Legend:
Library
Module
Module type
Parameter
Class
Class type
The part of the Execution_context that determines what to do when there is an unhandled exception.
Every Async computation runs within the context of some monitor, which, when the computation is running, is referred to as the "current" monitor. Monitors are arranged in a tree -- when a new monitor is created, it is a child of the current monitor.
If a computation raises an unhandled exception, the behavior depends on whether the current monitor is "detached" or "attached". If the monitor has been "detached", via one of the detach* functions, then whoever detached it is responsible for dealing with the exception. If the monitor is still attached, then the exception bubbles to monitor's parent. If an exception bubbles to the initial monitor, i.e., the root of the monitor tree, that prints an unhandled-exception message and calls exit 1.
Note about the toplevel monitor
It is important to note that in the toplevel monitor, exceptions will only be caught in the Async part of a computation. For example, in:
upon (f ()) g
if f raises, the exception will not go to a monitor; it will go to the next caml exception handler on the stack. Any exceptions raised by g will be caught by the scheduler and propagated to the toplevel monitor. Because of this it is advised to always use Scheduler.schedule or Scheduler.within. For example:
Scheduler.within (fun () -> upon (f ()) g)
This code will catch an exception in either f or g, and propagate it to the monitor.
This is only relevant to the toplevel monitor because if you create another monitor and you wish to run code within it you have no choice but to use Scheduler.within. try_with creates its own monitor and uses Scheduler.within, so it does not have this problem.
detach t detaches t so that errors raised to t are not passed to t's parent monitor. If those errors aren't handled in some other way, then they will effectively be ignored. One should usually use detach_and_iter_errors so that errors are not ignored.
val detach_and_iter_errors : t->f:(exn -> unit)-> unit
detach_and_iter_errors t ~f detaches t and passes to f all subsequent errors that reach t, stopping iteration if f raises an exception. An exception raised by f is sent to the monitor in effect when detach_and_iter_errors was called.
get_next_error t returns a deferred that becomes determined the next time t gets an error, if ever. Calling get_next_error t does not detach t, and if no other call has detached t, then errors will still bubble up the monitor tree.
val extract_exn : exn -> exn
extract_exn exn extracts the exn from an error exn that comes from a monitor. If it is not supplied such an error exn, it returns the exn itself. It removes the backtrace from the error (see discussion in try_with).
has_seen_error t returns true iff the monitor has ever seen an error.
val send_exn :
t->?backtrace:[ `Get | `This of Core_kernel.Backtrace.t ]->exn ->
unit
send_exn t exn ?backtrace sends the exception exn as an error to be handled by monitor t. By default, the error will not contain a backtrace. However, the caller can supply one using `This, or use `Get to request that send_exn obtain one using Backtrace.Exn.most_recent ().
try_with f runs f () in a monitor and returns the result as Ok x if f finishes normally, or returns Error e if there is an exception. It either runs f now, if run = `Now, or schedules a job to run f, if run = `Schedule. Once a result is returned, subsequent exceptions raised to the monitor are handled according to rest:
`Log: Logged to a global error log (cannot raise).
`Raise: Reraised to the monitor of try_with's caller.
`Call f: Passed to f within the context of the caller of try_with's monitor.
The name argument is used to give a name to the monitor the computation will be running in. This name will appear when printing errors.
try_with runs f () in a new monitor t that has no parent. This works because try_with calls detach_and_get_error_stream t and explicitly handles all errors sent to t. No errors would ever implicitly propagate to t's parent, although try_with will explicitly send them to t's parent with rest = `Raise.
If extract_exn = true, then in an Error exn result, the exn will be the actual exception raised by the computation. If extract_exn = false, then the exn will include additional information, like the monitor and backtrace.
handle_errors ?name f handler runs f () inside a new monitor with the optionally supplied name, and calls handler error on every error raised to that monitor. Any error raised by handler goes to the monitor in effect when handle_errors was called.
Errors that are raised after f () becomes determined will still be sent to handler, i.e., the new monitor lives as long as jobs created by f live.
protect f ~finally runs f () and then finally regardless of the success or failure of f. It re-raises any exception thrown by f or returns whatever f returned.
The name argument is used to give a name to the monitor the computation will be running in. This name will appear when printing the errors.