE really depend on the request and will be dynamically computed by CGI::Test.

For instance:

    -cgi_env => {
        HTTP_USER_AGENT     => "Mozilla/4.76",
        AUTH_TYPE           => "Digest",

See "CGI ENVIRONMENT VARIABLES" for more details on which environment variables are defined, and which may be superseded.

-doc_dir => path to document tree

This defines the root directory of the HTTP server, for path translation. It defaults to /var/www.

NOTE: CGI::Test only serves CGI scripts for now, so this setting is not terribly useful, unless you care about PATH_TRANSLATED.

-tmp_dir => path to temporary directory

The temporary directory to use for internal files created while processing requests. Defaults to the value of the environment variable TMPDIR, or /tmp if it is not set.

Object Interface

The following methods, listed in alphabetical order, are available:

GET url_string [, auth_user]

Issues an HTTP GET request of the specified URL, given as the string url_string. It must be in the http scheme, and must lie within the configured CGI space (i.e. under the base URL given at creation time via -base_url).

Optionally, you may specify the name of an authenticated user as the auth_user string. CGI::Test will simply setup the CGI environment variable REMOTE_USER accordingly. Since we're in a testing framework, you can pretend to be anyone you like. See "CGI ENVIRONMENT VARIABLES" for more information on environment variables, and in particular AUTH_TYPE.

GET returns a CGI::Test::Page polymorphic object, i.e. an object whose dynamic type is an heir of CGI::Test::Page. See CGI::Test::Page for more information on this class hierarchy.

POST url_string, input_data [, auth_user]

Issues an HTTP POST request of the specified URL. See GET above for a discussion on url_string and auth_user, which applies to POST as well.

The input_data parameter must be a CGI::Test::Input object. It specifies the CGI parameters to be sent to the script. Users normally don't issue POST requests manually: they are the result of submits on forms, which are obtained via an initial GET. Nonetheless, you can create your own input easily and issue a "faked" POST request, to see how your script might react to inconsistent (and probably malicious) input for instance. See CGI::Test::Input to learn how to construct suitable input.

POST returns a CGI::Test::Page polymorphic object, like GET does.


The base path in the URL space of the base URL configured at creation time. It's the URL with the scheme, host and port information removed.


The configured CGI root directory where scripts to be run are held.


The configured document root directory.


The host and port of the base URL you configured at creation time.

split_uri URI

Splits an URI object into server (host and port), path and query components. The path is simplified using UNIX semantics, i.e. /./ is ignored and stripped, and /../ is resolved by forgetting the path component that immediately precedes it (no attempt is made to make sure the translated path was indeed pointing to an existing directory: simplification happens in the path space).

Returns the list (host, path, query).


The temporary directory that is being used.


The CGI protocol defines a set of environment variables which are to be set by the web server before invoking the script. The environment created by CGI::Test conforms to the CGI/1.1 specifications.

Here is a list of all the known variables. Some of those are marked read-only. It means you may choose to set them via the -cgi_env switch of the new() routine, but your settings will have no effect and CGI::Test will always compute a suitable value.

Variables are listed in alphabetical order:


The authentication scheme used to authenticate the user given by REMOTE_USER. This variable is not present in the environment if there was no user specified in the GET/POST requests.

By default, it is set to "Basic" when present.


Read-only variable, giving the length of data to be read on STDIN by POST requests (as told by REQUEST_METHOD). If is not present for GET requests.


Read-only variable, giving the MIME type of data to be read on STDIN by POST requests (as told by REQUEST_METHOD). If is not present for GET requests.


The Common Gateway Interface (CGI) version specification. Defaults to "CGI/1.1".


The set of Content-Type that are said to be accepted by the client issuing the HTTP request. Since there is no browser making any request here, the default is set to "*/*".

It is up to your script to honour the value of this variable if it wishes to be nice with the client.


The charset that is said to be accepted by the client issuing the HTTP request. Since there is no browser making any request here, the default is set to "iso-8859-1".


Whether the connection should be kept alive by the server or closed after this request. Defaults to "Close", but since there's no connection and no real client...


This is the host processing the HTTP request. It is a read-only variable, set to the hostname and port parts of the requested URL.


The user agent tag string. This can be used by scripts to emit code that can be understood by the client, and is also further abused to derive the OS type where the user agent runs.

In order to be as neutral as possible, it is set to "CGI::Test" by default.


Read-only variable set to the extra path information part of the requested URL. Always present, even if empty.


This read-only variable is only present when there is a non-empty PATH_INFO variable. It is simply set to the value of PATH_INFO with the document rootdir path prepended to it (the value of the -doc_dir creation argument).


This very important read-only variable is the query string present in the requested URL. Note that it may very well be set even for a POST request.


The IP address of the client making the requst. Can be used to implement an access policy from within the script. Here, given that there's no real client, the default is set to "", which is the IP of the local loopback interface.


The DNS-translated hostname of the IP address held in REMOTE_ADDR. Here, for testing purposes, it is not computed after REMOTE_ADDR but can be freely set. Defaults to "localhost".


This read-only variable is only present when making an authenticated GET or POST request. Its value is the name of the user we are supposed to have successfully authenticated, using the scheme held in AUTH_TYPE.


Read-only variable, whose value is either GET or POST.


Read-only variable set to the filesystem path of the CGI script being run.


Read-only variable set to the virtual path of the CGI script being run, i.e. the path given in the requested URL.


The host name running the server, which defaults to the host name present in the base URL, provided at creation time as the -base_url argument.


The port where the server listens, which defaults to the port present in the base URL, provided at creation time as the -base_url argument. If no port was explicitely given, 80 is assumed.


The protocol which must be followed when replying to the client request. Set to "HTTP/1.1" by default.


The name of the server software. Defaults to "CGI::Test".


There are some, most probably. Please notify me about them.

The following limitations (in decreasing amount of importance) are known and may be lifted one day -- patches welcome:

  • There is no support for cookies. A CGI installing cookies and expecting them to be resent on further invocations to friendly scripts is bound to disappointment.

  • There is no support for testing a script in-situ, i.e. via a real web server, whereby CGI::Test would merely act as a client. Currently, scripts are run internally only, and therefore it is not possible to validate the installation procedure on the server.

  • There is no support for plain document retrieval: only CGI scripts can be fetched by an HTTP request for now.

  • There is no support for javascript (!). Plain buttons attached to scripts will do nothing when pressed...

  • There is no support for frames (!).

  • There is no support for Java (!). Perhaps if I work for Sun one day...

  • There is no support for the <ISINDEX> tag, which is deprecated.


You can find information about CGI::Test and other related modules at:


CGI::Test now has a publicly accessible CVS server provided by SourceForge ( You can access it by going to:


The original author is Raphael Manfredi <>.

Send bug reports, hints, tips, suggestions to Steven Hilton <>.


CGI(3), CGI::Test::Page(3), CGI::Test::Form(3), CGI::Test::Input(3), CGI::Test::Form::Widget(3), HTTP::Status(3), URI(3).

1 POD Error

The following errors were encountered while parsing the POD:

Around line 1074:

You forgot a '=back' before '=head1'