[Haskell-cafe] global variables

[Isaac Dupree]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Isaac Dupree wrote:
> Maybe some sort of ISOLATE, DON'T_OPTIMIZE (but CAF), or
> USED_AS_GLOBAL_VARIABLE pragma instead of just the insufficient NOINLINE
> would be a good first step... if successful it would remove the
> occasional need for -fno-cse for a whole module in GHC, at least.

ISOLATE, DON'T_OPTIMIZE are actually bad names for the whole effect,
which requires persistent CAF semantics.  An implementation that doesn't
make top-level definitions be CAFs, or even one that is willing to
garbage-collect them when memory is tight such that they need
recalculation later, would need a special case for global variables to
make them work.

i.e. I'm not sure if there exists a reasonable pragma while the code
still uses unsafePerformIO.

Hmm....

How about

so,
{-# NOINLINE var #-}
var :: IORef Int
var = unsafePerformIO (newIORef 3)

- -->

var :: IORef Int
var = {-# EVALUATE_THIS_TEXT_ONLY_ONCE #-} (unsafePerformIO (newIORef 3))

to capture the desired semantics: text-based uniqueness, no duplication,
no sharing of the IORefs (sharing the pure contents is fine), and no
need to actually evaluate it any times at all. {-#
EVALUATE_THIS_TEXT_ONLY_ONCE #-} is syntactically like a (special)
function.  Clearly it is an impossible demand for polymorphic things, so
the compiler could complain (at least a warning) if the (var :: IORef
Int) line was left off, for example. I guess it would also complain
about non-type(class) argument dependencies too such as (f x =
(unsafePerformIO (newIORef (x::Int))) )...

Food for thought :-)


Isaac
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFGU02YHgcxvIWYTTURAoCaAKCkDH7Pd7JbNt0TmNig9j7ujiUV9ACZAevI
QOjdmMbrPfVrKBafZshCh7c=
=9/5v
-----END PGP SIGNATURE-----