addConsNode currently unusable?

[CharJon]

Hi everyone,

so I might be missing something, but it seems to me as if the interface is lacking functionality to make use of addConsNode.
To implement some custom branching rules I want to use makeChild and addConsNode. The later expects a constraint as a Constraint object as input (next to a node to add the constraint to). But I can not wrap my head around how to transform my constraint of type ExprCons into a constraint. I tried adding it to the model via addCons, but this lead to problems, as I don't want the constraint to be part of the model in general, but just stick to one child (and it's children).
I might be misunderstanding something as my SCIP knowledge is fairly limited, but the constraint handlers make use of the createCons method, to generate special constraints. Do the branching rules need something similar, or am I doing something wrong?
I appreciate any help and/or comment.

[fserra]

does #391 answer your question?

[CharJon]

Hi, thanks for your answer. #391 is indeed closely related to my problem.
After reading the information there, I still feel like extending the Interface is the only solution for my case, as addCons does not work for me.
What I did for now to solve my problem is also wrapping SCIPcreateConsLinear and using Constraints created this way for addConsNode. This seems to work as expected, I'm just not familiar with wrapping and get a Process finished with exit code 139 (interrupted by signal 11: SIGSEGV) after everything finishes.

If this is the preferred solution and if someone could help me with the SIGSEGV I could create a pull-request to extend the functionality.

[fserra]

I think it would make sense to modify addConsNode to also accept ExprCons, maybe only when the ExprCons is linear.

[CharJon]

This works as well.

Right now I'm basically doing the same as Model._addLinCons does, just without PY_SCIP_CALL(SCIPaddCons(self._scip, scip_cons)) before and PY_SCIP_CALL(SCIPreleaseCons(self._scip, &scip_cons)) after PyCons = Constraint.create(scip_cons).
If I add the constraint creation via PY_SCIP_CALL(SCIPcreateConsLinear(...)) into addConsNode, how do I prevent a SIGSEGV?

[fserra]

I don't understand what you are saying. You can open a PR so we can see the code, but my guess is that it should look something like:

PY_SCIP_CALL(SCIPcreateConsLinear(...))
PY_SCIP_CALL(SCIPaddConsNode(...))
PyCons = Constraint.create(scip_cons)
PY_SCIP_CALL(SCIPreleaseCons(self._scip, &scip_cons))
...
return PyCons

You get a segfault when you do that?
I think it should work without having the increase the ref to cons, because when cons is a Constraint created from a conshdlr within SCIP, then cons is basically the consdata of the constraint and one doesn't want to delete the consdata. However, if the cons is an ExprCons it shouldn't be a problem to release the constraint....

[CharJon]

Thanks again for your feedback. I created #418 .
Adding PY_SCIP_CALL(SCIPreleaseCons(self._scip, &scip_cons)) to the end of the method solved my SIGSEGV error. Could you give some intuition why the constraint has to be "released"?

[fserra]

I wouldn't know why that resolves a segmentation fault. SCIP basically uses reference counting. So when you create a constraint it has reference 1, when you add it to scip it has reference 2. When a constraint reaches reference 0 it gets freed. So if you do not call release, it will never get down to 1 and so SCIP will not release it. That is a memory leak but not a segfault.
We call release at that point in time because we do not want the user to remember the constraint and then call release

[SanielDous]

@CharJon It would be really nice if you could post a code snippet of your workaround. Thereby, others like me could adapt it easier :)

[CharJon]

@CharJon It would be really nice if you could post a code snippet of your workaround. Thereby, others like me could adapt it easier :)

All code I used is in #418. I did not reply to your comment over there, because I do not know what happened to my code. It is flagged as merged, but I can't find the changes from current master branch and #485 is discontinued?

[SanielDous]

Thank you for replying. Unfortunately, I think your code never made it into master since the addConsLocal branch was never merged. This feature would be really useful...

[CGraczyk]

Hi,
I just saw your issue, tbh I dont remember what happened to that branch - could be that I made a mistake and just closed the MR instead of merging it?
I will try to find an answer for you, #605 is related to this as well correct @SanielDous ?

[SanielDous]

Yes, #605 is related. I tried to fix the problem using a constraint handler but it seems to make it way more complicated than what I want to achieve.