Skip to content

bmillwood/stepeval

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Goal

Given a Haskell expression as an AST, evaluate it step-by-step, showing each step (e.g. pattern matching, function application) to the user. Intended to be useful as an educational and perhaps debugging tool.

Project post-mortem

I largely stopped working on stepeval some years ago. It turned out to be harder than I expected, and I eventually grew exasperated with the mediocre, buggy results. I still think the goal is a good goal, but I haven't yet returned to it with the lessons I learned to see if I could make a better attempt at it.

At the time I first started writing it, I had never written a compiler or interpreter before, and that lack of experience was probably costly – I set myself too many problems to solve at once. Someone with a better background in that department and possibly with the formal education I was (am) lacking would probably have a better shot at it, even if they took essentially the same approach.

Something I would have looked into before starting this again would be using a typed AST, rather than manipulating Exp too freely – there are seemingly just too many functions Exp → Exp for me to write the correct one! Evaluation, after all, should be type-preserving, so it would seem wasteful to forget that information. Stepeval as it exists today had no access to type information, but would have needed it anyway if it were to get support for type classes.

Another useful invariant is preservation of normal form, although that's perhaps a bit vacuous (and anyway, it would be nice to support stepping through expressions that don't have a normal form, like infinite lists).

With the typed angle in mind, my next project after stepeval was an step-by-step explainer of type inference, which could conceivably fall in scope as part of the same project. You could imagine a first pass which annotated everything with its type, and then a second part which evaluated the resulting expression. I didn't get much of anywhere with that project, anyway.

If I were to come back to this project, I'd probably scale the ambition back a bit and instead write a step-by-step evaluator for a small sublanguage of Haskell, say let + lambda + case. Indeed, I think that was my original intention, but then I decided to get haskell-src-exts to do my parsing for me and then I got carried away a bit.

Anyway, it would be good to get a minimal language with its stepper in place, because I think a major so-far-undiscussed challenge is coming up with a useful and intuitive UI for the whole thing. Just repeatedly dumping the entire AST can stop being helpful if you're evaluating some tiny subcomponent of a large, complex expression. I had a few different ideas for dealing with that, again varying in ambition. The most ambitious would be some kind of graphical (probably web) UI where you can click on any subexpression to reduce it one step, choosing whatever reduction order you liked (or asking the computer to show you what it would do next). Less ambitious would just be perhaps taking subexpressions out into let bindings and evaluating them outside of their context, but it's hard to know what heuristic to use for when to do that. Least ambitious would be, don't do anything clever and just live with the clumsiness :)

Let me know if you produce something interesting based on any of these ideas, or your own similar ones!

About this implementation

stepeval comes bundled with an executable that can operate as a command-line utility or a CGI script. In either case it takes an expression input as a string and displays each step of its evaluation, until it reaches weak head normal form or, in the case of the CGI script, a time limit expires.

Note many features of Haskell are unsupported and three of the tests fail, one quite basic! I recommend against relying on the accuracy of the results of this program.

If you want to see things in action without downloading anything, try http://bm380.user.srcf.net/cgi-bin/stepeval.cgi where an example instance of the CGI script is (maybe) running.

Contact

haskell at benmachine dot co dot uk

or via IRC: benmachine on Freenode or QuakeNet

Build notes

The version constraints are conservative - only things that I've personally tested with. If you are told your version of whatever doesn't fit into them, relax them; if it works, tell me!

TODO

Rewrite from scratch, after having read the literature about how to evaluate functional languages properly. Perhaps leaving it in syntax form is basically unsatisfactory, and the first thing I should do is a visualisation of the program's graph.

I need to think about how function application is done. Problem is we can't easily show how guards are stepped through etc. because we can't modify the original definition. Perhaps we introduce the funbind to a local let binding and then modify that.

The testsuite has no means for requiring that evaluation stop at a given point.

In function application, results aren't always shared as much as they should be. Basically evaluation is often non-strict instead of lazy, as in the classic example let double x = x + x in double (double 10)

Better support for primitive operations like arithmetic. This may be related to class instance resolution.

EnumFrom syntax is not at all supported

Type inference would be nice, to resolve class instances.

It would be nice to have a UI in which one could easily change either the expression to be stepped or the environment it is being stepped in.

Everywhere in the program I have emphasised correctness and simplicity of code over performance. In a lot of small ways and a few big ones the efficiency of the algorithms involved is much poorer than it needs to be.

FIXME

Nested let-scopes aren't properly supported.

If you use an external definitions file, the evaluator doesn't know that those definitions can't be changed, so pattern bindings that require evaluation may go wrong. This isn't really a big deal though because such files are intended mainly for functions, which don't have this problem.

About

A program for evaluating a Haskell expression step-by-step

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •