A program for evaluating a Haskell expression step-by-step
Haskell
Pull request Compare This branch is 14 commits behind bmillwood:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
src
testsuite
.gitignore
LICENSE
Makefile
README
Setup.hs
stepeval.cabal

README

stepeval is a tool that can operate as a command-line utility or CGI script.
In either case, it is given a Haskell expression in string form, applies a
single evaluation operation (e.g. pattern matching, lambda application), and
prints the result, while also feeding it back into its evaluation mechanism
until the expression cannot be evaluated any further, or in the case of the
CGI script a time limit has expired.

Contact: haskell at benmachine dot co dot uk
 or via IRC: benmachine on Freenode or QuakeNet

If you want to see things in action without downloading anything, try
http://www.srcf.ucam.org/~bm380/cgi-bin/stepeval.cgi
where an example instance of the CGI script is usually running.

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:

In recent times there's been a tiny little tendency to prop up the old
architecture in the interests of avoiding radical restructurings. As a
result some design decisions need serious re-examination.

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.