portable and save reinterpret_cas t?

[ralpe]

Hi,

I have a template class foo which stores pairs of Keys and Ts in a
vector. Foo has an accessor function at(size_t) that returns a
reference to the pair at the specified index.

I do not want the users of foo to change the first value of a pair,
so I want to return a reference to a pair<const Key, T>, just like
std::map does. I cannot store pairs with a const first value in the
vector because such pairs would not be assignable.

My only solution so far is to use reinterpret_cas t. Is this a
portable solution that is guaranteed to work on all platforms?

template<typena me Key, typename T>
class foo
{
public:

std::pair<const Key, T>& at(std::size_t index)
{
return reinterpret_cas t<std::pair<con st Key, T>&>(pairs_.at( index));
}

void push_back(const std::pair<const Key, T>& pair)
{
pairs_.push_bac k(pair);
}

private:

std::vector<std ::pair<Key, T pairs_;
};

Thanks in advance,
Ralpe

[Ondra Holub]

ralpe napsal:

Hi,

I have a template class foo which stores pairs of Keys and Ts in a
vector. Foo has an accessor function at(size_t) that returns a
reference to the pair at the specified index.

I do not want the users of foo to change the first value of a pair,
so I want to return a reference to a pair<const Key, T>, just like
std::map does. I cannot store pairs with a const first value in the
vector because such pairs would not be assignable.

My only solution so far is to use reinterpret_cas t. Is this a
portable solution that is guaranteed to work on all platforms?

template<typena me Key, typename T>
class foo
{
public:

std::pair<const Key, T>& at(std::size_t index)
{
return reinterpret_cas t<std::pair<con st Key, T>&>(pairs_.at( index));
}

void push_back(const std::pair<const Key, T>& pair)
{
pairs_.push_bac k(pair);
}

private:

std::vector<std ::pair<Key, T pairs_;
};

Thanks in advance,
Ralpe

I think that simpler solution is to return std::pair<const something,
something_elsei nitalized with std::pair<somet hing, something_else>
value:

std::pair<const int, intTest()
{
std::pair<int, intp(1, 2); // This is only for illustration
return p;
}

[ralpe]

Ondra Holub wrote:

I think that simpler solution is to return std::pair<const something,
something_elsei nitalized with std::pair<somet hing, something_else>
value:

std::pair<const int, intTest()
{
std::pair<int, intp(1, 2); // This is only for illustration
return p;
}

No, that is no a solution because I need to return a reference to make
this possible:

myFoo.at(42).se cond = 4711;

[Ondra Holub]

ralpe napsal:

Ondra Holub wrote:

I think that simpler solution is to return std::pair<const something,
something_elsei nitalized with std::pair<somet hing, something_else>
value:

std::pair<const int, intTest()
{
std::pair<int, intp(1, 2); // This is only for illustration
return p;
}

No, that is no a solution because I need to return a reference to make
this possible:

myFoo.at(42).se cond = 4711;

You're right, with references it is problem. I think that
reinterpret_cas t should work everywhere for this case, becasue both
types differ only in const, so they should occupy exactly the same
place with the same alignment. I do not think, that const could lead to
some placement or alignment optimizations, but I am not sure.

Maybe 100% safe solution is to change API and do not return pair.

[ralpe]

Ondra Holub wrote:

You're right, with references it is problem. I think that
reinterpret_cas t should work everywhere for this case, becasue both
types differ only in const, so they should occupy exactly the same
place with the same alignment. I do not think, that const could lead to
some placement or alignment optimizations, but I am not sure.

The reinterpret_cas t-solution seems to work with Microsoft C++.
I think I'm going to use this solution if it will work with gcc too.

Maybe 100% safe solution is to change API and do not return pair.

Change of API is not a solution either. I need to write a class that
has the same API as std::map and is implemented in terms of a sorted
array/vector. So the class won't really have a at()-member function but
it will have iterators that return references to pair<const Key, T>.

Thanks,
Ralpe

[Sylvester Hesp]

Of course, if someone choses to specialize either (but not both)
std::pair<K, Vor std::pair<const K, Vto do something totally different
(for example, to swap the order of first and second in memory) you're
screwed. The question is whether this is important to you, but it is of
course perfectly legal C++ _and_ compatible with std::map.

Wouldn't it be a better solution if you let your map implementation
internally use a pair<const K, Vas well and const_cast away the const on
p.first whenever you need to change it? Then you won't have the issue of
pair<const K, Vand pair<K, Vnot having the same memory layout.

- Sylvester
"ralpe" <ra************ @gmx.netwrote in message
news:11******** ************@v3 3g2000cwv.googl egroups.com...

Ondra Holub wrote:

>You're right, with references it is problem. I think that
reinterpret_ca st should work everywhere for this case, becasue both
types differ only in const, so they should occupy exactly the same
place with the same alignment. I do not think, that const could lead to
some placement or alignment optimizations, but I am not sure.

The reinterpret_cas t-solution seems to work with Microsoft C++.
I think I'm going to use this solution if it will work with gcc too.

>Maybe 100% safe solution is to change API and do not return pair.

Change of API is not a solution either. I need to write a class that
has the same API as std::map and is implemented in terms of a sorted
array/vector. So the class won't really have a at()-member function but
it will have iterators that return references to pair<const Key, T>.

Thanks,
Ralpe

[ralpe]

Sylvester Hesp wrote:

Wouldn't it be a better solution if you let your map implementation
internally use a pair<const K, Vas well and const_cast away the const on
p.first whenever you need to change it? Then you won't have the issue of
pair<const K, Vand pair<K, Vnot having the same memory layout.

- Sylvester

I tried to use pair<const K, Vinternally, but didn't know how to do
it:

typedef std::pair<const int, intvalue_type;
std::vector<val ue_typev;
v.push_back(val ue_type(1, 2)); // DOES NOT COMPILE

Thanks
Ralpe

[Jim Langston]

"ralpe" <ra************ @gmx.netwrote in message
news:11******** **************@ 11g2000cwr.goog legroups.com...

Sylvester Hesp wrote:

>Wouldn't it be a better solution if you let your map implementation
internally use a pair<const K, Vas well and const_cast away the const
on
p.first whenever you need to change it? Then you won't have the issue of
pair<const K, Vand pair<K, Vnot having the same memory layout.

- Sylvester

I tried to use pair<const K, Vinternally, but didn't know how to do
it:

typedef std::pair<const int, intvalue_type;
std::vector<val ue_typev;
v.push_back(val ue_type(1, 2)); // DOES NOT COMPILE

it would be:
v.push_back( std::make_pair< const K, V>( 1, 2 ) );
but it doesn't seem to like the const either.

Thanks
Ralpe

[Sylvester Hesp]

Right, well, in that case that's no option :)

- Sylvester

"ralpe" <ra************ @gmx.netwrote in message
news:11******** **************@ 11g2000cwr.goog legroups.com...

Sylvester Hesp wrote:

>Wouldn't it be a better solution if you let your map implementation
internally use a pair<const K, Vas well and const_cast away the const
on
p.first whenever you need to change it? Then you won't have the issue of
pair<const K, Vand pair<K, Vnot having the same memory layout.

- Sylvester

I tried to use pair<const K, Vinternally, but didn't know how to do
it:

typedef std::pair<const int, intvalue_type;
std::vector<val ue_typev;
v.push_back(val ue_type(1, 2)); // DOES NOT COMPILE

Thanks
Ralpe

[Sylvester Hesp]

"Sylvester Hesp" <s.****@oisyn.n lwrote in message
news:45******** *************@n ews.xs4all.nl.. .

Right, well, in that case that's no option :)

- Sylvester

"ralpe" <ra************ @gmx.netwrote in message
news:11******** **************@ 11g2000cwr.goog legroups.com...

>Sylvester Hesp wrote:

>>Wouldn't it be a better solution if you let your map implementation
internally use a pair<const K, Vas well and const_cast away the const
on
p.first whenever you need to change it? Then you won't have the issue of
pair<const K, Vand pair<K, Vnot having the same memory layout.

- Sylvester

I tried to use pair<const K, Vinternally, but didn't know how to do
it:

typedef std::pair<const int, intvalue_type;
std::vector<val ue_typev;
v.push_back(val ue_type(1, 2)); // DOES NOT COMPILE

Thanks
Ralpe

Btw my sincere apoligies for top-posting, I just figured out that's
considered bad posting style. I'll bottom-post from now on ;)

- Sylvester