From: msergeant@... (Matt Sergeant)
Date: 2002-02-07T20:43:38+09:00
Subject: XML::SAX2 critique

This isn't exactly a critique of XML::SAX2 per-se, but more of the
tack taken in that project...

XML::SAX2 seems to be an almost verbatim copy of the Java SAX2 APIs. I
strongly believe that they have made a very poor choice in their API
by passing the node information to the callbacks as a list of
parameters. Here's an example in Ruby:

  handler.start_element(uri, localname, qname, attrs)

Now while the element data is passed as individual parameters, the
attrs are passed as an object (of the Attributes class). This is a
nasty design "smell". It also broke things for Java when they went
from SAX1 to SAX2, because the APIs were different, so they had to
change the name of the handler classes to avoid confusion and
breakage.

Now you may say "but XML is at 1.0 and isn't likely to change", but
that's a falsehood, because XML Schemas extend the infoset with type
information, which you cannot represent using the above SAX notation.

Now picture a cleaner API:

  handler.start_element(el)

Where element is a lightweight class Element [1], which encapsulates
the above information and the attributes. Simple to construct, but
most importantly: extensible. This is how the Perl SAX APIs have been
designed, right from the start, and it meant that Perl SAX1 was mostly
compatible with Perl SAX2 [2].

I really urge people to consider the implications of something like
this - it's a major design decision - if you carve in stone the Ruby
SAX2 API now, and copy the Java design, you may regret it.

I hope you don't see me as being purely critical. I think Maki has
done an amazing job in putting together all this code. I just don't
want you to lose out on a better option.

Matt.

[1] Could be as simple as:

class Element
  attr_reader :uri, :qname, :localname, :attrs
  attr_writer :uri, :qname, :localname, :attrs
  def initialize(uri, qname, localname, attrs)
    @uri = uri
    @qname = qname
    @localname = localname
    @attrs = attrs
  end
end

[2] http://sax.perl.org/