simonfxr

I guess you mean Futamura projections?

Okay you are right, GHC does not produce a stable ABI with dynamic libraries. But it could do so in principal, no problem with laziness itself. It just would be a lot slower without all the cross module inlining and so GHC chooses not to. The archlinux haskell packagers could just ship programs as statically linked binaries and not recompile them on each minor dependency upgrade (except if there is a security upgrade). The binaries would not have any dependency on haskell packages themselves and you would not be forced to upgrade the whole package graph when some dependency is bumped, that would also save a lot of download bandwidth. I'm not sure that anyone really needs haskell packages (besides standalone programs) from the system package manager. No haskell dev I know builds their programs this way.

I don't see how strict vs. lazy has anything to do with linkage. I guess you are talking about foreign functions defined in a dynamic library. Calling a foreign function (regardless of linkage) will just evaluate the arguments and pass them according to the calling convention of the foreign ABI. If you link to library dynamically and it gets updated (in an ABI compatible way) you don't have to recompile anything, it just works as you would expect.