Re: [Boost-users] [Serialization] Strange giant warning

newer
Re: [Boost-users] [Serialization]...

Avi Bahra

9 Mar 2010 9 Mar '10

1:38 p.m.

This is caused by a bug in the boost serialisation, well at least in 1.39 Where is assumes that shared ptr must be polymorphic I have already reported a similar problem. i.e

...

...
... boost_1_39_0/boost/ serialization/extended_type_info_typeid.hpp", line 88: warning #2414-D: delete of pointer to incomplete class BOOST_STATIC_WARNING(boost::is_polymorphic<T>::value); ^detected during: instantiation of "const boost::serialization::extended_type_info

*boost::serialization::extended_type_info_typeid<T>::get_derived_extended_type_info(const

...
T &) const [with T=Concrete]" at line 92 of

"/scratch/ma/emos/ma0/hpia64/boost/boost_1_39_0/boost/archive/shared_ptr_helper.hpp"

...
It looks like if we are serialising via a base ptr, and the base class is serialising concrete data member via a shared ptr, it has assumed that the type must be polymorphic. This does not seem right to me ?

The hacky way to stop your warning message would be add a dummy virtual function. Best regards, Ta, Avi

Attachments:

attachment.html (text/html — 1.3 KB)

Show replies by date

OvermindDL1

11 Mar 11 Mar

3:40 a.m.

New subject: [Serialization] Strange giant warning

On Tue, Mar 9, 2010 at 6:38 AM, Avi Bahra wrote:

...

This is caused by a bug in the boost serialisation, well at least in 1.39 Where is assumes that shared ptr must be polymorphic I have already reported a similar problem. i.e

Oh I *SO* hate that bug, so very much. On Tue, Mar 9, 2010 at 6:38 AM, Avi Bahra wrote:

...

...
...
It looks like if we are serialising via a base ptr, and the base class is serialising concrete data member via a shared ptr, it has assumed that the type must be polymorphic. This does not seem right to me ? The hacky way to stop your warning message would be add a dummy virtual function.

Which I had to do, which added a virtual table, which bloats the size of my small classes needlessly...

Robert Ramey

4:31 a.m.

New subject: [Serialization] Strange giant warning

OvermindDL1 wrote:

...

On Tue, Mar 9, 2010 at 6:38 AM, Avi Bahra wrote:

...
This is caused by a bug in the boost serialisation, well at least in 1.39 Where is assumes that shared ptr must be polymorphic I have already reported a similar problem. i.e

Oh I *SO* hate that bug, so very much.

what makes you think it's a bug in serialization library? If I recall correctly, the motivation for the warning is the following. class base { // no virtual functions here !! ... }; class derived : public base { ... }; BOOST_CLASS_EXPORT(derived); ... base * b = new derived; ar << b; // b will be "sliced" - only the base class part will be serialized !!! In this case the user is almost certainly making a mistake which will be exceeding difficult to find. What's more, he well probably have to send inumermable emails to this list alleging that the serialization library has a bug and I will have do debug his program for him.. If you want to get ride of the warning, either make the base class virtual or use derived *d = new derived; ar << d; I'm speaking from memory and it's possible that the situation has changed since 1.39 and the message is now spurious. If so send a small test case which shows that the library has a bug. Robert Ramey

OvermindDL1

10:02 a.m.

New subject: [Serialization] Strange giant warning

On Wed, Mar 10, 2010 at 9:31 PM, Robert Ramey wrote:

...

OvermindDL1 wrote:

...
On Tue, Mar 9, 2010 at 6:38 AM, Avi Bahra wrote:

...
This is caused by a bug in the boost serialisation, well at least in 1.39 Where is assumes that shared ptr must be polymorphic I have already reported a similar problem. i.e

Oh I *SO* hate that bug, so very much.

what makes you think it's a bug in serialization library?

If I recall correctly, the motivation for the warning is the following.

class base { // no virtual functions here !! ... };

class derived : public base { ... };

BOOST_CLASS_EXPORT(derived);

...

base * b = new derived; ar << b; // b will be "sliced" - only the base class part will be serialized !!!

In this case the user is almost certainly making a mistake which will be exceeding difficult to find. What's more, he well probably have to send inumermable emails to this list alleging that the serialization library has a bug and I will have do debug his program for him.. If you want

There is no hierarchy in my case, in fact it is just a struct with data members and shared_ptr's that link everything together, it does not compile, that is pretty irritating to have to create a freaking virtual function in a struct when it holds nothing but data.

David

2:01 p.m.

New subject: [Serialization] Strange giant warning

Sorry for not replying sooner... I was distracted from my comp for a few days. I put that virtual destructor dummy into the problem classes and voila, my warnings are gone. Thanks alot. I can see the need for it now, and consider it good practice. I'm using boost 1.41 btw. Thanks again. Thanks so much!

Robert Ramey

3:49 p.m.

New subject: [Serialization] Strange giant warning

OvermindDL1 wrote:

...

There is no hierarchy in my case, in fact it is just a struct with data members and shared_ptr's that link everything together, it does not compile, that is pretty irritating to have to create a freaking virtual function in a struct when it holds nothing but data.

Are you using BOOST_CLASS_EXPORT when it is in fact not necessary? Robert Ramey

...

_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users

5362

Age (days ago)

5364

Last active (days ago)

List overview

Download

5 comments

4 participants

participants (4)

Avi Bahra
David
OvermindDL1
Robert Ramey