-
Notifications
You must be signed in to change notification settings - Fork 76
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add opaqueness to Obj.magic under Flambda 2 #862
Conversation
1231148
to
4e2254d
Compare
You might want to patch line 2540 in |
Ooh thanks, have done that. |
18f6f13
to
2774504
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The typing change seems fine
25188da flambda-backend: Missed comment from PR802 (ocaml-flambda#887) 9469765 flambda-backend: Improve the semantics of asynchronous exceptions (new simpler version) (ocaml-flambda#802) d9e4dd0 flambda-backend: Fix `make runtest` on NixOS (ocaml-flambda#874) 4bbde7a flambda-backend: Simpler symbols (ocaml-flambda#753) ef37262 flambda-backend: Add opaqueness to Obj.magic under Flambda 2 (ocaml-flambda#862) a9616e9 flambda-backend: Add build system hooks for ocaml-jst (ocaml-flambda#869) 045ef67 flambda-backend: Allow the compiler to build with stock Dune (ocaml-flambda#868) 3cac5be flambda-backend: Simplify Makefile logic for natdynlinkops (ocaml-flambda#866) c5b12bf flambda-backend: Remove unnecessary install lines (ocaml-flambda#860) ff12bbe flambda-backend: Fix unused variable warning in st_stubs.c (ocaml-flambda#861) c84976c flambda-backend: Static check for noalloc: attributes (ocaml-flambda#825) ca56052 flambda-backend: Build system refactoring for ocaml-jst (ocaml-flambda#857) 39eb7f9 flambda-backend: Remove integer comparison involving nonconstant polymorphic variants (ocaml-flambda#854) c102688 flambda-backend: Fix soundness bug by using currying information from typing (ocaml-flambda#850) 6a96b61 flambda-backend: Add a primitive to enable/disable the tick thread (ocaml-flambda#852) f64370b flambda-backend: Make Obj.dup use a new primitive, %obj_dup (ocaml-flambda#843) 9b78eb2 flambda-backend: Add test for ocaml-flambda#820 (include functor soundness bug) (ocaml-flambda#841) 8f24346 flambda-backend: Add `-dtimings-precision` flag (ocaml-flambda#833) 65c2f22 flambda-backend: Add test for ocaml-flambda#829 (ocaml-flambda#831) 7b27a49 flambda-backend: Follow-up PR#829 (comballoc fixes for locals) (ocaml-flambda#830) ad7ec10 flambda-backend: Use a custom condition variable implementation (ocaml-flambda#787) 3ee650c flambda-backend: Fix soundness bug in include functor (ocaml-flambda#820) 2f57378 flambda-backend: Static check noalloc (ocaml-flambda#778) aaad625 flambda-backend: Emit begin/end region only when stack allocation is enabled (ocaml-flambda#812) 17c7173 flambda-backend: Fix .cmt for included signatures (ocaml-flambda#803) e119669 flambda-backend: Increase delays in tests/lib-threads/beat.ml (ocaml-flambda#800) ccc356d flambda-backend: Prevent dynamic loading of the same .cmxs twice in private mode, etc. (ocaml-flambda#784) 14eb572 flambda-backend: Make local extension point equivalent to local_ expression (ocaml-flambda#790) 487d11b flambda-backend: Fix tast_iterator and tast_mapper for include functor. (ocaml-flambda#795) a50a818 flambda-backend: Reduce closure allocation in List (ocaml-flambda#792) 96c9c60 flambda-backend: Merge ocaml-jst a775b88 flambda-backend: Fix ocaml/otherlibs/unix 32-bit build (ocaml-flambda#767) f7c2679 flambda-backend: Create object files internally to avoid invoking GAS (ocaml-flambda#757) c7a46bb flambda-backend: Bugfix for Cmmgen.expr_size with locals (ocaml-flambda#756) b337cb6 flambda-backend: Fix build_upstream for PR749 (ocaml-flambda#750) 8e7e81c flambda-backend: Differentiate is_int primitive between generic and variant-only versions (ocaml-flambda#749) git-subtree-dir: ocaml git-subtree-split: 25188da
We've hit a lot of cases where code using
Obj.magic
falls foul of the strict model of Flambda 2. There are almost certainly places we haven't fixed yet and that might be difficult to find. We could alleviate the risk ofInvalid
being produced by makingObj.magic
include opaqueness. This doesn't seem like it should de-optimize, since the argument is likely to be unknown.This PR follows the suggestion of @chambart that such opaqueness should only be used in the middle end, rather than compiling to
Copaque
in Cmm.There is a weirdness in the Menhir library included with the compiler as can be seen; warning 20 is generated using the new definition of
Obj.magic
(even if it is exposed asval
notexternal
inobj.mli
).