asm is getting stabilized upstream and the feature is not strictly necessary anymore for the implementation.
Note that only the host side needs it, currently exclusively for the purpose of executing syscalls.
Remove miri and asm feature - we lose potential for testing against UB on CI
Disable miri for tests involving host side - greatly diminishes the benefit of actually using miri
Disable miri selectively for each test involving host side and inline assembly - this is going to be quite tedious to maintain
It's not broken, so don't fix it - we can simply wait until inline assembly is supported by miri. That may take long and that means we have to maintain the feature within the implementation, but that is actually fairly trivial, since the host simply ignores the syscalls when feature is disabled. Note, that it is possible that will have some mechanism in place for selectively disabling/enabling host-side call handlers at runtime in the future, if we had such functionality, we could then simply disable all calls requiring asm when miri is enabled and drop the feature.
For now, I vote for 4. and suggest to revisit this in a few sprints.
asmis getting stabilized upstream and the feature is not strictly necessary anymore for the implementation. Note that only the host side needs it, currently exclusively for the purpose of executing syscalls.Unfortunately
miridoes not support inline assembly at the time of writing (https://github.com/rust-lang/miri/issues/1045, https://github.com/rust-lang/miri/issues/11 is the tracking issue)Few options here:
miriandasmfeature - we lose potential for testing against UB on CImirifor tests involving host side - greatly diminishes the benefit of actually usingmirimiriselectively for each test involving host side and inline assembly - this is going to be quite tedious to maintainmiri. That may take long and that means we have to maintain the feature within the implementation, but that is actually fairly trivial, since the host simply ignores the syscalls when feature is disabled. Note, that it is possible that will have some mechanism in place for selectively disabling/enabling host-side call handlers at runtime in the future, if we had such functionality, we could then simply disable all calls requiringasmwhenmiriis enabled and drop the feature.For now, I vote for 4. and suggest to revisit this in a few sprints.
Refs https://github.com/rust-lang/rust/pull/91728 https://github.com/enarx/enarx/issues/1178