[maypole.git] / doc / Flox.pod

=head1 Flox: A Free Social Networking Site

Friendster, Tribe, and now Google's Orkut - it seems like in early 2004,
everyone wanted to be a social networking site. At the time, I was too
busy to be a social networking site, as I was working on my own project
at the time - Maypole. However, I realised that if I could implement a
social networking system using Maypole, then Maypole could probably do
anything.

I'd already decided there was room for a free, open-source networking
site, and then Peter Sergeant came up with the hook - localizing it to
universities and societies, and tying in meet-ups with restaurant
bookings. I called it Flox, partially because it flocks people together
and partially because it's localised for my home town of Oxford and its
university student population.

Flox is still in, uh, flux, but it does the essentials. We're going to
see how it was put together, and how the techniques shown in the
L<Request.pod> chapter can help to create a sophisticated web
application. Of course, I didn't have this manual available at the time,
so it took a bit longer than it should have done...

=head2 Mapping the concepts

Any Maypole application should start with two things: a database schema,
and some idea of what the pages involved are going to look like.
Usually, these pages will be tying to displaying or editing some element
of the database, so these two concepts should come hand in hand.

When I started looking at social networking sites, I began by
identifying the concepts which were going to make up the tables of the
application. At its most basic, a site like Orkut or Flox has two
distinct concepts: a user, and a connection between two users.
Additionally, there's the idea of an invitation to a new user, which can
be extended, accepted, declined or ignored. These three will make up the
key tables; there are an extra two tables in Flox, but they're
essentially enumerations that are a bit easier to edit: each user has an
affiliation to a particular college or department, and a status in the
university. (Undergraduate, graduate, and so on.)

For this first run-through, we're going to ignore the ideas of societies
and communities, and end up with a schema like so:

    CREATE TABLE user (
        id int not null auto_increment primary key,
        first_name varchar(50),
        last_name varchar(50),
        email varchar(255),
        profile text,
        password varchar(255),
        affiliation int,
        unistatus int,
        status ENUM("real", "invitee"),
        photo blob,
        photo_type varchar(30)
    );

    CREATE TABLE connection (
        id int not null auto_increment primary key,
        from_user int,
        to_user int,
        status ENUM("offered", "confirmed")
    );

    CREATE TABLE invitation (
        id char(32) not null primary key,
        issuer int,
        recipient int,
        expires date
    );

Plus the definition of our two auxilliary tables:

    CREATE TABLE affiliation (
        id int not null auto_increment primary key,
        name varchar(255)
    );

    CREATE TABLE unistatus (
        id int not null auto_increment primary key,
        name varchar(255)
    );

Notice that, for simplicity, invitations and friendship connections are
quite similar: they are extended from one user to another. This means
that people who haven't accepted an invite yet still have a place in the
user table, with a different C<status>. Similarly, a connection between
users can be offered, and when it is accepted, its status is changed to
"confirmed" and a reciprocal relationship put in place.

We also have some idea, based on what we want to happen, of what pages
and actions we're going to define. Leaving the user aside for the
moment, we want an action which extends an invitation from the current
user to a new user. We want a page the new user can go to in order to
accept that invitation. Similarly, we want an action which offers a
friendship connection to an existing user, and a page the user can go to
to accept or reject it. This gives us five pages so far:

    invitation/issue
    invitation/accept

    user/befriend
    connection/accept
    connection/reject

Notice that the C<befriend> action is performed on a user, not a
connection. This is distinct from C<invitation/issue> because when
befriending, we have a real user on the system that we want to do
something to. This makes sense if you think of it in terms of object
oriented programming - we could say

    Flox::Connection->create(to => $user)

but it's clearer to say

    $user->befriend

Similarly, we could say

    Flox::User->create({ ... })->issue_invitation_to

but it's clearer to say

    Flox::Invitation->issue( to => Flox::User->create({ ... }) )

because it more accurately reflects the principal subject and object of
these actions.

Returning to look at the user class, we want to be able to view a user's
profile, edit one's own profile, set up the profile for the first
time, upload pictures and display pictures. We also need to handle the
concepts of logging in and logging out.

As usual, though, we'll start with a handler class which sets up the
database:

    package Flox;
    use base qw(Apache::MVC);
    Flox->setup("dbi:mysql:flox");
    Flox->config->{display_tables} = [qw[user invitation connection]];
    1;

Very simple, as these things are meant to be. Now let's build on it.

=head2 Authentication

The concept of a current user is absolutely critical in a site like
Flox; it represents "me", the viewer of the page, as the site explores
the connections in my world. We've described the authentication hacks
briefly in the L<Request.pod> chapter, but now it's time to go into a
little more detail about how user handling is done.

XXX

We also want to be able to refer to the current user from the templates,
so we use the overridable C<additional_data> method to give us a C<my>
template variable:

    sub additional_data { 
        my $r = shift; $r->{template_args}{my} = $r->{user}; 
    }

I've called it C<my> rather than C<me> because we it lets us check 
C<[% my.name %]>, and so on.

=head2 Viewing a user

The first page that a user will see after logging in will be their own
profile, so in order to speed development, we'll start by getting a
C<user/view> page up.

The only difference from a programming point of view between this action
and the default C<view> action is that, if no user ID is given, then we
want to view "me", the current user. Remembering that the default view
action does nothing, our C<Flox::User::view> action only needs to do
nothing plus ensure it has a user in the C<objects> slot, putting
C<$r-E<gt>{user}> in there if not:

    sub view :Exported {
        my ($class, $r) = @_;
        $r->{objects} = [ $r->{user} ] unless @{$r->{objects}||[]};
    }

Maypole, unfortunately, is very good at making programming boring. The
downside of having to write very little code at all is that we now have
to spend most of our time writing nice HTML for the templates.

XXX

The next stage is viewing the user's photo. Assuming we've got the photo
stored in the database already (which is a reasonable assumption for the
moment since we don't have a way to upload a photo quite yet) then we
can use the a variation of the "Displaying pictures" hack from the 
Requests chapter:

    sub view_picture :Exported {
        my ($self, $r) = @_;
        my $user = $r->{objects}->[0] || $r->{user};
        if ($r->{content_type} = $user->photo_type) {
           $r->{output} = $user->photo;
        } else {
           # Read no-photo photo
           $r->{content_type} = "image/png";
           $r->{output} = slurp_file("images/no-photo.png");
        }
    }

We begin by getting a user object, just like in the C<view> action: either
the user whose ID was passed in on the URL, or the current user. Then
we check if a C<photo_type> has been set in this user's record. If so,
then we'll use that as the content type for this request, and the data
in the C<photo> attribute as the data to send out. The trick here is
that setting C<$r-E<gt>{output}> overrides the whole view class processing
and allows us to write the content out directly.

In our template, we can now say

    <IMG SRC="/user/view_picture/[% user.id %]">

and the appropriate user's mugshot will appear.

However, if we're throwing big chunks of data around like C<photo>, it's
now worth optimizing the C<User> class to ensure that only pertitent
data is fetched by default, and C<photo> and friends are only fetched on
demand. The "lazy population" section of C<Class::DBI>'s man page
explains how to group the columns by usage so that we can optimize
fetches:

    Flox::User->columns(Primary   => qw/id/);
    Flox::User->columns(Essential => qw/status/);
    Flox::User->columns(Helpful   => qw/ first_name last_name email password/)
    Flox::User->columns(Display   => qw/ profile affiliation unistatus /);
    Flox::User->columns(Photo     => qw/ photo photo_type /);

This means that the status and ID columns will always be retrieved when
we deal with a user; next, any one of the name, email or password
columns will cause that group of data to be retrieved; if we go on to
display more information about a user, we also load up the profile,
affiliation and university status; finally, if we're throwing around
photos, then we load in the photo type and photo data.

These groupings are somewhat arbitrary, and there needs to be a lot of
profiling to determine the most efficient groupings of columns to load,
but they demonstrate one principle about working in Maypole: this is the
first time in dealing with Maypole that we've had to explicitly list the
columns of a table, but Maypole has so far Just Worked. There's a
difference, though, between Maypole just working and Maypole working
well, and if you want to optimize your application, then you need to
start putting in the code to do that. The beauty of Maypole is that you
can do as much or as little of such optimization as you want or need.

So now we can view users and their photos. It's time to allow the users
to edit their profiles and upload a new photo.

=head2 Editing users

XXX Editing a profile

I introduced Flox to a bunch of friends and told them to be as ruthless
as possible in finding bugs and trying to break it. And break it they
did; within an hour the screens were thoroughly messed up as users had
nasty HTML tags in their profiles, names, email addresses and so on. 
This spawned another hack in the request cookbook: "Limiting data for
display". I changed the untaint columns to use C<html> untainting, and
all was better:

    Flox::User->untaint_columns(
        html      => [qw/first_name last_name profile/],
        printable => [qw/password/],
        integer   => [qw/affiliation unistatus /],
        email     => [qw/email/]
    );

The next stage was the ability to upload a photo. We unleash the "Uploading
files" recipe, with an additional check to make sure the photo is of a
sensible size:

    use constant MAX_IMAGE_SIZE => 512 * 1024;
    sub do_upload :Exported {
        my ($class, $r) = @_;
        my $user = $r->{user};
        my $upload = $r->{ar}->upload("picture");
        if ($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> };

            use Image::Size;
            my ($x, $y) = imgsize(\$image);
            return $r->error("Image too big! ($x, $y) Maximum size is 350x350")
                if $y > 350 or $x > 350;
            $r->{user}->photo_type($ct);
            $r->{user}->photo($image);
        }

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

Now we've gone as far as we want to go about user editing at the moment.
Let's have a look at the real meat of a social networking site: getting
other people involved, and registering connections between users. 

=head2 Invitations

We need to do two things to invitations working: first provide a way to
issue an invitation, and then provide a way to accept it. Since what
we're doing in issuing an invitation is essentially creating a new
one, we'll use our usual practice of having a page to display the form
to offer an invitation, and then use a C<do_edit> method to actually do
the work. So our C<issue> method is just an empty action:

    sub issue :Exported {}

and the template proceeds as normal:

    [% PROCESS header %]
    <h2> Invite a friend </h2>

    <FORM ACTION="[%base%]/invitation/do_edit/" METHOD="post">
    <TABLE>

Now we use the "Catching errors in a form" recipe from L<Request.pod> and
write our form template:

    <TR><TD>
    First name: <INPUT TYPE="text" NAME="forename"
    VALUE="[%request.params.forename%]">
    </TD>
    <TD>
    Last name: <INPUT TYPE="text" NAME="surname"
    VALUE="[%request.params.surname%]"> 
    </TD></TR>
    [% IF errors.forename OR errors.surname %]
        <TR>
        <TD><SPAN class="error">[% errors.forename %]</SPAN> </TD>
        <TD><SPAN class="error">[% errors.surname %]</SPAN> </TD>
        </TR>
    [% END %]
    <TR>
    ...

Now we need to work on the C<do_edit> action. This has to validate the
form parameters, create the invited user, create the row in the C<invitation>
table, and send an email to the new user asking them to join.

We'd normally use C<create_from_cgi> to do the first two stages, but this time
we handle the untainting manually, because there are a surprising number of
things we need to check before we actually do the create. So here's the
untainting of the parameters:

    my ($self, $r) = @_;
    my $h = CGI::Untaint->new(%{$r->{params}});
    my (%errors, %ex);
    for (qw( email forename surname )) {
        $ex{$_} = $h->extract(
                "-as_".($_ eq "email" ? "email" : "printable") => $_
        ) or $errors{$_} = $h->error;
    }

Next, we do the usual dance of throwing the user back at the form in case
of errors:

    if (keys %errors) {
        $r->{template_args}{message} = "There was something wrong with that...";
        $r->{template_args}{errors} = \%errors;
        $r->{template} = "issue";
        return;
    }

We've introduced a new template variable here, C<message>, which we'll use
to display any important messages to the user.

The first check we need to do is whether or not we already have a user
with that email address. If we have, and they're a real user, then we
abort the invite progress and instead redirect them to viewing that user's 
profile.

    my ($user) = Flox::User->search({ email => $ex{email} });
    if ($user) {
        if ($user->status eq "real") {
            $r->{template_args}{message} =
                "That user already seems to exist on Flox. ".
                "Is this the one you meant?";

            $self->redirect_to_user($r,$user);
        } 

Where C<redirect_to_user> looks like this:

    sub redirect_to_user {
        my ($self, $r, $user) = @_;
        $r->{objects} = [ $user ];
        $r->{template} = "view";
        $r->{model_class} = "Flox::User"; # Naughty.
    }

This is, as the comment quite rightly points out, naughty. We're currently
doing a C</invitation/do_edit/> and we want to turn this into a
C</user/view/xxx>, changing the table, template and arguments all at once.
To do this, we have to change the Maypole request object's idea of the model
class, since this determines where to look for the template: if we didn't,
we'd end up with C<invitation/view> instead of C<user/view>.

Ideally, we'd do this with a Apache redirect, but we want to get that
C<message> in there as well, so this will have to do. This isn't good practice;
we put it into a subroutine so that we can fix it up if we find a better way
to do it.

Anyway, this is what we should do if a user already exists on the system
and has accepted an invite already. What if we're trying to invite a user but
someone else has invited them first and they haven't replied yet?

         } else {
            # Put it back to the form
            $r->{template_args}{message} =
            "That user has already been invited; ".
            "please wait for them to accept";
            $r->{template} = "issue";
        }
        return;
    }

Race conditions suck.

Okay. Now we know that the user doesn't exist, and so can create the new 
one:

    my $new_user = Flox::User->create({
        email => $ex{email},
        first_name => $ex{forename},
        last_name  => $ex{surname},
        status => "invitee"
    });

We want to give the invitee a URL that they can go to in order to
accept the invite. Now we don't just want the IDs of our invites to
be sequential, since someone could get one invite, and then guess the
rest of the invite codes. We provide a relatively secure MD5 hash as
the invite ID:

    my $random = md5_hex(time.(0+{}).$$.rand);

For additional security, we're going to have the URL in the form
C</invitation/accept/I<id>/I<from_id>/I<to_id>>, encoding the user ids
of the two users. Now we can send email to the invitee to ask them to
visit that URL:

    my $newid = $new_user->id;
    my $myid  = $r->{user}->id;
    _send_mail(to => $ex{email}, url => "$random/$myid/$newid", 
               user => $r->{user});

I'm not going to show the C<_send_mail> routine, since it's boring.
We haven't actually created the C<Invitation> object yet, so let's
do that now.

    Flox::Invitation->create({
        id => $random,
        issuer => $r->{user},
        recipient => $new_user,
        expires => Time::Piece->new(time + LIFETIME)->datetime
    });

You can also imagine a daily cron job that cleans up the C<Invitation>
table looking for invitations that ever got replied to within their
lifetime:

   ($_->expires > localtime && $_->delete)
       for Flox::Invitation->retrieve_all;

Notice that we don't check whether the ID is already used. We could, but,
you know, if MD5 sums start colliding, we have much bigger problems on
our hands.

Anyway, now we've got the invitation created, we can go back to whence we
came: viewing the original user:

    $self->redirect_to_user($r, $r->{user});

Now our invitee has an email, and goes B<click> on the URL. What happens?

XXX

=head2 Friendship Connections

=head2