[maypole.git] / doc / View.pod

=head1 Maypole View Classes

In a large application, you will almost certainly want to customize the
layout and design of the output pages. This task may even be the purview
of a separate team of HTML designers rather than the programmers. Since
a typical programmer will try to avoid touching HTML as much as possible
and a typical designer will try to avoid touching Perl code, programmers
have evolved a system of templating to separate the concerns of
programming and designing. 

One of the core concepts in Maypole is the I<view class>, and this is
responsible for routing the data produced in the model class into the
templates produced by the designers. Of course, there are a great many
possible templating systems and styles, and so there can be a great many
possible Maypole view classes. Each view class will take the data from
the controller, locate a template to be processed, and hand the whole
lot to its preferred templating module, which will then do the hard work
of filling in the template and coming up with the output.

You can choose whatever Maypole view class you want, but the default
view class is C<Maypole::View::TT>, and it feeds its data and templates
to a module called the Template Toolkit.

=head2 The Template Toolkit

The Template Toolkit, written by Andy Wardley, is a very powerful and
generic templating system. It provides its own little formatting language
which supports loops, conditionals, hash and array dereferences and
method calls, macro processing and a plug-in system to connect it to
external Perl modules. There are several good introductions to the
Template Toolkit available: you should have one installed as
L<Template::Tutorial::Datafile>; there's one at
L<http://www.perl.com/pub/a/2003/07/15/nocode.html>, and of course
there's the "Badger Book" - I<The Perl Template Toolkit>, by Andy et al.

We'll present a brief introduction here by deconstructing some of the 
templates used by Maypole applications. For more deconstruction, see
L<StandardTemplates.pod>, which is an entire chapter dealing with the
factory supplied templates.

Here's the template which is called for the front page of the standard
beer database application, C<custom/frontpage>.

    [% INCLUDE header %]

    <h2> The beer database </h2>

    <TABLE BORDER="0" ALIGN="center" WIDTH="70%">
    [% FOR table = config.display_tables %]
    <TR>
    <TD>
    <A HREF="[%table%]/list">List by [%table %]</A>
    </TD>
    </TR>
    [% END %]
    </TABLE>

The first thing to note about this is that everything outside of the
Template Toolkit tags (C<[%> and C<%]>) is output verbatim. That is,
you're guaranteed to see 

    <h2> The beer database </h2>

    <TABLE BORDER="0" ALIGN="center" WIDTH="70%">

in the output somewhere. Inside the tags, magic happens. The first piece
of magic is the C<[% INCLUDE header %]> directive. This goes away and
finds a file called F<header> - don't worry about how it finds that yet,
we'll come to that later on - and processes the file's contents as
though they were right there in the template. Our F<header> file happens
not to contain any C<[% %]> tags, but if it did, they would be processed
in the same way as the ones in F<frontpage>. 

The next piece of magic is this line:

    [% FOR table = config.display_tables %]

We're seeing a lot of things here at once. C<config> is where we should
start looking. This is a template variable, which is what templates are
all about - templating means getting data from somewhere outside and
presenting it to the user in a useful way, and the C<config> hash is a
prime example of data that we want to use. It's actually the hash of
configuration parameters for this Maypole application, and one of the
keys in that hash is C<display_tables>, the database tables that we're
allowed to play with. In the application, we probably said something
like

    BeerDB->config->{display_tables} = [qw[beer brewery pub style]];

This stores the four values - C<beer>, C<brewery>, C<pub> and C<style> -
in an array, which is placed in the config hash under the key
C<display_tables>. Now we're getting them back again.

The Template Toolkit's dot operator is a sort of do-the-right-thing
operator; we can say C<array.0> to get the first element of an array,
C<hash.key> to look up the C<key> key in a hash, and C<object.method> to
call C<method> on an object. So, for instance, if we said
C<config.display_tables.2>, we'd look up the C<display_tables> key in
the configuration hash and get our array back, then look up the 2nd
element and get C<pub>.

The C<FOR> loop will repeat the code four times, setting our new
variable C<table> to the appropriate array element. This code:

    [% FOR table = config.display_tables %]
        Hello [% table %]!
    [% END %]

will produce something like

    Hello beer!
    Hello brewery!
    Hello pub!
    Hello style!

In our case, though, we're printing out a table element linking to each
database table in turn.

Here's a slightly more complicated example, adapted from F<factory/pager>.
This template is responsible for printing the little page menu at the
bottom of a listing if there are more rows in the listing than we want
on a single page.

    [% PROCESS macros %]
    <P ALIGN="center">Pages:
    [%
         FOREACH num = [pager.first_page .. pager.last_page];
              IF num == pager.current_page;
                "["; num; "] ";
              ELSE;
                SET args = "?page=" _ num;
                SET label = "[" _ num _ "]";
                link(classmetadata.moniker, "list", args, label);
              END;
         END;
    %]
    </P>

Maypole will be providing a whole bunch of variables to this template,
and we'll look at them all in a moment, but the only ones we need to care
about are C<pager> and C<classmetadata>. 

We start by loading in a bunch of macros. Macros are Template Toolkit's
functions - you can provide them some parameters and they'll run a little
sub-template based on them. The C<macros> file contains some handy macros
that I've found useful for constructing Maypole templates; again, these
will be covered in full detail in L<StandardTemplates.pod>.

We're going to be displaying something like this:

    Pages: [1] [2] [3] [4]

with most of those numbers being a link to the appropriate page. This
mean we're going to have to have a list of numbers, and the C<FOREACH> loop
provides this: (C<FOREACH> and C<FOR> are identical, just like in Perl.)

         FOREACH num = [pager.first_page .. pager.last_page];

Here we're manually constructing an array of numbers, using the range
operator (C<..>) to fill in all the numbers from the C<first_page> (1)
to the C<last_page> (4). The same dot operator is used to ask the C<pager>
what its C<first_page> and C<last_page> is. Remember when we said 
C<config.display_tables>, we were looking up the C<display_tables> key
in the C<config> hash? Well, this time we're not looking anything up in
a hash. C<pager> is an object, and C<first_page> is a method. Thing is,
you don't have to care. You can pretend it's a hash if you want. The
syntax is the same, and Template Toolkit knows the right thing to do.

Now we're going to be executing this loop four times, once each for C<num>
being set to 1, 2, 3, and 4. At some point, we'll come across the page
that we're actually on right now:

      IF num == pager.current_page;

and in that case, we don't want to produce a link to it. We just want
to output it as text, surrounded by square brackets:

                "["; num; "] ";

We're using string literals to output the brackets. We don't have to
do that. We could say it this way:

    [% ...
      IF num == pager.current_page;
    %]
        [ [% num %] ] 
    [% ELSE %]
       ...
    [% END %]

But you know, I quite like it my way.

Now if the number we're printing isn't the number of the current page,
we want to make a link. Here's how we do it:

    SET args = "?page=" _ num;
    SET label = "[" _ num _ "]";
    link(classmetadata.table, "list", args, label);

C<SET> declares a new variable of our own. If there was anything called
C<args> before, there isn't now. It's going to be the result of our
statement C<"?page=" _ num>. C<_> is the concatenation operator, and
glues C<?page=> onto the front of our number. So if we want to link to
page 4, then the C<args> variable will contain C<?page=4>. Similarly,
the C<label> variable will be C<[4]>.

Now we call a macro, C<link> with these two variables and the value of
C<classmetadata.table>. This macro takes four arguments, C<table>, 
C<action>, C<args> and C<label>, and constructs a link of the form

    <A HREF="[% config.base_url %]/[% table %]/[% action %][% args %]">
    [% label %]
    </A>

In our case, it'll be filled in like so:

    <A HREF="[% config.base_url %]/[% classmetadata.table %]/list?page=4">
    [ 4 ]
    </A>

Where C<classmetadata.table> will actually be the name of the current
table, and C<config.base_url> will be replaced by the appropriate URL for
this application.

=head2 Locating Templates

Another feature of C<Maypole::View::TT> which may not be present in
alternate view class implementations - although they are strongly
encouraged to provide it - is the way that templates are located.
(Remember, I B<did> say I'd tell you about that later.) Template Toolkit
allows whatever uses it to provide a path for template files to be
located in. C<Maypole::View::TT> feeds it up to three possible
directories to look things up in, and it will try to find a template in
each of these in turn.

When you configure a Maypole application, you can tell it the base
directory of your templates like so:

    BeerDB->config->{template_root} = "/var/www/beerdb/templates";

If you don't do this, most Maypole front-ends will use the current
directory, which is generally what you want anyway. Off this directory,
Maypole will look for a set of subdirectories.

For instance, I said we were in the middle of processing the front page
and looking up a template file called F<header>. Maypole will first look
for this file in the F<custom> subdirectory. (say,
F</var/www/beerdb/templates/custom>) If it doesn't find one, then it
looks in the F<factory> subdirectory. If it doesn't find one there, then
it gives up and dies with an error. But that's your fault, since you've
called for a template which doesn't exist. Don't do that. 

This behaviour means that you can provide your own site-specific
templates, but if you don't do so, then you get to use a generic one
provided by Maypole. Maypole's "factory setting" templates are written
in such a way as to try and do the right thing no matter what your
application does. They are occasionally successful at this. 

Now the front page was a pretty simple example, since Maypole only looks
up two directories. In most cases, it checks an additional directory,
and this directory depends entirely on what Maypole is doing.

If you're writing an e-commerce application, for example, you may well
have a table which represents the product catalogue and all the products
you can buy. Let's call this the C<product> table. You'll also have a
data source which is specific to the user which contains all the
products that they're buying on this particular visit to the site. In
time-honoured tradition, we'll call this the C<basket> table.

Now it ought to be reasonably apparent that you don't want the basket
to be displayed in exactly the same way as the product catalogue. The
templates for C<product/list> and C<basket/list> need to be different.
This is where the third directory comes in. The other directory, which
Maypole checks very first of all, is specific to the table that you're
viewing. So if you go to C<http://your.shop.com/basket/list>, Maypole
will look in the F<basket> directory first for a file called F<list>,
and second in the F<custom> directory for a site-wide list template,
and then fall-back to the F<factory> directory for a generic list
template. It should be obvious that you probably want to provide all
of F<basket/list>, F<basket/view>, F<product/list>, F<product/view>
and any other combination of classes and actions that you can think of.

=head2 What Maypole provides to a template

C<Maypole::View::TT> provides quite a variety of template variables to
the template. As these are the building blocks of your pages, it's worth
looking at precisely what variables are available.

The most important variable is called C<objects>, and is a list of all
the objects that this page is going to deal with. For instance,
in the template F</beer/view>, C<objects> will contain the C<BeerDB::Beer>
object for the 23rd item in the database, while F</brewery/list> will
fill C<objects> will all the breweries; or at least, all the breweries
on the current page.

This variable is so important that to help design templates with it,
C<Maypole::View::TT> provides a helpful alias to it depending on
context. For instance, if you're writing your own F</brewery/list>
template, the data in C<objects> is also available in a template
variable called C<breweries>. If you're working on F</brewery/view>,
though, it's available in C<brewery>, since there's only one brewery to
be displayed.

Additionally, you can get the base URL for the application from the
C<base> template variable; this allows you to construct links, as we
saw earlier:

    <A HREF="[% base %]/brewery/edit/[% brewery.id %]">Edit this brewery</A>

You can also get at the rest of the configuration for the site with the
C<config> variable as we saw above, and the entire request object in 
C<request>, should you really need to poke at it. (I've only found this
useful when working with authentication modules which stash a current user
object in C<request.user>.)

To allow the construction of the "generic" templates which live in
F<factory>, Maypole also passes in a hash called C<classmetadata>,
which contains all sorts of useful information about the class under
examination:

=over 3

=item C<table>

This is the name of the table that is represented by the class.

=item C<class>

This is the Perl's idea of the class; you don't need this unless you're
doing really tricky things.

=item C<moniker>

This is a more human-readable version of the table name, that can be
used for display.

=item C<plural>

The same, but a correctly-formed plural. For instance, "breweries".

=item C<columns>

The list of columns for display; see the section "Customizing Generic
CRUD Applications" in L<StandardTemplates.pod>.

=item C<colnames>

This is a hash mapping the database's name for a column to a more
human-readable name. Again, see "Customizing Generic CRUD Applications>.

=item C<cgi>

This is a slightly trickier one. It is a hash mapping column names to
a C<HTML::Element> suitable for entering data into a new instance of
that class. That is, for the C<beer> table, C<classmetadata.cgi.style>
should be a C<HTML::Element> object containing a drop-down list of
beer styles. This is explained in L<StandardTemplates.pod>.

=item C<description>

This is the human-readable description provided by a class.

=item C<related_accessors>

This is a list of accessors which can be called on an object to get
lists of other things that this object "has". For instance, on a
brewery, it would return C<beers>, since calling C<brewery.beers> would
give you a list of beers produced by the brewery. Note that this only
caters for accessors defining one-to-many relationships, not the
ordinary one-to-one relationships, such as C<style>.

=back

=head2 Other view classes

Please note that these template variables, C<config>, C<classmetadata>,
C<objects> and its user-friendly alias, as well as the rest of them are
a function of one particular view class, the default
C<Maypole::View::TT> class. Other view classes may need to present an
entirely different set of template variables, since the default ones
might not make sense. The templates may look wildly different in other
view class implementations. But that's OK, because you couldn't
necessarily use the same templates with a different templating system
anyway.

For instance, in really dumb templating languages which can't handle
dereferencing hashes or arrays - no wait, that's most of them - passing
in a hash reference like C<classmetadata> won't help you since you can't
get at any of its elements. So you'll need to take a look at the
documentation for the appropriate view class to see what template
variables it provides.

So if, for some perverse reason, the Template Toolkit just isn't good
enough for you, then you can set your own view class while configuring
your application:

   package BeerDB;
   use base 'Apache::MVC';
   ...
   BeerDB->setup("dbi:SQLite:t/beerdb.db");
   BeerDB->config->{uri_base} = "http://localhost/beerdb/";
   BeerDB->config->{rows_per_page} = 10;
   BeerDB->config->{view} = "Maypole::View::Mason"; 

Where do these alternate view classes come from? Gentle reader, they
come from B<you>.

=head2 Building your own view class

I<You should probably skip this section for the first few readings of this manual. It's only intended for people extending Maypole.>

Imagine you've found a brand new templating system that's B<much better>
than the Template Toolkit. I know I'm stretching your imagination a bit
here, but try. You'd like to use it with Maypole, which means writing your
own view class. How is it done?

We'll demonstrate by implementing a view class for C<HTML::Mason>,
although no value judgement is implied. C<HTML::Mason> is a templating
system which embeds pure Perl code inside its magic tags. The good side
of this is that it can get into hash references and objects, and so
providing C<classmetadata>, C<config> and the Maypole request object
will work out just fine. The down side is that C<HTML::Mason> is used to
running more or less standalone, and having all the template variables
it wants already at its disposal through CGI parameters and the like, so
we have to fiddle a bit to get these variables into our template.

The key to building view classes is C<Maypole::View::Base>. This is the
base class that you're going to inherit from and, to be honest, it does
pretty much everything you need. It provides a method called C<vars>
which returns a hash of all the template variables described above, so
it would be good to feed those into C<HTML::Mason>. It also provides a
C<paths> method which turns returns the full filesystem path of the
three possible template paths as shown above. Again, it would be good to
use this as our component paths if we can. It also has some methods we
can override if we want to, but they're not massively important, so you
can see L<Maypole::View::Base> for more about them. 

The module will do the right thing for us if we agree to provide a
method called C<template>. This is responsible for taking the Maypole
request object (of which more later) and putting the appropriate output
either into C<$r-E<gt>{output}> or C<$r-E<gt>{error}>, depending, of
course, whether things are OK or whether we got an error.

Thankfully, C<HTML::Mason> makes things really easy for us. We B<can>
use multiple template roots, so we can use the C<paths> method; we
B<can> pass in a hash full of interesting data structures, so we can use
the C<vars> method too. In fact, we have to do very little to make
C<Maypole::View::Mason> work. Which is somewhat annoying, because it
makes a boring example. But it means I can leave the fun ones to you!

The doing-the-templating, in Mason and in any templating system, depends on
three things: the paths that we're going to use to find our templates, the
template name that we've been asked to fill out, and the set of variables that
are going to be fed to the template. We'll assemble these for reference:

    sub template {
        my ($self, $r) = @_;
        my @paths = $self->paths($r);
        my $template = $r->template;
        my %vars = $self->args($r);

We'll also declare somewhere to temporarily store the output:

        my $output;

Now comes the part where we have to actually do something
templating-language specific, so we open up our copy of "Embedding Perl
in HTML with Mason" and find the bit where it talks about running Mason
standalone. We find that the first thing we need to do is create a
C<HTML::Mason::Interp> object which knows about the component roots.
There's a slight subtlety in that the component roots have to be
specified as an array of arrays, with each array being a two-element
list of label and path, like so:

    comproot => [
        [ class   => "/var/www/beerdb/templates/brewery" ],
        [ custom  => "/var/www/beerdb/templates/custom" ],
        [ factory => "/var/www/beerdb/templates/factory" ],
    ]

We also find that we can set the output method here to capture Mason's
output into a scalar, and also that we can tell Mason to generate
sensible error messages itself, which saves us from having to worry
about catching errors. At the end of all this, we come up with a
constructor for our C<HTML::Mason::Interp> object which looks like this:

    my $label = "path0";
    my $mason = HTML::Mason::Interp->new(
        comproot => [ map { [ $label++ => $_ ] } @paths ],
        output_method => \$output,
        error_mode => "output" 
    );

The next thing we need to do is run the template with the appropriate
template variables. This turns out to be really easy:

    $mason->exec($template, %vars);

Now we've got the data in C<$output>, we can put it into the request object,
and return a true value to indicate that we processed everything OK. (If there
was an error, then Mason will have produced some suitable output, so we can
pretend that everything's OK anyway.)

    $r->{output} = $output;
    return 1;

And that's all we need to do. Barely twenty lines of code for the finished
product. Wasn't that easy? Don't you feel inspired to write Maypole view
classes for your favourite templating language? Well, don't let me stop you!
Patches are always welcome!