Are the Score Director methods for adding/removing entities truly deprecated?

[storm-TOS]

Certain Score Director methods for managing a variable number of planning entities (before/afterEntityAdded/Removed) have been marked as deprecated as of Timefold version 1.8. However, about half a year ago, @ge0ffrey suggested (in private communication) that this feature might receive continued support after all. What are your current plans regarding this deprecation?

Our task scheduler currently makes crucial and successful use of these methods. We recently tried to incorporate the connected ranges collector, but soon discovered that it bugs out in conjunction with a changing number of entities. Because of this, we expect that support will drop definitively sometime soon, and have been preparing to migrate to a different approach (using a large, fixed number of entities, of which a variable number will not participate in the actual planning). But of course, we would like to know for certain whether this is really necessary.

[triceo]

These methods are indeed and intentionally deprecated, since they aren't consistently supported across the codebase. (You mention some of the issues yourself.)

If you wish to change your problem by adding or removing entities, either do it through ProblemChange, or through a new instance of the solver altogether. (Under the hood, the two are very close.)

[storm-TOS]

It's not as simple as that, because finding a suitable number of entities is part of the solving process. (So we use custom moves that create and delete entities. There seems to be no way around this, including by modeling the problem in a different way etc.)

When we discussed this with Geoffrey a while back, he told us that in such cases, other users tend to use a stack of backup dummy entities during solving, which is probably our best bet as well. He said that he had never heard of users using the entity added/removed methods, but our case made him reconsider the deprecation, which is why I wanted to get back to this.

[triceo]

Yes, I have heard of this use case too. Unfortunately, this doesn't change the fact that these events are currently not reliable and making them reliable would likely require redoing many components in the solver.

Personally, I'd like for this to work. But at the moment, there are greater priorities and marking this functionality clearly as deprecated is, IMO, preferrable to misleading users into using a feature that only partially works.