[boost] Re: Serialization Draft #13 uploaded to files section

From: Robert Ramey (ramey_at_[hidden])
Date: 2003-11-03 14:52:37

Vladimir Prus wrote:

>Robert Ramey wrote:

>> Looking at the status table, I would
>> bet that finding/fixing an error in the text_archive will clear up a lot of
>> failures.  Then I would go for void_cast as its more self contained and
>> could explain the pointer failures.  Of course finding the last problem
>> is going to be the hardest.

>I found the first problem. When saving wchar_t, you were casing it to signed
>short. When loading, you were loading it as unsigned short. Don't know why
>this crashed tests, exactly, but when the stored value was negative, they
>crashed.

When I apply the attached patch, I get much nicer results. Again, see

>   http://zigzag.cs.msu.su:7813/serialization.html

>However, I'm really suspicious of passing wchar_t as unsigned shorts.
>wchar_t is 32 bits on some compilers, and gcc is one of them -- which means
>you're loosing some data.

very good, the whole wchar_t <-> int thing is going to have to be looked 
at a little more.  Oh then there's the case where wchar_t is just a typedef
for some int.  I notice that all the wide chare tests are failing so it seems
there's a little more to be done here in the short term as well.

I'm intrigues that xml_warchive tests pass as they use wchar as well.  You
might want to double check the output.  In my case they all intially came
up as "pass" even the gcc aborted imediately with a stack dump!

>I'll probably look at test_delete test next. Any ideas how to debug it, or
>what could go wrong?

I'm sure it fails in some sort of catastrophic way so I shouldn't be too hard.

Great job - good luck

Robert Ramey