Tue, 18 Sep 2001 01:56:45 -0400
> I can't think of a way to use unsafePtrCompare safely :-) The relative
> ordering of objects isn't guaranteed to be stable under GC.
Doh, that would throw a monkey wrench into things, wouldn't it? I know of
compacting GC algorithms, but I didn't consider that GHC might be using one.
At least I am now more enlightened on the inner workings of the magic beast...
I've considered many of the other implementation options, but as it isn't
essential to the working of the compiler, it hasn't been a priority yet. It
simply struck me that this would be a particularly quick and easy way to
implement reasonably good atom tables, only requiring a newtype declaration
and a few very simple function definitions.
Thanks to Simon for saving me from reinventing the wheel. The libraries
mentioned here should prove to be quite useful.
One's intuition would suggest that you could be safely implement mkAtom
without wrapping it in a IO monad. After all, at least at a abstract level,
an atom table is referentially transparent. The library documentation says
that lack of side effects and environmental independance is sufficent to
order for uses of unsafePerformIO to be safe. Is there a exact (or at least
better) criterion for safety?
"unsafePerformIO" is used in the implementation of mkFastString, so how is
it's side effects "safe". I experimented with unsafePerformIO with my Atom
table, but I could not get tthe code to work properly.