By decrementing things wherever possible, which makes the code easier to read and, therefore, to update/maintain.

We used to do handle them after having analysed the model which meant having to requalify a variable-based constant equation as an algebraic equation if it depended on one or several external variables. In other words, as a developer, we had to think about what consequence(s) an external variable might have on our analysed model. Until now, it wasn't too much a problem, but with our wish to support systems of non-linear algebraic equations (and then DAEs), this is going to become a nightmare.

... since we may have only generated warnings.

With:
```
mDependencies.erase(std::remove_if(mDependencies.begin(), mDependencies.end(),
                                   [=](const auto &dependency) -> bool {
                                       return dependency.lock()->variable() == nullptr;
                                   }),
```
it reports that the first line is not covered...!?

Even with:
```
mDependencies.erase(std::remove_if(mDependencies.begin(), mDependencies.end(), [=](const auto &dependency) -> bool {
                        return dependency.lock()->variable() == nullptr;
                    }),
                    mDependencies.end());
```
GCov reports that the first line is not covered...!? So, we now put everything on stupidly long line!