On Thu, Jan 26, 2012 at 5:54 AM, Tassilo Horn <tass...@member.fsf.org> wrote: > Cedric Greevey <cgree...@gmail.com> writes: > >> On Thu, Jan 26, 2012 at 2:28 AM, Tassilo Horn <tass...@member.fsf.org> wrote: >>> At least, it seems that (fn ^double [^double x] (+ x 0.5)) compiles to a >>> class supporting primitive, unboxed calls. >> >> Yes. That was my conclusion also. But apparently the only way to get >> the compiler to emit such a call is by binding the fn to a global Var >> with the appropriate hints and calling that. > > Yeah, I know. And now I can also see it in the Compiler's InvokeExpr > class. In its constructor, there's some extra work in case of a VarExpr > where the signature :tag metadata is pulled out from the :arglists > metadata. > > Why is the :arglists metadata added to the Var a function's bound to > instead of the function itself?
AFAICT, the compiler will need to know the call is definitely primitive and the fn is definitely primitive-compatible at compile time if it is to safely emit a primitive call. This is why I tried a ^Ifn$DD hint on the variable at one point. When the fn comes from a non-dynamic Var the compiler can easily be sure what primitive signatures it supports. When the fn is potentially runtime-variable, it seems to me that a hint on the local *should* suffice, and type inference from assignments *should* be possible, as it is for locals assigned primitive values. (If the hint was wrong, you'd get runtime ClassCastExceptions such as my_ns.my_defn$fn_3072 cannot be cast to clojure.lang.IFn$DD, much as you can get various CCEs with incorrect (reference) type hints elsewhere.) However, this seems to be either broken or not implemented yet. -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en
