This makes me thinking: Can we employ this "technique" (or some variation of it) to have an *interpreter-only* version of TH for cross-compiled targets? Many TH use-cases do not necessitate the loading and running of native libraries, and e.g. for generating some lenses the interpreter-only TH solution would be a great progress already. So, instead of adding the TH-feature to cross-compilation, one could strip a normal stage-2 build from its ability to load TH dylibs, and see how far we get compiling `lens` (this reminds me of https://www.fpcomplete.com/user/edwardk/editorial/procrustean-mathematics#ce...). Then we could try to figure out what it takes to transfer this feature set back to cross targets. Just my 2 cents, Gabor On 9/1/14, Sven Panne <svenpanne@gmail.com> wrote:
2014-09-01 9:26 GMT+02:00 Simon Marlow <marlowsd@gmail.com>:
Hi Sven - you would need to compile the module with -dynamic or -dynamic-too to have it be picked up by the new dynamically-linked GHCi in 7.8.
Ah, OK... Adding -dynamic makes this work, but with -dynamic-too, ghci still loads the interpreted version. I didn't follow the linking story in detail, so this is all a bit confusing. I think that at least http://www.haskell.org/ghc/docs/7.8.3/html/users_guide/ghci-compiled.html needs some update, because as it is, it doesn't reflect reality. _______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs