Hat Logo


Frequently Asked Questions

Why do you call Hat a tracer rather than a debugger?

The word debugger suggests: firstly that there is a fault to be found; and secondly, that the tool will fix it. In actual fact, the tracer is just a tool for observing what happened in any particular program run. You can use it to find the cause of a fault, but equally, you can use it simply to aid your understanding of a correct program. While the tracer will help you to find bugs, it cannot find them for you. And once the bug is found, you must fix it, the tracer cannot!

Why did you not implement a conventional tracer/debugger as they are well known for imperative languages?

Conventional tracers (debuggers) for imperative languages allow the user to step through the program computation, stop at given points and examine variable contents. In our opinion this tracing method is unsuitable for a lazy functional language such as Haskell, because its evaluation order is complex, function arguments are usually unwieldy large unevaluated expressions, and generally computation details do not match the user's high-level view of functions mapping values to values. Also the stack used by the Haskell runtime system does not resemble the stack of function calls used by runtime systems of strict languages.

However, the trace file written by the traced program contains all information in the order of lazy evaluation. It is possible to write a viewing tool that admits single stepping through a lazy computation.

How much does tracing cost me in time and space?

Currently a traced program runs between 50-150 times slower than its untraced counterpart, depending which compiler and options you use. We are working on reducing this factor. As a rule of thumb, a reduction step produces about 50 bytes in the trace file. So trace files can easily grow to tens and hundreds of megabytes. However, you can name some modules of your program as trusted, so that their reductions are not recorded in the trace.

I don't seem to be able to build Hat-2.02 with ghc-6.x

The hat-2.02 package is quite old now and does not build cleanly with ghc ≥ 6.0. All the build problems with ghc-6.x are fixed in the newer release hat-2.04.

I have trouble compiling my program for tracing.

Hat covers the complete Haskell'98 language and its libraries, or at least that is the intention. It also supports a few language extensions, such as the FFI, multi-parameter type class, functional dependencies, and so on. First make sure that your program is suitable for tracing with Hat [ check our Feature Summary to see exactly which language extensions are supported ], and if it still won't compile, report it as a bug in Hat.

Compiling for tracing gives me an error like: Ambiguous type variable `a' in top-level constraint `Num a'

You have been hit by Haskell's defaulting rules for numeric literals, which unfortunately means you need to add one or more type signatures to your code. In the rejected module, add the line "default ()", and recompile it normally (not for tracing). You will now see an error of similar form to the error you saw when compiling for tracing, except that the line number refers to your original code. Fix this error by adding a type signature, e.g. (0::Int) instead of just 0. Now if your original code compiles cleanly again, it will also compile properly for tracing.

hat-trans fails with an error like: Cannot handle field update with more than 2 labels; attempted arity 11

Hat has some numeric limits on the size of certain constructs it can handle. Usually, there is a simple syntactic change you can make to your original source code to get round the limitation. For instance:

The latest updates to these pages are available on the WWW from http://www.haskell.org/hat/

This page last modified: 12th October 2004
York Functional Programming Group