[Phphtmllib-devel] A Critique

[David M. <da...@re...>]

Hi all,

I fear that I might come across as putting down phpHtmlLib, so please
try and trust that that is not my intention.

I love the concept of phpHtmlLib, that of abstracting the generation of
HTML code.

But after working with it for 2 days, I've gone back to printing out raw
html for now.

In using phpHtmlLib, I've found that the abstractions can be more fiddly
to drive than raw html generation itself.

At first, I felt really enthusiastic, and started converting a web-based
project management application over to phpHtmlLib.

My app generates some intricate HTML - tables nested 3 deep, rows and
columns full of links which vary according to php script invocation
flags and database content. More, link and non-link column styles -
colours, italics, bold etc - also vary according to script arguments and
database content.

I gave up trying to convert the app. Instead, I started writing a
completely new script, borrowing slightly from the logic of the raw one.

But on writing the phpHtmlLib code, I found myself having to create tons
of temporary variables to keep track of things.

It all started to feel like programming in assembler. The abstraction
mechanisms of phpHtmlLib came at a price - a whole lot of concrete
tracking of details, and a whole lot of 'noise' in the script, which
completely destroyed its readibility. The code ended up far more
concrete than before.

[It's a shame we don't have 3D displays. Then, logic indenting could go
left-right, and tag indenting could go front-back. Or perhaps someone
with too much time on their hands could write a LISP major mode script
for phpHtmlLib to represent different levels of tags as different
colours']

But seriously, it would be unfair for me to just rant on without
offering some kind of ideas for how this could be fixed. Well, here
goes...

1) Create a 'cursor' object, so that every time a tag object gets
created, the cursor gets set to that object. Then, one could do
something like 'add("A line of text")', and know the text will be
written to the right place, without having to keep track of what the
current tag is.
2) Complementary to (1) - a function allowing an arbitrary 'pop' back to
any level of the stack. For example, if one is currently writing an <a>
within a <div> within a <td> within a <tr> within a <table> within a
<td> within a <tr>, one could pop the 'cursor' back to the outer level
'<tr>' tag.
3) Make reference the standard default type of 'push'. For situations
where someone really wants to do the 'copying'-style push, provide a tag
object copying function (or method) which completely copies a tag object
and all its children. With methods such as 'push_row()', the lack of a
reference push is a big drawback.
4) Allow for easy creation of template classes.
For example, it would be lovely to have a class which writes 10-point
bold green centred links within a table column, without having to create
n levels of tags each time.

I'll sleep on it tonight. I feel very tempted to have a go at creating
things like this myself, but I don't want to jump in without thinking at
depth how it could be done well.

Your thoughts?
Cheers
David