returning references

[pauldepstein]

Below is posted from a link for Stanford students in computer
science.

QUOTE BEGINS HERE
Because of the risk of misuse, some experts recommend never returning
a
reference from a function or method.
QUOTE ENDS HERE

I have never heard anyone else say that it is a problem for a function
to return a reference. Are there really experts who object, or is
this nothing other than the commonplace observation that reference-
returning is a somewhat difficult concept that needs to be learned
carefully? (I am just learning about this now.) Assuming programmers
have some degree of competence and are able to avoid returning
references to locals and so on, what (if anything) are the pitfalls?

Paul Epstein

 

[Christopher]

Below is posted from a link for Stanford students in computer
science.

QUOTE BEGINS HERE
Because of the risk of misuse, some experts recommend never returning
a
reference from a function or method.
QUOTE ENDS HERE

I have never heard anyone else say that it is a problem for a function
to return a reference. Are there really experts who object, or is
this nothing other than the commonplace observation that reference-
returning is a somewhat difficult concept that needs to be learned
carefully? (I am just learning about this now.) Assuming programmers
have some degree of competence and are able to avoid returning
references to locals and so on, what (if anything) are the pitfalls?

Paul Epstein

I don't see anything at all with returning references as long as the
const keyword is used correctly when appropriate. Surprisingly, a good
number of people do not use const when creating thier function stubs;

const MyClass & GetMyClass() const; // The typical accessor
SetMyClass(const MyClass & myclass); // The typical mutator

I constantly find the above two methods lacking the const keyword in
source when I am debugging and it leads to misuse by other programmers
who simply glance at a method without taking into account the impact
of making changes to the returned reference.

Another common problem I run into is you can't return NULL for a
reference. Sometimes it is nice to have the option to return NULL when
an error occured in which case I use a pointer and tell the caller to
check the returned value to determine if an error occured. Some people
don't like that idea though, but it is worth mentioning.

 

[Erik Wikström]

Below is posted from a link for Stanford students in computer
science.

QUOTE BEGINS HERE
Because of the risk of misuse, some experts recommend never returning
a
reference from a function or method.
QUOTE ENDS HERE

I have never heard anyone else say that it is a problem for a function
to return a reference. Are there really experts who object, or is
this nothing other than the commonplace observation that reference-
returning is a somewhat difficult concept that needs to be learned
carefully? (I am just learning about this now.) Assuming programmers
have some degree of competence and are able to avoid returning
references to locals and so on, what (if anything) are the pitfalls?

As far as I know there is only one pitfall, and that is returning a
reference to a variable local to the function, like so:

int& foo(int i)
{
int r = i+i;
return r;
}

int main()
{
int& twice = foo(5);
}

Since r is a local variable to foo() it will go out of scope as soon as
foo() is done executing, this means that the reference returned refers
to a variable that no longer exists.

On the other hand there are a number of valid reasons to return a
reference from a function, a function that returns an element from a
collection is a good example (look at the at() function in std::vector).

 

[kwikius]

Below is posted from a link for Stanford students in computer
science.

QUOTE BEGINS HERE
Because of the risk of misuse, some experts recommend never returning
a
reference from a function or method.
QUOTE ENDS HERE

hmmm... I assume that following this extremely wise and useful advice,
Stanford students will not be allowed to do ...

#include <iostream>
int main(){
std::cout << "hello world\n";
}


regards
Andy Little

(Thankfully Not a C++ expert!)

 

[Daniel T.]

Below is posted from a link for Stanford students in computer
science.

QUOTE BEGINS HERE
Because of the risk of misuse, some experts recommend never
returning a reference from a function or method.
QUOTE ENDS HERE

I have never heard anyone else say that it is a problem for a
function to return a reference.

"C++ Coding Standards" by Sutter & Alexandrescu. Item 42: Don't give
away your internals.

Don't volunteer too much: Avoid returning handles to internal data
managed by your class, so clients won't uncontrollably modify
state that your object thinks it owns.

For context, "handles" above is defined as non-const references, and
pointers to non-const data.

So now you have heard of at least two acknowledged experts who almost
state the same thing.

Are there really experts who object, or is this nothing other than
the commonplace observation that reference- returning is a somewhat
difficult concept that needs to be learned carefully? (I am just
learning about this now.) Assuming programmers have some degree of
competence and are able to avoid returning references to locals and
so on, what (if anything) are the pitfalls?

The only valid reference that can be returned from a non-member function
is something that either the calling code had access to anyway, or
something that is buried in a module.

The only valid reference that can be returned from a member-function is
of something that either the calling code had access to anyway, or
something that is private within the class.

In either case, if you are returning a non-const reference, then the
object returned better not have anything to do with the invariant of
that class/module or the class/module is asking for trouble
(encapsulation is broken.)

 

[Christopher]

"C++ Coding Standards" by Sutter & Alexandrescu. Item 42: Don't give
away your internals.

Don't volunteer too much: Avoid returning handles to internal data
managed by your class, so clients won't uncontrollably modify
state that your object thinks it owns.

For context, "handles" above is defined as non-const references, and
pointers to non-const data.

So now you have heard of at least two acknowledged experts who almost
state the same thing.

I beg to differ with the statement that two quotes from so called
experts _almost_ say the same thing. They are extremely different
considering the impact of const correctness. the former quote is
moronic and the latter is imo quite correct.

The only valid reference that can be returned from a non-member function
is something that either the calling code had access to anyway, or
something that is buried in a module.

The only valid reference that can be returned from a member-function is
of something that either the calling code had access to anyway, or
something that is private within the class.

What's stopping me from returning a reference to protected data
exactly?

In either case, if you are returning a non-const reference, then the
object returned better not have anything to do with the invariant of
that class/module or the class/module is asking for trouble
(encapsulation is broken.)

agreed.

 

[Salt_Peter]

Below is posted from a link for Stanford students in computer
science.

QUOTE BEGINS HERE
Because of the risk of misuse, some experts recommend never returning
a
reference from a function or method.
QUOTE ENDS HERE

I have never heard anyone else say that it is a problem for a function
to return a reference. Are there really experts who object, or is
this nothing other than the commonplace observation that reference-
returning is a somewhat difficult concept that needs to be learned
carefully? (I am just learning about this now.) Assuming programmers
have some degree of competence and are able to avoid returning
references to locals and so on, what (if anything) are the pitfalls?

Paul Epstein

Nothing wrong in returning references, example: various operators.
You already coverred the case of returning a temporary.
One pitfall is side-effects where an accessor might _get_ nasty.
Thats where CV-qualifiers play an important role.

#include <iostream>
#include <ostream>

class N
{
int m_n;
public:
N(int n) : m_n(n) { }
// accessors
int& getnasty() { return m_n; } // nasty
int const& getsafe() const { return m_n; }
void set(const int& n) { m_n = n; }
// friend op<<
friend std:stream&
operator<<(std:stream& os, const N& r)
{
return os << r.m_n;
}
};

int main()
{
N instance(9);
std::cout << instance << std::endl;

instance.getnasty() = -1; // nasty side-effect
std::cout << instance << std::endl;

instance.set(8);
std::cout << instance << std::endl;
std::cout << instance.getsafe() << std::endl;
}

/*
9
-1
8
8
*/

 

[Barry]

hmmm... I assume that following this extremely wise and useful advice,
Stanford students will not be allowed to do ...

#include <iostream>
int main(){
std::cout << "hello world\n";

well this is allowed, while
std::cout << "hello world" << std::endl;
is not,

write
std::cout << "hello world";
std::cout << std::endl;
instead.

 

[James Kanze]

I beg to differ with the statement that two quotes from so called
experts _almost_ say the same thing. They are extremely different
considering the impact of const correctness. the former quote is
moronic and the latter is imo quite correct.

The real difference is that they are addressing two radically
different levels. The Sutter/Alexandrescu quote has nothing to
do with references per se, but is more concerned with design;
I'm sure that neither Sutter nor Alexandrescu would condemn
std::vector<>:perator[] because it returns a reference (to
internal data, no less), nor the operator<< and operator>> in
iostream because they return a reference to support chaining.
(Alexandrescu definitly doesn't like iostream, but I don't think
that the fact that operator<< returns a reference has anything
to do with it.)

The real difference is that Sutter and Alexandrescu address a
real issue with their comment: how to intelligently design a
class. Where as the quoted recommendation simply misses the
point entirely, by concentrating on an irrelevant syntax issue.
(How is returning a reference riskier than returning a pointer,
for example?)

What's stopping me from returning a reference to protected
data exactly?

agreed.

But what does that say, anyway? And how is returning a
reference any more of a problem here than returning a pointer.

There are two issues with returning a reference; permitting
access to internal details, and the lifetime of what is referred
to. They're both design issues, both affect other constructs as
well, and neither affects all use of references. Many standard
idioms and patterns involve returning references, and I've never
heard of any expert who recommended never returning a reference.

 

[Erik Wikström]

well this is allowed, while
std::cout << "hello world" << std::endl;
is not,

write
std::cout << "hello world";
std::cout << std::endl;
instead.

The reference will be returned regardless if you use it or not, but I
guess they make an exception for library functions

PS. Please do not quote signatures

 

[Daniel T.]

I beg to differ with the statement that two quotes from so called
experts _almost_ say the same thing. They are extremely different
considering the impact of const correctness. the former quote is
moronic and the latter is imo quite correct.

Only if you assume that the former quote includes "const reference" when
it says "reference". It may not.

My personal standing rule is, "functions return const reference, pointer
to const, or by object, unless they are returning one of the parameters
passed in." (Of course there are exceptions, there are exceptions to
every rule, but if I find myself returning a non-const reference or a
pointer to a non-const object, I make sure I have a real good reason.)

What's stopping me from returning a reference to protected data
exactly?

Granted. The point of the post is that if you are returning a non-const
reference to something the calling code doesn't have access to (for
whatever reason,) you are probably making a mistake.

I suspect that at this point, any difference between our positions is
merely picking nits.

 

[Bo Persson]

Daniel T. wrote:
:: (e-mail address removed) wrote:
::
::: Below is posted from a link for Stanford students in computer
::: science.
:::
::: QUOTE BEGINS HERE
::: Because of the risk of misuse, some experts recommend never
::: returning a reference from a function or method.
::: QUOTE ENDS HERE
:::
::: I have never heard anyone else say that it is a problem for a
::: function to return a reference.
::
:: "C++ Coding Standards" by Sutter & Alexandrescu. Item 42: Don't
:: give away your internals.
::
:: Don't volunteer too much: Avoid returning handles to internal
:: data managed by your class, so clients won't uncontrollably
:: modify state that your object thinks it owns.
::
:: For context, "handles" above is defined as non-const references,
:: and pointers to non-const data.
::
:: So now you have heard of at least two acknowledged experts who
:: almost state the same thing.
::
::: Are there really experts who object, or is this nothing other than
::: the commonplace observation that reference- returning is a
::: somewhat difficult concept that needs to be learned carefully?
::: (I am just learning about this now.) Assuming programmers have
::: some degree of competence and are able to avoid returning
::: references to locals and so on, what (if anything) are the
::: pitfalls?
::
:: The only valid reference that can be returned from a non-member
:: function is something that either the calling code had access to
:: anyway, or something that is buried in a module.
::
:: The only valid reference that can be returned from a
:: member-function is of something that either the calling code had
:: access to anyway, or something that is private within the class.
::
:: In either case, if you are returning a non-const reference, then
:: the object returned better not have anything to do with the
:: invariant of that class/module or the class/module is asking for
:: trouble (encapsulation is broken.)

But even if you return a const reference to internal data, the
encapsulation is still broken - just a bit less. By returning a
reference, you have promised to somehow keep the referred-to object.
If you return by value, you are more free to change your
implementation later.


Also, we should all note that Sutter & Alexandrescu say "avoid" in
their guidelines, not "never do this". That's an important difference!


Bo Persson

 

[Daniel T.]

But even if you return a const reference to internal data, the
encapsulation is still broken - just a bit less. By returning a
reference, you have promised to somehow keep the referred-to object.
If you return by value, you are more free to change your
implementation later.

Changing a function from returning const-reference to returning const
object can't break any existing clients.

IE. if I have this:

class Object {
public:
const Foo& getFoo() const;
};

and later, I don't want to store a Foo, I can easily change the class to:

class Object {
public:
const Foo getFoo() const; // I could even dump the first const
};

and clients will not be affected. As I understand it, the two returns
are as interchangeable as if they were function parameters.

At least this is my understanding...

 

[James Kanze]

Changing a function from returning const-reference to returning const
object can't break any existing clients.

IE. if I have this:

class Object {
public:
const Foo& getFoo() const;
};

and later, I don't want to store a Foo, I can easily change
the class to:

class Object {
public:
const Foo getFoo() const; // I could even dump the first const
};

and clients will not be affected. As I understand it, the two
returns are as interchangeable as if they were function
parameters.

Unless, of course, the type doesn't support copy. Or if the
client takes the address of the return, and later uses it. Or
any number of other cases as well. The lifetime of the object
is *not* the same.

 

[Daniel T.]

Unless, of course, the type doesn't support copy. Or if the
client takes the address of the return, and later uses it. Or
any number of other cases as well. The lifetime of the object
is *not* the same.

Well, in the former case, Object has no choice but to return a reference
(or pointer.) In the latter case... I don't concern myself with such
pathological behavior by the client. I don't guarantee that state
returned by const reference will last past the next sequence point in
any case.

 

[James Kanze]

[...]

Well, in the former case, Object has no choice but to return a
reference (or pointer.)

And the case is rather frequent, don't you think.

In the latter case... I don't concern myself with such
pathological behavior by the client. I don't guarantee that
state returned by const reference will last past the next
sequence point in any case.

In sum, you require the results of a lookup in a map to be used
in the expression doing the lookup, and not later. That's very,
very counter-intuitive and unexpected. And also very difficult
to use correctly, since the client does have to test whether the
lookup found something before using it.

 

[Daniel T.]

Daniel T wrote:
[Returning a const-reference to internal state can be
converted to a return of a const object of that state]

Unless, of course, the type doesn't support copy. Or if the
client takes the address of the return, and later uses it. Or
any number of other cases as well. The lifetime of the object
is *not* the same.

Well, in the former case, Object has no choice but to return a
reference (or pointer.)

And the case is rather frequent, don't you think.

That depends on how often you use value semantics.

[What] if the client takes the address of [a const-reference]
return, and later uses it.

In the latter case... I don't concern myself with such
pathological behavior by the client. I don't guarantee that state
returned by const reference will last past the next sequence
point in any case.

In sum, you require the results of a lookup in a map to be used in
the expression doing the lookup, and not later. That's very, very
counter-intuitive and unexpected. And also very difficult to use
correctly, since the client does have to test whether the lookup
found something before using it.

A map does not return a non-const reference to any internal state, so I
don't see how that relates in any way.

 

[James Kanze]

Jame Kanze wrote:
Daniel T wrote:
[Returning a const-reference to internal state can be
converted to a return of a const object of that state]
Unless, of course, the type doesn't support copy. Or if the
client takes the address of the return, and later uses it. Or
any number of other cases as well. The lifetime of the object
is *not* the same.
Well, in the former case, Object has no choice but to return a
reference (or pointer.)

And the case is rather frequent, don't you think.

That depends on how often you use value semantics.

You use value semantics when value semantics are called for.
Entity objects almost never have reference semantics. And in
many cases, entity objects have to be looked up using an
externally specified key.

[What] if the client takes the address of [a const-reference]
return, and later uses it.
In the latter case... I don't concern myself with such
pathological behavior by the client. I don't guarantee that state
returned by const reference will last past the next sequence
point in any case.

In sum, you require the results of a lookup in a map to be used in
the expression doing the lookup, and not later. That's very, very
counter-intuitive and unexpected. And also very difficult to use
correctly, since the client does have to test whether the lookup
found something before using it.

A map does not return a non-const reference to any internal
state, so I don't see how that relates in any way.

std::map certainly does return references to internal state, as
does every other map I've every seen or heard of.

 

[Daniel T.]

std::map certainly does return references to internal state, as
does every other map I've every seen or heard of.

Really, could you give an example? Maybe I'm mistaken...

 

[jkherciueh]

Really, could you give an example? Maybe I'm mistaken...

std::map< int, int > the_map;
...
the_map[5] = 6; // the_map[5] returns an int&.

I conjecture that the two of you are in disagreement about whether that is
an "internal state" of the map. Since I have no idea what "internal state"
is supposed to mean with regard to std::map, I will not offer an opinion.


Best

Kai-Uwe Bux