The API style is now more uniform: both are passed by mutable ref, and
only expose accessors. Additionally, the methods that were initially
accessed through the scheduler field are now directly implemented on
`Context`.
Now that `step_by` returns an error anyway (it was unfaillible before),
there is no more incentive to keep it as a separate method.
The `step_until` method now accepts an `impl Deadline`, which covers
both cases (`Duration` and `MonotonicTime`).
TODO: return the list of models involved in a deadlock.
Note that Many execution errors are not implemented at all at the
moment and will need separate PRs, namely:
- Terminated
- ModelError
- Panic
This makes it possible to concurrently control and monitor the
simulation when using gRPC.
Accordingly, the gRPC server now runs on 2 threads so it can serve
control and monitoring requests concurrently.
The `schedule_*event_in` and `schedule_*event_at` pairs of methods are
each merged into a single overloaded method accepting either a relative
`Duration`or an absolute `MonotonicTime`.
This is a pretty large patch that impacts the API.
Until now, it was not possible to cancel events that were scheduled for
the current simulation time slice, making it necessary for the user to
use complex workarounds (see former version of the espresso machine
example).
The new implementation makes this possible but the generation of a key
associated to an event has now a non-negligible cost (basicaly it
creates three references to an Arc). For this reason, the API now
defaults to NOT creating a key, and new methods were added for
situations when the event may need to be cancelled and a key is
necessary.
See the much simplified implementation of the espresso machine example
for a motivating case.
