STL-container used with references

S

Severin Ecker

hi!

normally i would simply do the following:

std::vector<Element> vec;
void somefunc() {
Element e;
vec.push_back(e);
}

now Element e is in the vector. thats fine as long as no polymorphic
behaviour is needed.

std::vector<Element &> vec;
void somefunc() {
Derived_from_Element e;
vec.push_back(e); //bad idea,.. e is gone after functionscope is left
}

what i want to avoid (if possible) are 2 things:
dynamic allocation of the objects, and having an extracontiner for every
type derived from the basetype to store the elements.

i hope, i made clear what i'm trying to do... so is there a solution (except
for the 2 mentioned above?)

thx, regards,
sev
 
T

tom_usenet

hi!

normally i would simply do the following:

std::vector<Element> vec;
void somefunc() {
Element e;
vec.push_back(e);
}

now Element e is in the vector. thats fine as long as no polymorphic
behaviour is needed.

std::vector<Element &> vec;

You can't hold references in containers - references are just aliases,
not real objects.
void somefunc() {
Derived_from_Element e;
vec.push_back(e); //bad idea,.. e is gone after functionscope is left
}

what i want to avoid (if possible) are 2 things:
dynamic allocation of the objects, and having an extracontiner for every
type derived from the basetype to store the elements.

i hope, i made clear what i'm trying to do... so is there a solution (except
for the 2 mentioned above?)

It is obviously hard to avoid dynamic allocation of the objects, since
all of your derived types can have different sizes and alignment
requirements. There are techniques for doing it (as long as all
derived classes are known in advance), but they would be premature
optimization in this case I am sure. Your best bet is to use a
container of smart pointers:

std::vector<shared_ptr<Element> > vec;
vec.push_back(shared_ptr<Element>(new Derived_from_Element));

See www.boost.org for shared_ptr.

Tom
 
H

Howard

tom_usenet said:
...Your best bet is to use a
container of smart pointers:

std::vector<shared_ptr<Element> > vec;
vec.push_back(shared_ptr<Element>(new Derived_from_Element));

See www.boost.org for shared_ptr.

Just curious...What if you don't have (or want) boost? Is there an STL
solution similar to shared_ptr?

-Howard
 
J

John Harrison

Howard said:
Just curious...What if you don't have (or want) boost? Is there an STL
solution similar to shared_ptr?

No, but it's really very simple to roll your own basic smart pointer. It
would not have all the functionality of boost's but would certainly support
polymorphism and automatic cleanup.

Have a look at Scott Meyers book for example code (I think its the More
Effective C++ one).

john
 
T

tom_usenet

Just curious...What if you don't have (or want) boost? Is there an STL
solution similar to shared_ptr?

boost::shared_ptr has been proposed for standardization as part of the
library technical report. Look out for std::tr1::shared_ptr, coming to
your compiler soon.

There is no current standard solution, except to write your own smart
pointer class (not at all recommended - matching boost::shared_ptr's
functionality is non-trivial).

Tom
 
M

Mike Wahler

tom_usenet said:
boost::shared_ptr has been proposed for standardization as part of the
library technical report. Look out for std::tr1::shared_ptr, coming to
your compiler soon.

Curiosity:

What's the significance of the (namespace?) name 'tr1'?

-Mike
 
P

Pete Becker

Mike said:
What's the significance of the (namespace?) name 'tr1'?

The C++ standards committee has a Technical Report in the works,
incorporating recommended library extensions. It's known informally as
TR1, and its extensions go in namespace std::tr1.
 
K

Kevin Goodsell

Pete said:
The C++ standards committee has a Technical Report in the works,
incorporating recommended library extensions. It's known informally as
TR1, and its extensions go in namespace std::tr1.

I don't like the sound of this. Are they going to permanently put these
things in std::tr1::? Why wouldn't they just use std::? If they put it
in std:: later, will they have to also support it in std::tr1:: for
compatibility?

-Kevin
 
P

Pete Becker

Kevin said:
I don't like the sound of this. Are they going to permanently put these
things in std::tr1::?

TR1 puts them in std::tr1. Future TRs and future standards could do
something different.
Why wouldn't they just use std::?

Because they're recommended extensions and not part of the standard.
If they put it
in std:: later, will they have to also support it in std::tr1:: for
compatibility?

Maybe. Maybe the new stuff won't ever go into the standard.
 
K

Kevin Goodsell

Pete said:
Because they're recommended extensions and not part of the standard.

OK. I didn't make the connection that a TR is different from a TC, and
was thinking that this /would be/ part of the standard (the comment
"coming to your compiler soon" threw me off). If it's just a
possibility, this makes more sense.

-Kevin
 
S

Sergei Matusevich

std::vector<Element &> vec;
[...]

what i want to avoid (if possible) are 2 things:
dynamic allocation of the objects, and having an extracontiner for every
type derived from the basetype to store the elements.

i hope, i made clear what i'm trying to do... so is there a solution (except
for the 2 mentioned above?)

Severin,

You cannot have a container of references, but you very well can have
a container of pointers, i.e. something like std::vector<Element*>.
Better yet, you can use smart pointers, e.g. boost::shared_ptr<Element>.
In any case, remember that you have to dereference pointers (e.g. using
functors) when you access the data (for example, using standard
algorithms).

You may also want to make your Element class a lightweight proxy that
holds a reference to an actual data stored outside the container.
(Some kind of a Featherweight pattern?)

Hope this helps!
Sergei.
 
T

Thorsten Ottosen

[snip]
You cannot have a container of references, but you very well can have
a container of pointers, i.e. something like std::vector<Element*>.
Better yet, you can use smart pointers, e.g. boost::shared_ptr<Element>.

maybe even better, use a container that takes ownership of the pointers.

checkout the ptr_container lib in the boost sandbox.

br

Thorsten
 
T

tom_usenet

The C++ standards committee has a Technical Report in the works,
incorporating recommended library extensions. It's known informally as
TR1, and its extensions go in namespace std::tr1.

How long till Dinkumware starts to track TR1 extensions? Will you wait
until its almost finalised or even finished (the conservative,
possibly sensible approach)?

Tom
 
P

Pete Becker

tom_usenet said:
How long till Dinkumware starts to track TR1 extensions? Will you wait
until its almost finalised or even finished (the conservative,
possibly sensible approach)?

If all goes well, the technical work on TR1 will be finished in October.
A significant number of changes from the original proposals have been
the result of our work implementing it.
 
D

Dietmar Kuehl

tom_usenet said:
How long till Dinkumware starts to track TR1 extensions? Will you wait
until its almost finalised or even finished (the conservative,
possibly sensible approach)?

I cannot speak for Dinkumware, of course, but I know that library writers
are already implementing what is currently in the TR. Actually, quite a
few of the defects we processed two weeks ago seem to come from people
implementing the TR.

Of course, this is no indication when there will be a release of the TR
components, even if they closely track what is going on: since there are
probably still some changes which will be applied, I think it would be
unwise to release an implementation of the TR libraries just now. I would
expect that you will be able to purchase or download a version relatively
soon after the TR is finalized. I'm always confusing the data but I think
the plan is to try and essentially finalize the TR at the next meeting ie.
in October in Redmont.

BTW, does anybody have a good open source implementation of the special
functions in the lib TR? Well, not necessarily in C or C++ but at least
the same functionality - or an idea how this stuff can be implemented?
Some people apparently know how to do such stuff but I considered numerics
a waste of time while at the university and in any case they didn't talk
about computing these functions anyway (as far as I can remember...).
 
P

P.J. Plauger

I cannot speak for Dinkumware, of course, but I know that library writers
are already implementing what is currently in the TR. Actually, quite a
few of the defects we processed two weeks ago seem to come from people
implementing the TR.

Of course, this is no indication when there will be a release of the TR
components, even if they closely track what is going on: since there are
probably still some changes which will be applied, I think it would be
unwise to release an implementation of the TR libraries just now. I would
expect that you will be able to purchase or download a version relatively
soon after the TR is finalized. I'm always confusing the data but I think
the plan is to try and essentially finalize the TR at the next meeting ie.
in October in Redmont.

Correct, and as Pete Becker indicated, Dinkumware has been implementing all
of
the pieces of TR1 for over a year now. We hope to have a *full* version soon
after the document freezes, which we still hope will be this October. Note
that it's a *very big* addition, including all of the functions added to C
with C99.
BTW, does anybody have a good open source implementation of the special
functions in the lib TR? Well, not necessarily in C or C++ but at least
the same functionality - or an idea how this stuff can be implemented?
Some people apparently know how to do such stuff but I considered numerics
a waste of time while at the university and in any case they didn't talk
about computing these functions anyway (as far as I can remember...).

The C committee has asked for an appendix to their special functions TR
to indicate at least some way to implement each of these functions. Our
work to date has unearthed a few free functions that are excellent, quite
a few that are barely adequate to float precision, and numerous approaches
that are mediocre but make a reasonable starting point. If there's a
complete, high quality set out there we haven't found it yet. They're
pretty tough, in general.

P.J. Plauger
Dinkumware, Ltd.
http://www.dinkumware.com
 
H

Harald Nowak

maybe even better, use a container that takes ownership of the pointers.
checkout the ptr_container lib in the boost sandbox.

br

Thorsten

Now THIS is finally what I've been waiting for for the last 5 years -
a crucial missing piece - I do certainly hope that the ptr_container
will make it into the official boost branch and from there as quickly
as possible into the standard (or the tr).
In the last years I've continuously pointed out that omission and
tried to provide workarounds that deal with STL containers of pointers
- most of the time people seemed just happy with containers of
shared_ptr's but the main point is and was, that it is dangerous and
yet at the same time possible to use normal pointers in STL
containers: dangerous because algorithms can be innocently applied (as
long as the functors are written appropriately) and yet have
devastating effects (leaks and crashes - see my Short Tip in C++ Users
Journal: "A remove_if for vector", C/C++ Users Journal, July 2001. ).
If finally a dedicated version of the STL containers exist (like
ptr_container delivers) users will be guided the correct way - thats
why I appreciate this library so much.
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Members online

Forum statistics

Threads
473,731
Messages
2,569,432
Members
44,835
Latest member
KetoRushACVBuy

Latest Threads

Top