11

We use Integer type represent index variables most of the time. But in some situations, we're forced to choose 

std::vector<int> vec; .... for(int i = 0; i < vec.size(); ++i) ....

This will cause the compiler to raise the warning that mixed use of signed/unsigned variables. if I make the index variable as for( size_t i = 0; i < vec.size(); i++ ), (or an unsigned int )it will sort out the issues.

When it come more specific to use windows types, most of the Windows APIs are dealing with DWORD (which typedef-ed as unsigned long).

So when I use similar iteration, will again cause the same warning. Now if I rewrite it as 

DWORD dwCount; .... for(DWORD i = 0; i < dwCount; ++i) ....

I find this a bit weird. It might be the problem withe perceptions.

I agree that we are supposed to use the same type of index variable to avoid the range problems can happen with the index variables. For e.g if we're using 

_int64 i64Count; // .... for(_int64 i = 0; i < i64Count; ++i) ....

But in the case of DWORD, or unsigned integers, are there any problems in rewriting it as 

for(int i = 0; (size_t)i < vec.size(); ++i)

How most of the people are working with similar issues?

Improve this question
3
  • 4
    Why would you use a signed integer to represent index? That's like using vector of integers to store a string. Commented Aug 30, 2011 at 9:46
  • 3
    @Let_Me_Be: Because it makes the boundary conditions easier to test. For example on a count down to zero, a less-than test before executing the loop body cannot work with an unsigned value. Similarly a count up to the maximum doesn't work. Of course in that case a signed integer won't work either (because it can't represent such a large value). Commented Jan 13, 2012 at 20:03
  • "This will cause the compiler to raise the warning that mixed use of signed/unsigned variables." That's only one of the two problems you'll have to contend with. In many cases std::size_t is a higher rank than int (or even long). If the vector's size ever exceeds std::numeric_limits<int>::max(), you'll regret having used int. Commented Jan 24, 2018 at 0:31

6 Answers 6

Reset to default
11

vector has a typedef that tells you the correct type to use :-

for(std::vector<int>::size_type i = 0; i < thing.size(); ++i) { }

It's almost always defined to be size_t though but you can't rely on that

Improve this answer
6
  • 9
    Not really an improvement in readability, IMHO. Commented Aug 30, 2011 at 13:12
  • No, but as it's the only way to tell what is the correct type to use for an index into the vector, that doesn't really matter... Commented Aug 30, 2011 at 13:26
  • 4
    These days in c++11 just use auto Commented Jan 1, 2014 at 9:37
  • 8
    @JohnB You mean like auto i = 0? That doesn't help at all, i become an int. Commented Mar 27, 2015 at 3:42
  • 1
    Readability can be improved, with using index_t = std::vector<int>::size_type;. Commented Oct 19, 2017 at 9:00
4 
std::vector<int> vec; for(int i = 0; i < vec.size(); ++i)

Use an iterator for this, not a for loop.

For the others, as long as the variable type is of the same size, static_cast should work just fine (i.e. DWORD to int16_t)

Improve this answer
1
  • 3
    for (std::vector<int>::iterator i = vec.begin(); i != vec.end(); ++i) is a pain to write. Having for (auto i = vec.begin();... is a whole lot more readable. Of course, foreach is also in C++11. Commented Aug 30, 2011 at 14:12
4

The case you described is one of the things I dislike in C++, too. But I have learned to live with that, either by using

for( size_t i = 0; i < vec.size(); i++ )

or

for( int i = 0; i < (int)vec.size(); i++ )

(of course, the latter only when there is no risk of getting some int overflow).

Improve this answer
4

The reason it's warning you about comparison between signed and unsigned is because the signed value will likely be converted to unsigned, which might not be what you expect.

In your example (comparing int to size_t), int will be implicitly converted to size_t (unless int somehow has a larger range than size_t). Thus, if the int is negative, it will likely be greater than the value you're comparing it to due to wraparound. This won't be a problem if your index is never negative, but you'll still get that warning.

Instead, use an unsigned type (such as unsigned int, size_t, or, as John B recommends, std::vector<int>::size_type) for your index variable:

for(unsigned int i = 0; i < vec.size(); i++)

Be careful when counting down, though:

for(unsigned int i = vec.size()-1; i >= 0; i--) // don't do this!

The above will not work because i >= 0 is always true when i is unsigned. Instead, use the "arrow operator" for loops that count down:

for (unsigned int i = vec.size(); i-- > 0; ) vec[i] = ...;

As other answers point out, you normally want to use an iterator to traverse a vector. Here's the C++11 syntax:

for (auto i = vec.begin(); i != vec.end(); ++i)
Improve this answer
1
  • 1
    This still runs the risk that an unsigned int isn't large enough to hold the size. Commented Jan 24, 2018 at 0:39
2

A new option for C++11, you can do things like the following

for(decltype(vec.size()) i = 0; i < vec.size(); ++i) {...}

and

for(decltype(dWord) i = 0; i < dWord; ++i) {...}

While it repeats a little more than the basic for-loop would, it's not nearly as long-winded as pre-'11 ways of specifying values, and using this pattern consistently will work for most, if not all, possible terms you'd want to compare against, which makes it great for code refactoring. It even works for simple cases like this:

int x = 3; int final = 32; for(decltype(final) i = x; i < final; ++i)

Furthermore, while you should use auto whenever you're setting i to some intelligent value (like vec.begin()), decltype works when you're setting to a constant like zero, where auto would just resolve that to int because 0 is a simple integer literal.

To be honest, I'd like to see a compiler mechanism for extending the auto type-determination for loop incrementers to look at the value being compared against.

Improve this answer
1

I use a cast to int, as in for (int i = 0; i < (int)v.size(); ++i). Yes, it's ugly. I blame it on the stupid design of the standard library where they decided to use unsigned integers to represent sizes. (In order to.. what? extend the range by one bit?)

Improve this answer
5
  • 1
    In what situation would a collection size of negative anything be meaningful? Using unsigned integers for various collection sizes seems like the sensible choice, to me. Last I checked, taking the length of a string rarely returned a negative result either... Commented Aug 31, 2011 at 8:11
  • 2
    In what situation would it be meaningful for simple test such as if(v.size()-1 > 0) { ... } to return true for an empty container? The problem is that sizes are also often used in arithmetic, esp. with index-based containers, which is asking for trouble given that they're unsigned. Basically, using unsigned types for anything else than 1) bitwise manipulations, or 2) modular arithmetic is calling for trouble. Commented Sep 1, 2011 at 6:11
  • 2
    good point. While I don't really see the point of your particular example (I would probably just write if(v.size() > 1) { ... } since that makes the intent more clear, and as an added bonus the issue of signed/unsigned becomes void), I do see how in some specific cases signedness might be useful. I stand corrected. Commented Sep 2, 2011 at 20:59
  • 1
    @Michael: I agree it was contrived example. Still, I often write algorithms with nested loops: for(i=0;i<v.size()-1;++i) for(j=i+1;j<v.size();++j) .. if v is empty, the outer loop executes (size_t)-1 times. So I either have to check v.empty() before the loop or cast v.size() to a signed type, both of which I personally think are ugly workarounds. I choose a cast as it's fewer LOC, no if()s => fewer possibilites for mistake. (Also, in 2nd complement, conversion oveflow returns a negative number, so the loop doesn't execute at all.) Commented Sep 3, 2011 at 9:44
  • 1
    Extending the range by 1 bit was (and continues to be) very useful in 16-bit systems. Commented Jan 24, 2018 at 0:41

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.