Re: [Boost-users] Serialisation across libraries.

From: Robert Ramey (ramey_at_[hidden])
Date: 2007-05-08 21:57:07

I strongly suspect that this is a manifestation of a subtle
bug/feature of serialization 133 and maybe 1.34.

Its hard to explain in a short email.  But the short
version is that serialization for the same type is
included more than one place - a shared library
as well as mainline or another shared library.

So you can consider how your code is distributed
among the libraries with the idea that serialization
for the same type is not spread accross various
shared libraries.

Robert Ramey

Hughes, James wrote:
> Hi Robert,
>
> I was worried there might not be enough information, so here goes. See
> below.
>
>> -----Original Message-----
>> From: boost-users-bounces_at_[hidden]
>> [mailto:boost-users-bounces_at_[hidden]] On Behalf Of Robert
>> Ramey Sent: 08 May 2007 15:57
>> To: boost-users_at_[hidden]
>> Subject: Re: [Boost-users] Serialisation across libraries.
>>
>> A little extra information would be helpful.
>>
>> a) what compiler/linker etc are you using
>
> gcc 3.4.3  (Cross compiling to Arm)
>
>
>> b) which version of boost serialization are you using.
>
> A completely standard 1.33.1
>
>> c) what kind of library: static, DLL, or?
>
> All the libs involved are shared.
>
>> d) Is this an issue related to shared_ptrs or more general?
>>
>
> There are shared_ptr's involved. I will endeavour to describe the data
> structure in the main library in more detail (limited to the Text case
> which is all I am testing currently).
>
> Message    - the basic data to be serialised
>  contains some basic data and a vector of shared_ptrs to Panels (which
> is a base class of a number of different panel types)
>
> Panels    - Contains some basic member data
>
> TextPanel - Derived from Panels, contains some member data and a
> structure, PanelFieldsList
>
> PanelFieldsList contains a vector of shared_ptrs to Fields (which is a
> base class to a number of different field types).
>
> Fields
> Base class contains a vector of raw pointers back to any panels which
> reference them (but not currently serialised - trying to get the
> basics working first - may add these in a post process of the loaded
> data to save brain ache)
>
>
> Now....The first area of problem, and the one described in my first
> post, is that when I attempt to serialise a data structure that is
> contained in an instance of Panels, it saves OK but on loading back
> in I get this released twice/corruption error. The data structure is
> defined in a header file from a different library (where is
> implementation is). The header file for that DS contains the
> serialisation code, and a CPP file in this other library contains a
> BOOST_SERIALIZATION_EXPORT_GUID statement. Therefor this GUID is NOT
> available to the first library until link time. I do wonder if this
> is where the problem arises. I would like to keep the serialisation
> intrusive, in the appropriate headers in the appropriate library
> rather than have to move to a different location.
>
>
> Now that I have described the DS in more detail, I can also add
> another problem. Depending on which bits I attempt to serialise, on
> loading I keep getting bus errors at various points, depending on
> which parts of the DS I try to serialise. The save appears to work
> fine - I have enclosed a dump of the saved DS.
>
> I hope someone has some idea of what's going on, because I have been
> staring at this all day to no avail.
>
> Regards
>
> James
>
>
> Dump :::: Sorry about wrapping.....Is there any way I can reduce the
> number of spaces output for indentation???
>
>
> <?xml version="1.0" encoding="UTF-8" standalone="yes" ?>
> <!DOCTYPE boost_serialization>
> <boost_serialization signature="serialization::archive" version="3">
> <Message class_id="0" tracking_level="0" version="0">
>        <Name>Message1</Name>
>        <Filename class_id="1" tracking_level="0" version="0">
>                <path>/home/jhughes/msg.xml</path>
>        </Filename>
>        <PanelList class_id="2" tracking_level="0" version="0">
>                <count>1</count>
>                <item class_id="3" tracking_level="0" version="1">
>                        <px class_id="4" class_name="TextPanel"
> tracking_level="1" version="0" object_id="_0">
>                                <LCPanel class_id="5"
> tracking_level="0" version="0">
>                                        <RasterHeight>7</RasterHeight>
>                                        <PanelWidth>0</PanelWidth>
>                                        <PanelHeight>7</PanelHeight>
>                                </LCPanel>
>                                <TextPanelFields class_id="6"
> tracking_level="1" version="0" object_id="_1">
>                                        <LCPanelFields class_id="7"
> tracking_level="1" version="0" object_id="_2">
>                                                <LCPanelFieldsFacade
> class_id="8" tracking_level="1" version="0"
> object_id="_3"></LCPanelFieldsFacade>
>                                                <Fields class_id="9"
> tracking_level="0" version="0">
>
>
> <count>1</count> <item
> class_id="10" tracking_level="0" version="1">
>                                                                <px
> class_id="12" class_name="StaticTextField" tracking_level="1"
> version="0" object_id="_4">
>
> <LCTextField class_id="13" tracking_level="1" version="0"
> object_id="_5">
>
> <LCField class_id="11" tracking_level="1" version="0"
> object_id="_6"></LCField>
>
> <Text></Text>
>
> </LCTextField>
>                                                                </px>
>                                                        </item>
>                                                </Fields>
>                                        </LCPanelFields>
>                                </TextPanelFields>
>                        </px>
>                </item>
>        </PanelList>
> </Message>
> </boost_serialization>
>
>
>
>
>> Robert Ramey
>>
>> Hughes, James wrote:
>>> Hello all,
>>>
>>> I am a bit stumped by a problem I am seeing with serialisation, and
>>> wondered whether it may be due to the way I sue some libraries.
>>>
>>> I have two libraries, one which contains the main code that uses
>>> serialisation, and another which contains some small
>> classes used by
>>> the main code.
>>>
>>> The main code serialises a tree like data structure containing some
>>> shared pointers, each node of the 'tree' contains member
>> data, and its
>>> one of these members that is causing a double free or
>> corruption error
>>> on loading. This particular member is a structure from the other
>>> library. It contains a serialisation template in the class header
>>> which appears to save correctly - the XML output all
>> appears correct.
>>> However on loading the double free error occurs.
>>>
>>> I cannot see anything wrong with the code at all. The only
>> thing I can
>>> think of is that there are limitations in doing
>> serialisation across
>>> the library boundary. Perhaps I end up with two
>> instantiations of the
>>> template for serialisation? One in the main library, and one in the
>>> small library? Am I clutching at straws? DO I somehow have to force
>>> the instantiation of the serialisation for the class in the sub
>>> library to be in there, rather than being created again in the main
>>> library?
>>>
>>> Any thoughts?
>>>
>>> James
>>>
>>> This message (including any attachments) contains
>> confidential and/or
>>> proprietary information intended only for the addressee.
>>> Any unauthorized disclosure, copying, distribution or
>> reliance on the
>>> contents of this information is strictly prohibited and may
>>> constitute a violation of law.  If you are not the intended
>>> recipient, please notify the sender immediately by responding to
>>> this e-mail,
>> and delete
>>> the message from your system.  If you have any questions about this
>>> e-mail please notify the sender immediately.
>>
>>
>>
>> _______________________________________________
>> Boost-users mailing list
>> Boost-users_at_[hidden]
>> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>>
>
> This message (including any attachments) contains confidential
> and/or proprietary information intended only for the addressee.
> Any unauthorized disclosure, copying, distribution or reliance on
> the contents of this information is strictly prohibited and may
> constitute a violation of law.  If you are not the intended
> recipient, please notify the sender immediately by responding to
> this e-mail, and delete the message from your system.  If you
> have any questions about this e-mail please notify the sender
> immediately.