Hi
How much in Core has an "obvious" translation from Haskell, such that different implementations or versions, can compile the same Haskell code and get the exact same Core text out?
Not a whole lot. There is also the issue of optimisations, even the very basic ones will permute expressions around. You can get the same top-level declarations regardless, but that is about it.
If only local things like the details of how let/lambda-bound variables are named, vary, there is more possible separate compilation of separate files (or something like that...).
The top level interface can be made very solid, so separate compilation still works fine.
I think (line,column) is sufficiently unique identification? Maybe 200_4 would represent line 200, column 4. (or "200.4"? By itself, that looks like a fractional number, which is bad)
That may be sufficient, or not. Consider: f x = (a,b) where (a,b) = x This compiles to something like: f x = (fst x, snd x) However, fst and snd are actually lambda's, which could conceivably have the same source position. I do think line_col could be made to work though, with some thought, and it would be a very handy scheme. Thanks Neil