Assigning c-array to c++ vector

Question

I have a question about lifetime of onstack created arrays and about transforming them to c++ vectors. For ex I have two functions:

void getVector()
{ 
    auto myVector = createVectorFromArray();
}

vector<float> createVectorFromArray()
{
    float myArr[10];
    readDataFromSomewhere(myArr, 10); //some read data function with signature void(float*, size_t)
    vector<float> resVector;
    resVector.assign(myArr, myArr+10);
    return resVector;
}

As I understand array myArr will be killed as soon as we leave function createVectorFromArray. So iterators of vector myVector in function getVector will point to nowhere. Am I right or this works in another way? And how should I better make vector from array to return it from function in this situation?

Thank you!

You can't invalidate elements of a std::vector that simply, unless they are pointers. — LogicStuff
– LogicStuff, Commented Aug 31, 2016 at 8:24
Have you had problems doing this? It seems really easy to just compile it and test it out. — sokkyoku
– sokkyoku, Commented Aug 31, 2016 at 8:30
@sokkyoku, yep, you're right. I should check it before asking such a simple q( — Daniil Vlasenko
– Daniil Vlasenko, Commented Aug 31, 2016 at 8:36

Garf365 · Accepted Answer · 2016-08-31 08:56:01Z

3

You can rewrite createVectorFromArray like that and you don't have anymore trouble with assigning an array to vector:

vector<float> createVectorFromArray()
{
    vector<float> resVector(10);

    readDataFromSomewhere(&resVector[0], resVector.size());

    return resVector;
}

std::vector is contiguous memory container, so you can use it as interface with plain old C functions

edited Aug 31, 2016 at 8:56

answered Aug 31, 2016 at 8:30

Garf365

3,7075 gold badges33 silver badges42 bronze badges

Sign up to request clarification or add additional context in comments.

33 Comments

Richard Hodges Over a year ago

@Nidhoegger in the current implementation, the floats will be uninitialised. This answer is correct and idiomatic. You're wrong to downvote it.

Richard Hodges Over a year ago

@Nidhoegger you're wrong to say that. It's absolutely optimal.

Richard Hodges Over a year ago

@sameerkn &vector[0] does not break encapsulation. A vector's elements are guaranteed by the standard to be contiguous. This use case is idiomatic.

Richard Hodges Over a year ago

@Garf365 the vector is returned efficiently by RVO. There is no overhead.

Richard Hodges Over a year ago

@Nidhoegger RVO is ubiquitous. As of c++17 it is mandated by the standard. You can rely on it.

|

Community · Accepted Answer · 2020-06-20 09:12:55Z

This answer is designed to discuss the various approaches, explain them, and put them into context.

option 1: the native to vector copy:

vector<float> createVectorFromArray()
{
    // data copied once
    float myArr[10];
    readDataFromSomewhere(myArr, 10);

    // vectors are lightweight handles. zero-cost construction
    // since the optimiser will see that the next statement assigns to
    // the vector
    vector<float> resVector;

    //
    // data copied twice
    //
    resVector.assign(myArr, myArr+10);

    //
    // it's almost guaranteed that RVO will elide this copy. As of c++17
    // it's a guarantee. Returning the vector is fine
    return resVector;
}

Problems:

data copied twice
vector will require memory allocation

option 2: use the vector directly

vector<float> createVectorFromArray()
{
    // Will allocate space and
    // default-initialise the floats (i.e. they won't be initialised)
    vector<float> resVector(10);

    //
    // one data copy. perfectly safe and idiomatic.
    // std::addressof expressed intent clearly.
    //

    readDataFromSomewhere(std::addressof(resVector[0]), resVector.size());

    //
    // it's almost guaranteed that RVO will elide this copy. As of c++17
    // it's a guarantee. Returning the vector is efficient.
    return resVector;
}

Problems:

vector will require memory allocation

Better...

option 3 : re-use an existing vector

void createVectorFromArray(vector<float>& resVector)
{
    //
    // if resVector has been used before and has a capacity() of 10
    // or more, no memory allocation necessary
    //
    resVector.resize(10);

    // one copy of data
    readDataFromSomewhere(std::addressof(resVector[0]), resVector.size());
}

Problems:

perhaps not quite to easy to use the reference interface.

How would I choose between option 2 and 3?

Option 2 is more readable (IMHO) but would be expensive if used in a loop. So for a one-off, I'd go for that.

If I'm reading data into a buffer in a loop, I'd probably want to avoid un-necessary memory allocations. So re-using the vector's capacity would be a wise move.

e.g.

std::vector<float> buf;
while (thereIsMoreData())
{
    createVectorFromArray(buf);   // option 3
    useTheData(buf);
    // data no longer needed, so we can re-use the vector
}

the alternative:

while (thereIsMoreData())
{
    // the passing of the vector is efficient, particularly as of c++11
    // however, we still suffer one memory allocation per loop.
    // probably undesirable in a high performance environment.
    useTheData(createVectorFromArray());   // option 2
}

Finally...

option 4:

Provide both. Allow the user the performant approach or the 'readable' approach as he/she wishes

void createVectorFromArray(vector<float>& resVector)
{
    //
    // if resVector has been used before and has a capacity() of 10
    // or more, no memory allocation necessary
    //
    resVector.resize(10);

    // one copy of data
    readDataFromSomewhere(std::addressof(resVector[0]), resVector.size());
}

// overload
std::vector<float> createVectorFromArray()
{
    std::vector<float> result;
    createVectorFromArray(result);
    return result;
}

Can you add some use case, specially for 2 last options ? Where it better to use one rather than other ?
@Garf365 sure. basically in loops. You don't want to have to reallocate buffers all the time

Collectives™ on Stack Overflow

Assigning c-array to c++ vector

2 Answers 2

33 Comments

option 1: the native to vector copy:

option 2: use the vector directly

option 3 : re-use an existing vector

option 4:

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

33 Comments

option 1: the native to vector copy:

option 2: use the vector directly

option 3 : re-use an existing vector

option 4:

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related