Simon Peyton-Jones wrote:
I don't think there is any reason in principle why GHC can't generate DLLs that "just work", but plainly it's deficient at the moment.
The fundamental reason is that the DLL mechanism itself doesn't allow initialization/ shutdown do be hidden from the user of a DLL, because the order of loading/unloading DLLs is unspecified by the MS Windows operating system. Therefore it's not a GHC problem - it's just a fact of life with DLLs...
The other thing is that we could do with more "how to" documentation about GHC and DLLs. We have a bit here: http://haskell.org/haskellwiki/GHC/Using_the_FFI Could you expand it in the light of your experience?
I've added a section http://haskell.org/haskellwiki/GHC/Using_the_FFI#Beware_of_DllMain.28.29.21 which attempts to explain why init/exit code can't be put in DllMain, as well as an *untested* example to show what I meant by putting init/exit code in Begin/End functions instead. I've put a note next to the example to say that it's untested, and to appeal to anyone with more knowledge of static linking between Haskell and C to fix anything I've missed or replace with a better example. (Apologies for not testing it but static linking to stubs generated by GHC seems to require the DLL to be compiled by gcc and use of gcc on Windows seems to require Cygwin since gcc doesn't seem to understand Windows paths and I don't have Cygwin installed - unfortunately at the moment I don't have time to pursue this further.) Best regards, Brian. -- http://www.metamilk.com