Discourage use of References

[mkristien]

void foo(int& val); or
void foo(int* val); ?

I do not like references because they look like passing values around but they behave like pointers. This can lead to programmers' errors.

[neilweidinger]

Can you specify what kind of errors? I'm of the opinion that we should actually be encouraging usage of references, especially const myDataType& variable_name, since this avoids accidentally changing the state of the value passed in and is closer to idiomatic C++11 than raw pointers.

[mkristien]

The error happens if people assume value semantics while the code has pointer semantics.
Last year, original embrakes code had this:
Data stateMachineData = Data::getInstance();
This created a copy of the shared data structure, thus updates to it were done to the copy and not to the one and only. This could have been prevented if Data::getInstance() returned Data* instead of Data&.

[brano2]

@mkristien in that example, the real problem is that the singleton pattern for the Data class was not implemented properly (I suppose the constructor that converts a reference to a value was not deleted). It does not work any more afaik even though Data::getInstance() still returns a reference.

[brano2]

I don't think references are confusing as they are the closest thing C++ has to languages like Java or Python which only have one way of passing things around. Hence most programmers should be more skilled at working with references than working with pointers/values. If there is evidence that people were actually messing up with references much more often than with values and pointers, I'll change my opinion.