Make boot/ocamlc read directly byterun/primitives

[bobot]

boot/* are often responsible for merge conflict in github or rebase conflict during development, because they are binaries. People needs to modifies boot/ocamlc each time the set of primitives in the default runtime is modified.

This merge request proposes to use for boot/ocamlc the option -use-prims byterun/primitives. It allows to add new primitive and use it in the standard library without bootstrapping. Compilation with ./ocamlc,./ocamlopt are not modified.

For example you can compile with the following patch without the errors:

File "_none_", line 1:
Error: Error while linking boot/stdlib.cma(Weak):
The external function `caml_test_add_prim' is not available

diff --git a/byterun/weak.c b/byterun/weak.c
index 3614d11..29e1845 100644
--- a/byterun/weak.c
+++ b/byterun/weak.c
@@ -190,3 +190,8 @@ CAMLprim value caml_weak_blit (value ars, value ofs,
   }
   return Val_unit;
 }
+
+CAMLprim value caml_test_add_prim (value ar, value n)
+{
+  return ar;
+}
diff --git a/stdlib/weak.ml b/stdlib/weak.ml
index 71385c9..5dc37b4 100644
--- a/stdlib/weak.ml
+++ b/stdlib/weak.ml
@@ -303,3 +303,7 @@ module Make (H : Hashtbl.HashedType) : (S with type data = H.t) = struct
   ;;

 end;;
+
+external test_add_prim: int -> int -> int = "caml_test_add_prim"
+
+let _ = test_add_prim 1 2

I think that work also for removing some primitives but I don't know which exactly.

This could wait after 4.03 since it only simplifies compiler development.

[gasche]

I don't know enough of the compiler build machinery to understand if that is correct, but I warmly welcome the change. Indeed bootstrapping on primitive changes is a pain and makes some sorts of contributions much more delicate. (It also generates conflicts that makes handling out-of-tree branches painful, etc.)

Thanks for this very productive work.

[damiendoligez]

@gerdstolpmann should look at this, as it's getting close to cross-compilation territory.

Anyway, let's wait for 4.04, there's a lot going on now and I don't want to risk breaking the makefiles at this time.

[gerdstolpmann]

Basically, we have primitives that are hard-coded in the compiler (e.g. look at the beginnning of Translcore, it assumes caml_equal and others), and primitives that are used in the standard library. The -use-prims option is correct for the latter. My thinking is that there should also be a list of hard-coded primitives that is merged with the loaded list.

But anyway, given that there are rarely new hard-coded primitives the new approach is better than the old one. I don't see any problems with cross-compilation. On the contrary, host and target compilers could now even deviate with respect to the primitives.

[alainfrisch]

Thanks! Merged (manually -- somehow Git got overly confused while trying to merge, apparently by Windows EOL in .nt files).