Support SystemTime with enabled isolation

[newpavlov]

Right now MIRI with enabled isolation fails in code which uses SystemTime::now() with the following error:

error: unsupported operation: `clock_gettime` with `REALTIME` clocks not available when isolation is enabled
   --> /home/rust/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/sys/pal/unix/time.rs:130:22
    |
130 |         cvt(unsafe { libc::clock_gettime(clock, t.as_mut_ptr()) }).unwrap();
    |                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ `clock_gettime` with `REALTIME` clocks not available when isolation is enabled
    |

It means that most loggers will trigger this error. In some cases it's possible to work around this by disabling timestamps, but ideally it would work out of the box.

MIRI already has emulation of monotonic clocks. I think it could be used to emulate sytem time as well by re-interpreting the monotonic clock as UNIX timestamp, i.e. MIRI tests would see timestamps for 1970-01-01.

[RalfJung]

We quite deliberately did not do that as it seemed even more confusing than the fake monotonic clock. The intended reaction to a program using SystemTime is -Zmiri-disable-isolation.

(Also, it's Miri, not MIRI. ;)

[RalfJung]

FWIW if you have a good motivation for why you want this with isolation enabled, we could be persuaded. I just stated our reasoning for not having done this already. :)

[newpavlov]

It's just a matter of convenience. It would be nice in my opinion to have this feature, but I understand your position above. I've asked the logger crate which we use in our tests to add support for Instant-based timestamps, which is an alternative way to resolve this.

Also, it's Miri, not MIRI.

I always thought it's an abbreviation of "Mid-level Intermediate Representation Interpreter". :)

[RalfJung]

I always thought it's an abbreviation of "Mid-level Intermediate Representation Interpreter". :)

That is indeed the origin of the name, but we don't write it in all-caps. :)