Initializations in C++

Initialization has been a vague part of my C++ knowledge. Although I always knew that most user defined classes has appropriate constructors, so that I can expect reasonable behavior (e.g. std::string initializes to an empty string "" ), this hasn't been the case for built-in arrays of vectors.

What do you expect from this code:

float * test = new float[10];
copy(test, test+10, ostream_iterator<float>( cout, " ");

Honestly, I expect anything to happen though it sometimes behaves reasonably and spits out zeros. As a matter of fact, valgrind will list hundreds of errors for this tiny peace of code, most of which are of the form "Conditional jump or move depends on uninitialised value". The way to make sure everything is zero initialized is to use std::fill or std::fill_n (but please, no memset) after the allocation. On the other hand, this is not always possible.

Assume you have a templated function which takes an array of T's where T can be pretty much anything. If it is a non-POD (plain old data) class, then you can look at the documentation and make sure the default constructor does what you want it to do (in my case, assignment to zero). But how to handle built-in types without writing a specialization to cover them? std:fill trick won't work since it might not make sense to the non-POD class you're dealing with.

Here is the loophole :


float * test = new float[10] () ;
copy(test, test+10, ostream_iterator<float>( cout, " ");

Note the empty set of parantheses as the initializer, which makes the array elements default constructed and the C++ standard says that a default constructed POD type is zero-initialized. Neat, huh? (btw, valgrind reports 0 errors this time)