[boost] Re: [serialization] Regressions?

From: Robert Ramey (ramey_at_[hidden])
Date: 2005-05-06 11:57:26

I've just had a chance to review your patches.

>> split_member.hpp.

I see no problem here.  Actually I have that in my next pending upload

>> interface_oarchive.hpp

As explained in my previous posting, I would consider this erroneous and I
believe it should be backed out.  Did you re-re-run the serialization tests
on your local machine before you checked this in?  I would be surprised if
this change doesn't generate a number of test failures in the test_const*
tests as well as array serialization on the borland compiler.  Compilers are
so permited to convert a const to a non-const by making a copy (although I
havn't seen this) so I think this is wrong for that reason as well.

>> iserializer.hpp

This is somewhat puzzling to me.   Actually I suspect it is a secondary
effect from the above change.   The issue of const-ness has resulted in
tweaks in the archive overrides.  The above change would have subtle effects
in archives derived from basicarchive.  This is due to the CRTP design.
After the above change is backed out, This should be checked again.

Robert Ramey