I'm probably re-inventing the wheel, but I'm currently writing a CGI script that returns web pages built from templating information. (An example of the template file is [link|http://yceran.org/static/staticpages.qf|here]* if you're curious.) The script itself is written in [link|http://unicon.sourceforge.net/|Icon], so things like strings, lists and tables are first-class datatypes and thus easy to use. There is extensive string manipulation available, too, for those who don't know anything about Icon.

One of the things I've determined I need in the template language is an RPN processor (a bit like that use in HP calculators) to enable some conditional control et al. I've got all sorts of options for implementing this, but I'm not happy with some aspect of most of them. Has anyone else ever implemented an RPN processor?

Wade.

* Argh! Your browser will probably try to render it as it has HTML in it. Grr. Just do a browse source if your browse does that. The mime type is correct, BTW; text/plain.

But Icon is not an interpreted language. It's a translated language (it has it's own runtime p-code interpreter). If it were an interpreted language, I probably would have done what you suggested, yes.

Actually, I sorta began to head down that way. It's straightforward to get a function pointer to any builtin operator at runtime and call it with whatever you like. I was going to do something like that for most of the operators. Among other things, it gives me access to functions like repl() which repeats a string a given number of times. My original extension of this idea would have had the code build a list of operations at parse time so when it got to expanding all the tokens, if it found a list where it expected a string, then it would do a simple loop through the list, calling function pointers and moving values around in a mechanical and conditional free way. Unfortunately, this approach requires some very creative algorithms at a few points to solve some data-hiding and edge-condition problems.

Wade.