I still think an explicit TNF task would be useful. But I don't think it makes sense to work on this before the change to the component interaction problem is done.

I currently have no plans to pursue this. Florian, how do you feel about this? Should we close this issue (we can always reopen it if our plans change)?

Introducing the ExplicitTask class is issue700.

As we discussed offline, I suggest we split this issue into two or three parts: 
1) introducing the class ExplicitTask
2) introducing the class TransitionNormalFormTask
3) using TNF for potential heuristics

This should make it easier to discuss and make the dependencies more explicit.

Now that we've decided to remove GlobalOperator, we can move ahead on this issue. 
I have merged it with the latest master revision and updated it to reflect our 
latest guidelines. Florian, could you have a look at the code, please?

I'll have another look after the next Fast Downward meeting, where we wanted to 
discuss the classes ExplicitOperator and GlobalOperator.

Thanks for the comments. I think I took care of them all. Do you want to have 
another look, Florian?

I'm done with my changes. You can have a look at the code now.

We currently check that parent tasks of DomainAbstractedTask don't have axioms 
and conditional effects by calling the local "static bool 
has_conditional_effects(const AbstractTask &task)" function, i.e., on the level 
of AbstractTask. In this pull request, we check the same thing for 
TransitionNormalFormTask by using 

TaskProxy parent_task_proxy(*parent);
verify_no_axioms(parent_task_proxy);
verify_no_conditional_effects(parent_task_proxy);

i.e., on the level of TaskProxy. For TNFTask this is possible since the TNFTask 
is hidden in the .cc file and it is the TNFTask factory that makes the check.

We should probably discuss offline if we can use a similar approach for 
DomainAbstractedTask.

I'll try to implement the necessary methods now.

There still is a TODO about merging issue660. Should I wait for that?

Done with the next round of comments. Florian, could you have a look, please?

I made the changes and the pull request is ready for review again.

I added some comments. Can you ping me again, once you made the change to
explicit task we discussed?

I have made a pull request at 
https://bitbucket.org/jendrikseipp/downward/pull-requests/52

Florian, can you have a look when time allows?

We'd like to use TNF for computing the admissibility constraints for potential 
heuristics.