[Haskell-cafe] Re: Non-technical Haskell question

[Simon Marlow]

On 06 December 2004 14:39, John Goerzen wrote:

> On 2004-12-06, Keith Wansbrough <Keith.Wansbrough at cl.cam.ac.uk> wrote:
>> The static vs dynamic linking question has been discussed many times.
>> The summary is: GHC is a highly-optimising compiler, and the binary
>> interface _necessarily_ changes with every minor revision (even
>> patchlevel revision) of the compiler and each library.  So you can't
> 
> We already have a way to deal with that using sonames.  It's not that
> hard, and is routinely used.
> 
> BTW, is this a problem even if no -O options are given when building a
> library?

Certainly it's possible to get a more stable library ABI by not
compiling with -O.  The question is, do you want to?  You certainly
don't want to do this for the Prelude and other essential libraries.
For example, GHC wouldn't be able to see the definition of 'instance Num
Int', and hence (x + 1 :: Int) would be compiled as 

  (+) dNumInt x (I# 1)

instead of (after inlining (+) and dNumInt):

  case x of I# x# -> I# (x# +# 1#)

which can then be further optimised: the compiler can see that it is
strict in x, so might be able to unbox x, and so on.  This is pretty
essential to get any kind of reasonable performance, but it might be
less important for higher-level libraries, as long as you don't have any
important monads that need to be inlined, for example.

Dynamic linking is (almost) a separate issue.  GHC 6.4 will have some
support for dynamic linking in the native code generator thanks to
Wolfgang Thaller, but it needs someone to push it the final mile on
x86/Linux and Windows.  Dynamically linked libraries will work (albeit
slightly slower than statically linked ones), but you still have the
versioning issue.  Since libraries are getting really big nowadays (eg.
wxHaskell) making dynamic linking work even without worrying about
versioning seems like it would be worthwhile.

Cheers,
	Simon