| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
"What's wrong with an API which non-deterministically returns either
type A or type B(Set(A, ...))?"
This is pretty much what the exception handling behavior of the
parallel collections does: If exceptions of type A occur, either an
exception of type A or an exception of type B, wrapping multiple
exceptions of A's, is returned.
This behavior is incredibly broken and so unintuitive, that even
people writing tests don't handle it correctly, as seen in test
files/run/t5375.scala.
Concerning “non-deterministic”:
How many exceptions are observed before the operation is aborted
depends on the machine, the available cores and hyper-threading,
the operating system, the threadpool implementation and
configuration, the size of the collection and the runtime itself.
In fact, files/run/t5375.scala can be made to fail reproducible
on both jdk7u and Avian, if we run on a single-core machine
like in a virtual machine.
With this change, we just pass the "first" exception which occurs.
This is
- consistent with the behaviour of sequential collections,
- doesn't require users to know more about parallel collections
than they already do ("elements might be processed out of order"),
- in line with what Java 8 does.
“Why don't we wrap everything in CompositeThrowables?”
Even consistently returning CompositeThrowable doesn't make much
sense (because we have fail-fast behaviour and don't wait until
all tasks have finished or have thrown an exception).
Therefore, there is no useful semantic in having a
CompositeThrowable which returns "some" exceptions.
I have done extensive research into C#'s approach (which is very
similar to what Scala did until now, but even more messy) and
the key observation from asking multiple C# developers is that
not a single one had understood how PLINQ handled exceptions or
could describe the semantics of it.
As a consequence, either
a) gather and return all exceptions in a CompositeThrowable or
b) just return one, unwrapped,
instead of non-deterministically wrapping a non-deterministic
number of exceptions into a completely unrelated wrapper type.
Considering that changing the parallel collection semantics in
such a profound way as described in a) is out of question, b)
is chosen.
As soon as Scala targets Java > 7 Throwable#addSurpressed can be
used to add further exceptions to the one which gets returned.
This would allow passing more information to the caller without
compromising the simplicity of the API.
|
