Could you make the monotonic constraints work with gblinear as well?

Currently the monotonic constraints only work when using gbtree or dart as the booster. Could we let it work with gblinear as well? By doing so we will able to control the sign of weights for variables, which is crucial to make the final model match the domain knowledge.

An intuitive idea will be to check the sign of weights after each iteration, zero out the ones conflicting with the monotonic constraints and then start the next iteration.

Thank you.