Lots more documentation.

[maypole.git] / doc / Request.pod

=head1 Maypole Request Hacking Cookbook

Hacks; design patterns; recipes: call it what you like, this chapter is a
developing collection of techniques which can be slotted in to Maypole
applications to solve common problems or make the development process easier.

As Maypole developers, we don't necessarily know the "best practice" for
developing Maypole applications ourselves, in the same way that Larry Wall
didn't know all about the best Perl programming style as soon as he wrote
Perl. These techniques are what we're using at the moment, but they may
be refined, modularized, or rendered irrelevant over time. But they've
certainly saved us a bunch of hours work.

=head2 Frontend hacks

These hacks deal with changing the way Maypole relates to the outside world;
alternate front-ends to the Apache and CGI interfaces, or subclassing chunks
of the front-end modules to alter Maypole's behaviour in particular ways.

=head3 Debugging with the command line

You're seeing bizarre problems with Maypole output, and you want to test it in
some place outside of the whole Apache/mod_perl/HTTP/Internet/browser circus.

B<Solution>: Use the C<Maypole::CLI> module to go directly from a URL to
standard output, bypassing Apache and the network altogether.

C<Maypole::CLI> is not a standalone front-end, but to allow you to debug your
applications without having to change the front-end they use, it temporarily
"borgs" an application. If you run it from the command line, you're expected
to use it like so:

    perl -MMaypole::CLI=Application -e1 'http://your.server/path/table/action'

For example:

    perl -MMaypole::CLI=BeerDB -e1 'http://localhost/beerdb/beer/view/1?o2=desc'

You can also use the C<Maypole::CLI> module programatically to create
test suites for your application. See the Maypole tests themselves or
the documentation to C<Maypole::CLI> for examples of this.

=head3 Changing how URLs are parsed

You don't like the way Maypole URLs look, and want something that either
fits in with the rest of your site or hides the internal workings of the
system.

C<Solution>: So far we've been using the C</table/action/id/args> form
of a URL as though it was "the Maypole way"; well, there is no Maypole
way. Maypole is just a framework and absolutely everything about it is 
overridable. 

If we want to provide our own URL handling, the method to override in
the driver class is C<parse_path>. This is responsible for taking
C<$r-E<gt>{path}> and filling the C<table>, C<action> and C<args> slots
of the request object. Normally it does this just by splitting the path
on C</>s, but you can do it any way you want, including getting the
information from C<POST> form parameters or session variables. 

For instance, suppose we want our URLs to be of the form
C<ProductDisplay.html?id=123>, we could provide a C<parse_path> method
like so:

    sub parse_path {
        my $r = shift;
        $r->{path} ||= "ProductList.html";
        ($r->{table}, $r->{action}) = 
            ($r->{path} =~ /^(.*?)([A-Z]\w+)\.html/);
        $r->{table}  = lc $r->{table};
        $r->{action} = lc $r->{action};
        my %query = $r->{ar}->args;
        $self->{args} = [ $query{id} ];
    }

This takes the path, which already has the query parameters stripped off
and parsed, and finds the table and action portions of the filename,
lower-cases them, and then grabs the C<id> from the query. Later methods
will confirm whether or not these tables and actions exist.

See L<BuySpy.pod> for another example of custom URL processing.

=head3 Maypole for mobile devices

You want Maypole to use different templates to display on particular
browsers.

B<Solution>: There are several ways to do this, but here's the neatest
we've found. Maypole chooses where to get its templates either by
looking at the C<template_root> config parameter or, if this is not
given, calling the C<get_template_root> method to ask the front-end to
try to work it out. We can give the front-end a little bit of help, by
putting this method in our driver class:

    sub get_template_root {
        my $r = shift;
        my $browser = $r->{ar}->headers_in->get('User-Agent');
        if ($browser =~ /mobile|palm|nokia/i) {
            "/home/myapp/templates/mobile";
        } else {
            "/home/myapp/templates/desktop";
        }
    }

(Maybe there's a better way to detect a mobile browser, but you get the
idea.)

=head2 Content display hacks

These hacks deal primarily with the presentation of data to the user,
modifying the C<view> template or changing the way that the results of
particular actions are displayed.

=head3 Null Action

You need an "action" which doesn't really do anything, but just formats
up a template.

B<Solution>: There are two ways to do this, depending on what precisely
you need. If you just need to display a template, C<Apache::Template>
style, with no Maypole objects in it, then you don't need to write any
code; just create your template, and it will be available in the usual
way.

If, on the other hand, you want to display some data, and what you're
essentially doing is a variant of the C<view> action, then you need to
ensure that you have an exported action, as described in
L<StandardTemplates.pod>:

    sub my_view :Exported { }

=head3 Template Switcheroo

An action doesn't have any data of its own to display, but needs to display
B<something>.

B<Solution>: This is an B<extremely> common hack. You've just issued an
action like C<beer/do_edit>, which updates the database. You don't want
to display a page that says "Record updated" or similar. Lesser
application servers would issue a redirect to have the browser request
C</beer/view/I<id>> instead, but we can actually modify the Maypole
request on the fly and, after doing the update, pretend that we were
going to C</beer/view/I<id>> all along. We do this by setting the
objects in the C<objects> slot and changing the C<template> to the
one we wanted to go to.

In this example from L<Flox.pod>, we've just performed an C<accept>
method on a C<Flox::Invitation> object and we want to go back to viewing
a user's page.

    sub accept :Exported {
        my ($self, $r) = @_;
        my $invitation = $r->objects->[0];
        # [... do stuff to $invitation ...]
        $r->{objects} = [$r->{user}];
        $r->{model_class} = "Flox::User";
        $r->{template} = "view";
    }

This hack is so common that it's expected that there'll be a neater
way of doing this in the future.

=head3 XSLT

Here's a hack I've used a number of times. You want to store structured
data in a database and to abstract out its display.

B<Solution>: You have your data as XML, because handling big chunks of
XML is a solved problem. Build your database schema as usual around the
important elements that you want to be able to search and browse on. For
instance, I have an XML format for songs which has a header section of
the key, title and so on, plus another section for the lyrics and
chords:

    <song>
        <header>
            <title>Layla</title>
            <artist>Derek and the Dominos</artist>
            <key>Dm</key>
        </header>
        <lyrics>
          <verse>...</verse>
          <chorus>
            <line> <sup>A</sup>Lay<sup>Dm</sup>la <sup>Bb</sup> </line> 
            <line> <sup>C</sup>Got me on my <sup>Dm</sup>knees </line> 
            ...
 
I store the title, artist and key in the database, as well as an "xml"
field which contains the whole song as XML.

To load the songs into the database, I can C<use> the driver class for
my application, since that's a handy way of setting up the database classes
we're going to need to use. Then the handy C<XML::TreeBuilder> will handle
the XML parsing for us:

    use Songbook;
    use XML::TreeBuilder;
    my $t = XML::TreeBuilder->new;
    $t->parse_file("songs.xml");

    for my $song ($t->find("song")) {
        my ($key) = $song->find("key"); $key &&= $key->as_text;
        my ($title) = $song->find("title"); $title = $title->as_text;
        my ($artist) = $song->find("artist"); $artist = $artist->as_text;
        my ($first_line) = $song->find("line");
        $first_line = join "", grep { !ref } $first_line->content_list;
        $first_line =~ s/[,\.\?!]\s*$//;
        Songbook::Song->find_or_create({
            title => $title,
            first_line => $first_line,
            song_key => Songbook::SongKey->find_or_create({name => $key}),
            artist => Songbook::Artist->find_or_create({name => $artist}),
            xml => $song->as_XML
        });
    }

Now we need to set up the custom display for each song; thankfully, with
the C<Template::Plugin::XSLT> module, this is as simple as putting the
following into F<templates/song/view>:

    [%
        USE transform = XSLT("song.xsl");
        song.xml | $transform
    %]

We essentially pipe the XML for the selected song through to an XSL
transformation, and this will fill out all the HTML we need. Job done.

=head3 Displaying pictures

You want to serve a picture, a Word document, or something else which
doesn't have a content type of C<text/html>, out of your database.

B<Solution>: Fill the content and content-type yourself.

Here's a subroutine which displays the C<photo> for either a specified
user or the currently logged in user. We set the C<output> slot of the
Maypole request object: if this is done then the view class is not called
upon to process a template, since we already have some output to display.
We also set the C<content_type> using one from the database.

    sub view_picture :Exported {
        my ($self, $r) = @_;
        my $user = $r->{objects}->[0];
        $r->{content_type} = $user->photo_type;
        $r->{output} = $user->photo;
    }

Of course, the file doesn't necessarily need to be in the database
itself; if your file is stored in the filesystem, but you have a file
name or some other pointer in the database, you can still arrange for
the data to be fetched and inserted into C<$r-E<gt>{output}>.

=head3 REST

You want to provide a programmatic interface to your Maypole site.

B<Solution>: The best way to do this is with C<REST>, which uses a
descriptive URL to encode the request. For instance, in L<Flox.pod> we
describe a social networking system. One neat thing you can do with
social networks is to use them for reputation tracking, and we can use
that information for spam detection. So if a message arrives from
C<person@someco.com>, we want to know if they're in our network of
friends or not and mark the message appropriately. We'll do this by
having a web agent (say, L<WWW::Mechanize> or L<LWP::UserAgent>) request
a URL of the form
C<http://flox.simon-cozens.org/user/relationship_by_email/person%40someco.com>.
Naturally, they'll need to present the appropriate cookie just like a
normal browser, but that's a solved problem. We're just interested in
the REST request.

The request will return a single integer status code: 0 if they're not
in the system at all, 1 if they're in the system, and 2 if they're our
friend.

All we need to do to implement this is provide the C<relationship_by_email>
action, and use it to fill in the output in the same way as we did when
displaying a picture. Since C<person%40someco.com> is not the ID of a
row in the user table, it will appear in the C<args> array:

    use URI::Escape;
    sub relationship_by_email :Exported {
        my ($self, $r) = @_;
        my $email = uri_unescape($r->{args}[0]);
        $r->{content_type} = "text/plain";
        my $user;
        unless (($user) = Flox::User->search(email => $email)) {
            $r->{content} = "0\n"; return;
        }

        if ($r->{user}->is_friend($user)) { $r->{content} = "2\n"; return; };
        $r->{content} = "1\n"; return;
    }

=head3 Component-based Pages

You're designing something like a portal site which has a number of
components, all displaying different bits of information about different
objects. You want to include the output of one Maypole request call while
building up another. 

B<Solution>: Use C<Maypole::Component>. By inheriting from this, you can
call the C<component> method on the Maypole request object to make a 
"sub-request". For instance, if you have a template

    <DIV class="latestnews">
    [% request.component("/news/latest_comp") %]
    </DIV>

    <DIV class="links">
    [% request.component("/links/list_comp") %]
    </DIV>

then the results of calling the C</news/latest_comp> action and template
will be inserted in the C<latestnews> DIV, and the results of calling
C</links/list_comp> will be placed in the C<links> DIV. Naturally, you're
responsible for exporting actions and creating templates which return 
fragments of HTML suitable for inserting into the appropriate locations.

Alternatively, if you've already got all the objects you need, you can
probably just C<[% PROCESS %]> the templates directly.

=head3 Bailing out with an error

Maypole's error handling sucks. Something really bad has happened to the
current request, and you want to stop processing now and tell the user about
it.

B<Solution>: Maypole's error handling sucks because you haven't written it
yet. Maypole doesn't know what you want to do with an error, so it doesn't
guess. One common thing to do is to display a template with an error message
in it somewhere.

Put this in your driver class:

    sub error { 
        my ($r, $message) = @_;
        $r->{template} = "error";
        $r->{template_args}{error} = $message;
        return OK;
    }

And then have a F<custom/error> template like so:

    [% PROCESS header %]
    <H2> There was some kind of error... </H2>
    <P>
    I'm sorry, something went so badly wrong, we couldn't recover. This
    may help:
    </P>
    <DIV CLASS="messages"> [% error %] </DIV>

Now in your actions you can say things like this:

    if (1 == 0) { return $r->error("Sky fell!") }

This essentially uses the template switcheroo hack to always display the
error template, while populating the template with an C<error> parameter.
Since you C<return $r-E<gt>error>, this will terminate the processing
of the current action.

The really, really neat thing about this hack is that since C<error>
returns C<OK>, you can even use it in your C<authenticate> routine:

    sub authenticate {
        my ($self, $r) = @_;
        $r->get_user;
        return $r->error("You do not exist. Go away.")
            if $r->{user} and $r->{user}->status ne "real";
        ...
    }

This will bail out processing the authentication, the model class, and
everything, and just skip to displaying the error message. 

Non-showstopper errors or other notifications are best handled by tacking a
C<messages> template variable onto the request:

    if ((localtime)[6] == 1) {
        push @{$r->{template_args}{messages}}, "Warning: Today is Monday";
    }

Now F<custom/messages> can contain:

    [% IF messages %]
    <DIV class="messages">
    <UL>
        [% FOR message = messages %]
           <LI> [% message %] </LI>
        [% END %]
    </UL>
    </DIV>
    [% END %]

And you can display messages to your user by adding C<PROCESS messages> at an
appropriate point in your template; you may also want to use a template
switcheroo to ensure that you're displaying a page that has the messages box in
it.

=head2 Authentication hacks

The next series of hacks deals with providing the concept of a "user" for
a site, and what you do with one when you've got one.

=head3 Logging In

You need the concept of a "current user".

B<Solution>: Use something like
C<Maypole::Authentication::UserSessionCookie> to authenticate a user
against a user class and store a current user object in the request
object.

C<UserSessionCookie> provides the C<get_user> method which tries to get
a user object, either based on the cookie for an already authenticated
session, or by comparing C<username> and C<password> form parameters
against a C<user> table in the database. Its behaviour is highly
customizable, so see the documentation, or the authentication paper at
C<http://maypole.simon-cozens.org/docs/authentication.html> for examples.

=head3 Pass-through login

You want to intercept a request from a non-logged-in user and have
them log in before sending them on their way to wherever they were
originally going.

B<Solution>:

    sub authenticate {
        my ($self, $r) = @_;
        $r->get_user;
        return OK if $r->{user};
        # Force them to the login page.
        $r->{template} = "login";
        return OK;
    }

This will display the C<login> template, which should look something
like this:

    [% INCLUDE header %]

      <h2> You need to log in </h2>

    <DIV class="login">
    [% IF login_error %]
       <FONT COLOR="#FF0000"> [% login_error %] </FONT>
    [% END %]
      <FORM ACTION="/[% request.path%]" METHOD="post">
    Username: 
        <INPUT TYPE="text" NAME="[% config.auth.user_field || "user" %]"> <BR>
    Password: <INPUT TYPE="password" NAME="password"> <BR>
    <INPUT TYPE="submit">
    </FORM>
    </DIV>

Notice that this request gets C<POST>ed back to wherever it came from, using
C<request.path>. This is because if the user submits correct credentials,
C<get_user> will now return a valid user object, and the request will pass
through unhindered to the original URL.

=head3 Logging Out

Now your users are logged in, you want a way of having them log out
again and taking the authentication cookie away from them, sending
them back to the front page as an unprivileged user.

B<Solution>: This action, on the user class, is probably overkill, but
it does the job:

    sub logout :Exported {
        my ($class, $r) = @_;
        # Remove the user from the request object
        my $user = delete $r->{user};
        # Destroy the session
        tied(%{$r->{session}})->delete;
        # Send a new cookie which expires the previous one
        my $cookie = Apache::Cookie->new($r->{ar},
            -name => $r->config->{auth}{cookie_name},
            -value => undef,
            -path => "/"
            -expires => "-10m"
        );
        $cookie->bake();
        # Template switcheroo
        $r->template("frontpage");
    }

=head3 Multi-level Authentication

You have both a global site access policy (for instance, requiring a
user to be logged in except for certain pages) and a policy for
particular tables. (Only allowing an admin to delete records in some
tables, say, or not wanting people to get at the default set of methods
provided by the model class.) 

You don't know whether to override the global C<authenticate> method or
provide one for each class.

B<Solution>: Do both. Have a global C<authenticate> method which calls 
a C<sub_authenticate> method based on the class:

    sub authenticate {
        ...
        if ($r->{user}) {
            return $r->model_class->sub_authenticate($r)
                if $r->model_class->can("sub_authenticate");
            return OK;
        }
        ...
    }

And now your C<sub_authenticate> methods can specify the policy for
each table:

    sub sub_authenticate { # Ensure we can only create, reject or accept
        my ($self, $r) = @_;
        return OK if $r->{action} =~ /^(issue|accept|reject|do_edit)$/;
        return;
    }

=head2 Creating and editing hacks

These hacks particularly deal with issues related to the C<do_edit>
built-in action.

=head3 Limiting data for display

You want the user to be able to type in some text that you're later
going to display on the site, but you don't want them to stick images in
it, launch cross-site scripting attacks or otherwise insert messy HTML.

B<Solution>: Use the C<CGI::Untaint::html> module to sanitize the HTML
on input. C<CGI::Untaint::html> uses C<HTML::Sanitizer> to ensure that
tags are properly closed and can restrict the use of certain tags and
attributes to a pre-defined list.

Simply replace:

    App::Table->untaint_columns(
        text      => [qw/name description/]
    );

with:

    App::Table->untaint_columns(
        html      => [qw/name description/]
    );

And incoming HTML will be checked and cleaned before it is written to
the database.

=head3 Getting data from external sources

You want to supplement the data received from a form with additional
data from another source.

B<Solution>: Munge the contents of C< $r-E<gt>params > before jumping
to the original C<do_edit> routine. For instance, in this method,
we use a C<Net::Amazon> object to fill in some fields of a database row based
on an ISBN:

    sub create_from_isbn :Exported {
       my ($self, $r) = @_;
       my $response = $ua->search(asin => $r->{params}{isbn});
       my ($prop) = $response->properties;
       # Rewrite the CGI parameters with the ones from Amazon
       @{$r->{params}{qw(title publisher author year)} =            
           ($prop->title,
           $prop->publisher,
           (join "/", $prop->authors()),
           $prop->year());
       # And jump to the usual edit/create routine
       $self->do_edit($r);
    }

The request will carry on as though it were a normal C<do_edit> POST, but
with the additional fields we have provided.

=head3 Uploading files and other data

You want the user to be able to upload files to store in the database.

B<Solution>: It's messy.

First, we set up an upload form, in an ordinary dummy action. Here's
the action:

    sub upload_picture : Exported {}

And here's the template:

    <FORM action="/user/do_upload" enctype="multipart/form-data" method="POST">

    <P> Please provide a picture in JPEG, PNG or GIF format:
    </P>
    <INPUT TYPE="file" NAME="picture">
    <BR>
    <INPUT TYPE="submit">
    </FORM>

(Although you'll probably want a bit more HTML around it than that.)

Now we need to write the C<do_upload> action. At this point we have to get a
little friendly with the front-end system. If we're using C<Apache::Request>,
then the C<upload> method of the C<Apache::Request> object (which
C<Apache::MVC> helpfully stores in C<$r-E<gt>{ar}>) will work for us:

    sub do_upload :Exported {
        my ($class, $r) = @_;
        my $user = $r->{user};
        my $upload = $r->{ar}->upload("picture");

This returns a C<Apache::Upload> object, which we can query for its
content type and a file handle from which we can read the data. It's
also worth checking the image isn't going to be too massive before we
try reading it and running out of memory, and that the content type is
something we're prepared to deal with. 

    if ($upload) {
        my $ct = $upload->info("Content-type");
        return $r->error("Unknown image file type $ct")
            if $ct !~ m{image/(jpeg|gif|png)};
        return $r->error("File too big! Maximum size is ".MAX_IMAGE_SIZE)
            if $upload->size > MAX_IMAGE_SIZE;

        my $fh = $upload->fh;
        my $image = do { local $/; <$fh> };

Now we can store the content type and data into our database, store it
into a file, or whatever:

        $r->{user}->photo_type($ct);
        $r->{user}->photo($image);
    }

And finally, we use our familiar template switcheroo hack to get back to
a useful page:

        $r->objects([ $user ]);
        $r->{template} = "view";
    }

Now, as we've mentioned, this only works because we're getting familiar with
C<Apache::Request> and its C<Apache::Upload> objects. If we're planning to use
C<CGI::Maypole> instead, or want to write our application in a generic way so
that it'll work regardless of front-end, then we need to replace the C<upload>
call with an equivalent which uses the C<CGI> module to get the upload data.
This is convoluted and horrific and we're not going to show it here, but it's
possible.

Combine with the "Displaying pictures" hack above for a happy time.