I have little constructive to say on the topic, except
that I typically use hugs in preference to "monstrous"
ghc/ghci. It's lightweight and its language is properly
documented (in the Haskell 98 report), neither of which
can be said of ghc. Its rudimentary instrumentation
(:set +s) is more useful than ghc's equivalent, though
admittedly ghc has many debugging features I haven't
explored.
That said, it should be noted that I do not use
Haskell (nor, nowadays, any other language) for
writing production code.
Apropos of language, a couple of years ago I noted
that ghc implements 2^99 languages, not one. (There
were 99 non-antonymous language pragmas; there may
be more now.) Who can know what terrors lurk there.
My first attempt to investigate the field (turn on
all 99) caused a compiler panic, since fixed.
Early on, I encountered a hugs bug: garbage collection
in the middle of a bignum operation caused havoc.
Having never looked at the hugs source before, I was
able quite quickly pinpoint the offending code; the
maintainers corrected it almost overnight. That
heart-warming experience hasn't worn off, even though
the maintainers have moved on.
Doug
