Thread: Container allocators based in malloc and realloc

Originally Posted by ccafe

I've found several container classes out there that use malloc() internally as their allocator, and realloc() for growing/shrinking. Of course, they use placement new.

They claim the only requirement is that these container classes cannot be used for objects "whose copy constructor has side effects". Is that always the case? I mean, is having a "copy constructor without side effects" the only requirement for being it safe to reallocate C++ objects to a different memory address?

I make this question because one of the first things I learnt about C++ is that objects are not supposed to move from one memory address to another, so I'm wondering if this kind of container classes violate some rule from the standard.

It would be good for you to provide an example of such a container class and its associated documentation for readers here to examine since it is possible that you may have misinterpreted the claim, or someone attempting to answer your questions may misinterpret your paraphrasing of the claim.

In general though, C++ programmers tend to be conditioned not to rely on copy constructors having side effects since copy constructors can be elided.

Originally Posted by ccafe

Said this, is there any succinct and complete definition of what's a "copy constructor without side effects"? Typically I hear the explanation that they don't do any other thing but copying the members, but I find that explanation quite vague and imprecise.

I think it is used in the general computer science sense of "modifying outside the local environment context" rather than something explicitly defined by the standard, that's why the explanations you hear are "quite vague and imprecise".

Originally Posted by ccafe

By the way, is there any keyword in the C++ language that lets you tag a copy constructor as either having side effects or not?

Not that I know of, unless you count comments of the structured type that ends up in generated documentation.