Creating a BodyFilter

A BodyFilter is just a derived class that implements some methods called by the proxy. Of all the methods presented below, only "filter()" must be defined in the derived class.

filter()

The signature of the filter() method is the following:

    sub filter {
        my ( $self, $dataref, $message, $protocol, $buffer ) = @_;
        ...
    }
    

where $self is the filter object, $dataref is a reference to the chunk of body data received, $message is a reference to either a HTTP::Request or a HTTP::Response object, and $protocol is a reference to the LWP::Protocol protocol object.

Note that this subroutine signature looks a lot like that of the call- backs of LWP::UserAgent (except that $message is either a HTTP::Request or a HTTP::Response object).

$buffer is a reference to a buffer where some of the unprocessed data can be stored for the next time the filter will be called (see "Using a buffer to store data for a later use" for details). Thanks to the built-in HTTP::Proxy::BodyFilter::* filters, this is rarely needed.

It is possible to access the headers of the message with "$message->headers()". This HTTP::Headers object is the one that was sent to the client (if the filter is on the response stack) or origin server (if the filter is on the request stack). Modifying it in the "filter()" method is useless, since the headers have already been sent.

Since $dataref is a reference to the data string, the referent can be modified and the changes will be transmitted through the filters that follows, until the data reaches its recipient.

A HTTP::Proxy::BodyFilter object is a blessed hash, and the base class reserves only hash keys that start with "_hpbf".

new()

The constructor is defined for all subclasses. Initialisation tasks (if any) for subclasses should be done in the "init()" method (see below).

init()

This method is called by the "new()" constructeur to perform all initisalisation tasks. It's called once in the filter lifetime.

It receives all the parameters passed to "new()".

begin()

Some filters might require initialisation before they are able to handle the data. If a "begin()" method is defined in your subclass, the proxy will call it before sending data to the "filter()" method.

It's called once per HTTP message handled by the filter, before data processing begins.

The method signature is as follows:

    sub begin {
        my ( $self, $message ) = @_
        ...
    }
    

end()

Some filters might require finalisation after they are finished handling the data. If a "end()" method is defined in your subclass, the proxy will call it after it has finished sending data to the "filter()" method.

It's called once per HTTP message handled by the filter, after all data processing is done.

This method does not expect any parameters.

will_modify()

This method return a boolean value that indicate if the filter will modify the body data on the fly.

The default implementation returns a true value.

Using a buffer to store data for a later use

Some filters cannot handle arbitrary data: for example a filter that basically lowercases tag name will apply a simple regex such as "s/<\s*(\w+)([^>]*)>/<\L$1\E$2>/g". But the filter will fail is the chunk of data contains a tag that is cut before the final ">".

It would be extremely complicated and error-prone to let each filter (and its author) do its own buffering, so the HTTP::Proxy architecture handles this too. The proxy passes to each filter, each time it is called, a reference to an empty string ($buffer in the above signature) that the filter can use to store some data for next run.

When the reference is "undef", it means that the filter cannot store any data, because this is the very last run, needed to gather all the data left in all buffers.

It is recommended to store as little data as possible in the buffer, so as to avoid (badly) reproducing what HTTP::Proxy::BodyFilter::complete does.

In particular, you have to remember that all the data that remains in the buffer after the last piece of data is received from the origin server will be sent back to your filter in one big piece.

The store and forward approach

HTTP::Proxy implements a store and forward mechanism, for those filters which need to have the whole message body to work. It's enabled simply by pushing the HTTP::Proxy::BodyFilter::complete filter on the filter stack.

The data is stored in memory by the "complete" filter, which passes it on to the following filter once the full message body has been received.

Standard BodyFilters

Standard HTTP::Proxy::BodyFilter classes are lowercase.

The following BodyFilters are included in the HTTP::Proxy distribution:

lines

This filter makes sure that the next filter in the filter chain will only receive complete lines. The "chunks" of data received by the following filters with either end with "\n" or will be the last piece of data for the current HTTP message body.

htmltext

This class lets you create a filter that runs a given code reference against text included in a HTML document (outside "<script>" and "<style>" tags). HTML entities are not included in the text.

htmlparser

Creates a filter from a HTML::Parser object.

simple

This class lets you create a simple body filter from a code reference.

save

Store the message body to a file.

complete

This filter stores the whole message body in memory, thus allowing some actions to be taken only when the full page has been received by the proxy.

tags

The HTTP::Proxy::BodyFilter::tags filter makes sure that the next filter in the filter chain will only receive complete tags. The current implementation is not 100% perfect, though.

Please read each filter's documentation for more details about their use.