unknown RTS option: -xm20000000

[paulyoung]

If I try to build clash-ghc using Nix, clash-prelude fails in the checkPhase with unknown RTS option: -xm20000000 when trying to run doctests and unittests.

I'm using an M1 Macbook Pro.

If I do the following in nix/overlay.nix the error goes away.

       clash-prelude =
-        hprev.callCabal2nixWithOptions
+        hprev.callCabal2nix
           "clash-prelude"
           ../clash-prelude
-          "--flag workaround-ghc-mmap-crash"
           { };

I've tried to override this using various Nix mechanisms (e.g. override, overrideAttrs, a custom overlay, pkgs.haskell.lib.dontCheck) and while some of these work when building clash-prelude in isolation, it doesn't help when building clash-ghc for some reason that I don't understand.

[paulyoung]

If I do the following in nix/overlay.nix the error goes away.

I spoke too soon. While the build succeeds, actually trying to use clash results in a runtime error.

nix build .#clash-ghc
./result/bin/clash --help
clash: unknown RTS option: -xm20000000
...

[paulyoung]

I made an additional change to nix/overlay.nix and unsurprisingly the runtime error is gone.

       clash-ghc =
         let
           unmodified =
-            hprev.callCabal2nixWithOptions
+            hprev.callCabal2nix
               "clash-ghc"
-              ../clash-ghc
-              "--flag workaround-ghc-mmap-crash" {
+              ../clash-ghc {
               inherit (hfinal) clash-lib clash-prelude;
             };

nix build .#clash-ghc
./result/bin/clash --help
Usage:

    clash [command-line-options-and-input-files]

...

[paulyoung]

I've tried to override this using various Nix mechanisms

To clarify, what I meant here is that I have the flake in this repository as an input to my own flake.

[DigitalBrains1]

Thanks for the bug report and the steps to work around it!

It would seem the -xm flag is only on x86_64:
https://gitlab.haskell.org/ghc/ghc/-/blob/0df8ce27f1c418fee1ba860f1c6575f66cae2ca7/rts/RtsFlags.c?page=2#L1756

We had not realised this. We'll need to make it such that the flag is only passed on x86_64.

[paulyoung]

Thanks! I’ll probably use a fork in the meantime.

I think Rosetta should also allow people to work around this by doing something like:

…
let
  supportedSystems = [
    flake-utils.lib.system.aarch64-darwin
    flake-utils.lib.system.x86_64-darwin
  ];
in
  flake-utils.lib.eachSystem supportedSystems (system:
    let
      pkgs = import nixpkgs {
        system:
          if system == flake-utils.lib.system.aarch64-darwin
          then flake-utils.lib.system.x86_64-darwin
          else system;
      };
…

[rowanG077]

Should be fixed with #2657. I run aarch64 but didn't use nix flakes so I didn't hit this before.