Ian Bicking: the old part of his blog

Smart vs. Dumb Templates

Sometimes people think that layout is simple. You need a big, smart language like Python to do programming -- but something more limited will do for the layout and rendering. In fact, there's usually a little arrogance in there -- the templating language should be simple, and it should be underpowered, because you want to hobble the template writers lest they escape into the Dark Forest of Programming, taking maintainability with them.

At first it seems like this would work. But all programs start out simple (at least if you are writing them correctly). I think people believe layout can be simple because rendering is sloppy. Good rendering doesn't lead to "You have 0 message(s)" (and if you're going to be sloppy, at least make it obvious by saying "Messages: 0" -- use natural language right, or just don't use it at all).

There's lots of other layout issues. Many pages render improperly when the content is too short, or has long lines, and so on. And these are problems just for rendering relatively static material, as opposed to interactive web applications.

Okay... so people do sloppy programming, that's fine. Sloppy output is a much smaller sin than other kinds of programmer sloppiness. But a templating language should be able to stay with you from sloppiness to refinement. If your template language is underpowered you're putting up a future barrier.

And instead, layout starts creeping back into the "control" code. You have to provide the template with partially rendered input, or precalculate values for consumption only by your templates. You can separate hard programming from easy programming, or you can separate logic from layout, but you cannot do both.

Things become even worse if you are dealing with an interactive web application. A web UI is a subtle thing, and prone to overall suckiness. (It is ironic that I am a web programmer, yet I rather dislike web interfaces) "Layout" and "display" become very complicated -- if you are writing your UI well! A simple display gives people only small snippets of information at a time, leads to low-content interstitial pages, forces information traversal instead of nested display, and lacks helpful or predictive direction. A good UI involves all sorts of little rules, exceptions, warnings, qualified feedback and compound screens. Wimpy template languages are not up to the job. Though you might ask: are any template langauges up to the job?

PyDS (which I use to publish this weblog) uses Cheetah Templates for much of its display. Cheetah is a full-featured templating language (perhaps the most full featured of Python templating options, though empy might be close). PyDS uses quite a few templates to control the process -- but they are very dense. Some sorts of layout control should be easy, like rearranging elements or tweaking the granuarity of CSS classes. But reading the PyDS templates isn't easy, they are dense with code and difficult to understand. So I'm not sure, should template languages be powerful, or should we leave them dumb and couple them with Python code?

(I might add: PyDS templates, or other complex templates, might be simpler to work with if they used Cheetah's inheritance features; a Cheetah editor would also be neat ;)

Created 31 Oct '03
Modified 14 Dec '04


Actually I tried to use the Cheetah inheritance feature in the start of PyDS. I failed miserably - don't know wether it was a bug or my stupidity or both, though ;-)

There were some weird things about the inheritance that got me to change over to the way I now do it. The templates are actually quite simple - there are *Template.tmpl files, that describe your desktop. There are *Rendering.tmpl files that describe your clould. There are Base*.tmpl files that describe your outer layout (where to place what). All other templates are for tools and carry the tool name in their filename.

Templates are in a very simple way: the tools push their contents into the tool template. What comes out goes into the base template. The rest is just pushing around objects and prefilling the template namespace.

So for layout changes you mostly work on the Base*.tmpl files, as they define your layout. The others are more "push this out of the code" templates. But the template situation is still a bit weird with PyDS, as not all areas are templated (many tools don't use templates but could make use of them: for example the aggregator should be changed to templates one day, so you could influence your aggregator layout better).
# Georg Bauer

I'm on the fence when it comes to templating. My first instinct is to say that presentation layers should only wory about presentation. i.e., where bits should go and how they should look. Everything else should be code, because it involves data processing and non-presentation logic. Thus, one does not have to work in two (or more!) distinct programming languages and deal with the inevitable impedence mismatches that result. So by this argument, template languages should be a fancy syntax for nothing more than simple string concatination and replacement engines.

However, the flaw in this seems to be that many people view a templating language as a vehicle for non-programmers to act like programmers without getting into too much trouble when they make mistakes. So this mind set uses wants template languages to be a way of adding safe, managed code to projects by potentially unsafe developers.
And down that path seems to be maddness.
# Kevin Jacobs

But my entire point was that a simple templating language isn't up to the task of presentation. If you want the templating language to handle presentation, it must be a powerful and complete language, because presentation isn't an easy problem. I don't think tools should be crippled in order to keep people from doing what they shouldn't -- carrots are better than sticks.

PHP, for instance, has "powerful" templates, with nothing except templates. So stuff gets all mixed up. But the problem isn't that the templates in PHP are too powerful. Rather they lack the carrot to keep people from putting code in templates. In Zope DTML was a mess, but that had a lot to do with the lack of Python Scripts. You still get growth issues, but those are refactoring issues. When things get messy you refactor them -- that's the natural development process, and messiness is an inevitable side-effect of development. But those things should be fixed -- the dumb template alternative is to avoid messiness and refactoring by keeping people from doing anything at all, or by forcing them to create sub-standard presentations.
# Ian Bicking

After many attempts I would say: reduce templates to a minimun. In the past I've tried ASP.NET, XSLT and Cheetah (with Webware) to separate code and layout, but I always ended up to have a mess in the (X)HTML. So recently i started experimenting with XHTML and DOM manipulation.

Most of the times what I really need is a "Master Page" concept that hold common elements shared among site pages and a way to children pages to replace blocks in the mater pages with specilized content. This is what Cheetah does, but it should really stop there.

For example in Python it's very easy to generate a [select] and [options] fields plus-item-currently-selected with a list comprehension, but in Cheetah it'is not that compact, you have to use a #for, an #if/#else but doing this broke the WYSIWYG HTML editor. So you have a language that looks quite like Python but lacks of the most advanced (and useful) features. I don't like it.

Also delegate all the presentation to CSS helps a lot. You can changes fonts, background colors and, sometimes, elements position via CSS bypassing XHTML altogether.

To my knownledge there are a couple similar approaches in Python: PyMeld and HTMLTemplate.

# dee

FYI: there's a useful analysis of templates at http://wact.sourceforge.net/index.php/TemplateView
# Harry Fuecks

I've worked with non-Python programmers and templating systems in the past. And I see your point about sufficient power in the templating system.

There are a number of considerations involving templating languages that do not have to do with power.

One is the nature to do a form of round trip. Can a template be given back to a designer (or designer's software) and can this designer then make adjustments without getting hopelessly confused or breaking the logic of the template all the time? Both ZPT and Twisted Woven are attempts in this direction. Lots of program code in a template can break this.

The other, which is a recent insight I'm still exploring, is how much of the program has to be exposed to the template designer at all. Most templating systems allow calling into the program's API, which can have a complex object model below it. This has two problems:

* the template designer can be exposed to lots of aspects of the API that this designer has nothing to do with. It's not clear whether the designer needs most of this information; the designer may get confused or the designer may cause parts of the program to be called that were not intended to be called from the template directly at all.

* Bugs in the software are exposed to the template designer. The template designer can create perfect template logic, and be exposed to errors in the software that have nothing to do with his own performance as a template designer. The template
designer is exposed to complex error messages in the application logic.

I think that a template designer needs enough power but should not have to deal with the junk in the underlying system so much. I also suspect the template language does not need *that* much power if the data that is fed into the template already is prepared for various circumstances; i.e. Python logic on a view.

I don't mind if the designer writes parts of the view code (outside the template), too. It will probably result in better factored templates as common expressions can disappear into simple view methods.

Most view systems are 'pull models'; API calls pull the information from
the object model. The problems I mentioned above are hard to avoid in those. There are also view systems with aspects of a 'push model'; information is pushed outside the application into a bundle that can be easily accessed by the view, and which only contains the information needed by the views. Bugs in application logic will often be exposed in the push phase, not afterwards. Streams of data will be combined before they reach the view, so it's not the job of the view to query various part of the program directly, only what is produced by the first phase.

What is pushed out can be strictly defined. XML pipeline
models such as in cocoon have push properties, and are therefore interesting in this respect. The stage before the view template produces an XML
document with the relevant information to display the view. Programmers
can debug what appears in this stage without bothering the view
developer, who only has access to the output of this stage.

# Martijn Faassen

Have you looked at Zope Page Templates
# David Sumner

A case can be made that most of these so-called "underpowered" templates are actually 1) *overpowered*, and 2) based on the wrong paradigm.

By contrast, check out the templating in PSO (Python Service Objects), Twisted's Woven, and my own peak.web.templates. Each is based on the idea of designating portions of output as being rendered by Python components. But, the components aren't responsible for generating actual output markup. That is, the Python components don't know anything about markup language, they simply manipulate markup provided to them as parameters.

From this approach, you can do anything that a typical templating language does, but you're forced to do it modularly, in Python. The end result is that there's no code in your template, and no template in your code. Instead, UI developers simply choose from available structural manipulations to apply to the markup. The result is VB-like, in that you just add controls and parameterize them (with markup). If new controls are needed, you write them in Python, producing the best of both worlds.
# Phillip J. Eby

i've been using the perl templating module HTML::Template, for years and have never felt at all limited by it, even though it probably falls closer to the Dumb end of the spectrum.

as i've started working more in python, i discovered an underappreciated benefit of HTML::Template: language independence. the template syntax itself isn't tied to perl or any other language and clones exist for python, php, java, and probably other languages. so the designers i work with haven't had to learn a new template syntax just because i've changed to a new application language, and i've been able to successfully port apps from perl to python without touching the template files.

also, perrin harkins has a pretty good discussion of various templating systems (perl oriented but the basic ideas carry over): http://www.perl.com/pub/a/2001/08/21/templating.html
# anders

We've discussed this internally many times and I personally have come to the conclusion that your system should have an API that your delivery layer calls. The delivery layer is the combined code and templates needed to render the componenent that goes onto the screen. Each of these components then gets called/included/whatever to place it onto the screen framework.

so we get

1) Code to handle data
2) Code to convert data into form ready for displaying (date formats, ordering, paging whatever)
3) Markup to render the converted data
4) Page build that combined different componenets together

# Tim Parkin

All of my new projects use a very smooth combination of XSLT and CSS, and I like it a lot. Most of the logic is in my PHP code, but then some of the display logic can just be handled using XSLT stuff. XSLT may not be useful to graphic designers, but it's perfect for programmers.
# Manuzhai

Check out
Enforcing ModelView Separation in Template Engines, a recently-posted paper fro Terence Parr, author of ANTLR (ANother Tool for Language Recognition, formerly PCCTS) and co-founder of jGuru (and a former boss of mine, by way of jGuru).
# Tripp Lilley

Argh. Want. Editable. Comments.
# Tripp Lilley

"you want to hobble the template writers lest they escape into the Dark Forest of Programming, taking maintainability with them." - What a nice saying, Ian! ROFL!!!

"You can separate hard programming from easy programming, or you can separate logic from layout, but you cannot do both." - Aha! Now I understand why the problems of template languages is one of the most painful problem I know. Thank you!

BTW, I use CheetahTemplate for rendereing http://phd.pp.ru/ , and I use the full power of inheritance. See http://phd.pp.ru/about/technical.html . Look at the templates for an every page.

# Oleg Broytmann