taskflow
Exception handling
Yeah... once again: Exception handling.
There are a couple of issues about this, e.g. #72 or #193
My proposal: Make exception handling optional and allow the user to specify the behaviour via (combinable) flags.
Exception handling also shouldn't be limited to a single exception, but rather store all exceptions that were thrown. future.get() should return the first exception that was thrown.
I think these two flags would be appropriate to start with:
TF_ENABLE_EXCEPTION_HANDLING -> Exceptions are caught and stored in order that std::abort does not get called.
TF_CANCEL_ON_EXCEPTION -> Only valid in combination with TF_ENABLE_EXCEPTION_HANDLING, automatically calls cancel once an exception has been caught.
Things that need to be worked out
- Where / how would be the flags be specified?
- How are detached Subflows handled?
- How are nested Subflows handled?
- What exactly gets cancelled? IMHO it would make sense to cancel the whole graph, as it would never be able to complete. However there are also detached Subflows...
What do you think?
