Re: [Boost-users] [Serialization] Runtime error in load_pointer

Subject: Re: [Boost-users] [Serialization] Runtime error in load_pointer
From: Robert Ramey (ramey_at_[hidden])
Date: 2010-01-15 12:51:53

This is likely a secondary effect from some other error.  There's no
way I can help with this.

But still, I'll give it a try.

I know from previous questions that you're using serialization
pretty extensively in more complex scenarios: derived classes,
abstract classes, dlls, pointers etc.  So when you have a problem
the source of the failure isn't easy to find.  I don't know if you
already do this, but I'll suggest it anyway in case you don't.

Take a look at the tests in the serialization library.  There
is approximately 60 of them.  Most of them are run for
each of the archive classes delivered with the serialization
library. So that results in a couple of hundred distinct tests.

For each one of your classes, you should make a test.  This
test should include the archive class as a test/template
parameter like the serialization library tests do.  Your tests
should be run in particular order - base classes first.

test A
#include "A.hpp"
main(... // code create, save and load an instance of A and pointer to A
// maybe collections of A as well?

test B
#include "B.hpp" // which includes A
main(.. // code create, save and load an instance of B and pointer to B

Now I know that this "seems" like a huge amount of work.  But
in practice it's a lot less work than it seems.  This amount of investment
save huge amounts of time in the long run.

a) Once you've created the first test_A, subsequent ones are created by
copying this first one and changing A to B and some other changes.  It
generally takes only a few minutes.

b) Everytime anyone makes a change to a class A, the tests which use
this component are re-run.  This guarentees that you're not moving
backwards.  My principle environment is MSVC IDE.  Each test
is a separate project.  The post-build step is setup to run the test.
The project for test A is dependent upon the
code in A, the code in test A an the serialization library.
of these change, the test is automatically re-compiled, re-linked
and the test is re-run  automatically.  If any test fails, the build
fails and I can't go on without addressing the problem.  That is
I stop myself from building crappy code which I would otherwise do.

c) This forces you to maintain a hierarchy of composable components
and stops one's code from morphing into a "rat's nest" of inter-module
calls which is impossible to debug.

e) If something fails, you automatically have a test case to submit.

d) This makes bug-free code development MUCH faster.

I don't know if this helps - but it makes me feel better.  Good Luck.

Robert Ramey

elizabeta petreska wrote:
> Hello
> I do not have small working example, and can't reduce the problem to
> such. I know that is impossible to say what can be the problem from
> the following information, but still
> I 'll give it a try.
> I am getting runtime error in STL ( vector is out of range error )
> when trying to load my shared pointer.
> The runtime error happens in basic_iarchive.cpp line line 431 :
>
>        cobject_id_vector[i].bpis_ptr = bpis_ptr;
>
>
>  I am using boost 1.41. Anyone has seen this error and maybe knows
> what can get to it.
>
> Thanks
>
>
>
>
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users