Thread: User input for array size.

Because you also have to allocate memory for whatever variables a vector has internally.

It doesn't make program that slow.

[edit]
Does it? How many variables it allocates for each push_back()?

It doesn't make program that slow.

I agree with you. I'm just suggesting reasons for the cannon statement.

There are very few.

If the code is too complex, vectors should be used to ease coding, debugging and maintenance.

If the code is too simple, vectors should be used because they are still simpler than an array to implement and we already have a head start in case we need to upgrade the code.

Arrays should be used on the internals of complex classes or hierarchies where the speed factor is critical. The differences in speed are minimal... So only when minimal has a real meaning, should they be used. And needless to say if we need some of the facilities provided by the algorithms, we have to at least be as good programmers as the library implementators to be able to code them with arrays in mind.

I am comparing dynamically allocated memory to vectors. Static arrays are faster of course.

A vector cannot be (if it is, the library implementator is deaf) implemented just to give access to its elements. It must do it in Constant Time ( O(1) ). Any performance hit can only be the result of the constraints derived from the fact a vector is indeed implemented on top of an array.

But because it is still implemented in constant time and because the hit is very small, only with big amounts of data or with the need to traverse its contents repeatedly an incredible number of times can it become noticeable.

And then probably the difference in time on my computer is not asnearly significant on yours, if your processor is just a nudge as faster than mine.

In the end, really... no. They have the same speed. They run both at constant time

EDIT: And you are right Daved. Forgot about the fact many of the algorithms can work with arrays.

Originally Posted by Mario F.

Arrays should be used on the internals of complex classes or hierarchies where the speed factor is critical. The differences in speed are minimal... So only when minimal has a real meaning, should they be used. And needless to say if we need some of the facilities provided by the algorithms, we have to at least be as good programmers as the library implementators to be able to code them with arrays in mind.

The only other thing I use arrays for is buffers and the like which are filled by API function calls. Then I immediately create a vector (or string) and delete[] the buffer.

I'm nervous about letting an API call directly copy memory into a vector. I know the standard guarantees it should work without problems, though, at least for vector and string (but none of the other containers). I always considered it bad form, though.

Or compile in release mode to prevent it from inserting debug symbols in your app.

Originally Posted by Daved

Only for vector, string has no such guarantee.

Huh. Guess I learned something new today.