The issue uses consistency with -odir and -hieDir as a rationalization. I think it's not quite right because inputs can have a search path while outputs don't, but if it is true that -odir simultaneously sets the output and input dir for `*.o`, that seems confusing too. Shouldn't -odir set *only* output, and all inputs should either be lumped under the `-i` search path, or even split up into separate flags as outputs are? If you are building in --make mode then you will naturally want to read .o files from the last place you wrote them, but in that case you can explicitly say so. I'm not actually advocating for this since I don't have a use for it at the moment, and I don't want to potentially break every single user of ghc --make, but it seemed to me that the argument for consistency could run the other direction.
I think this issue is coming up due to a mismatch between the traditional Make style builds which assume a single mutable input and output directory, and the newer functional / hermetic style builds which require immutable inputs and a mutable output. It's a familiar concept, just transposed to the file system! _______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs
