[boost] Re: Serialization Draft # 12

From: Robert Ramey (ramey_at_[hidden])
Date: 2003-09-17 18:52:36

Date: Wed, 17 Sep 2003 14:30:50 -0400
From: "Jeff Flinn" <TriumphSprint2000_at_[hidden]>
To: 
Subject: [boost] 
Message-ID: <bka99b$cm2$1_at_[hidden]>
References: 
        <01C37B64.851AD4A0_at_[hidden]><bk5fbe$to5$1_at_[hidden]>
        <bk6vmc$tan$1_at_[hidden]>
Precedence: list
Message: 10

Robert,

Jeff Flinn wrote:

>I've been able to build the library in release mode successfully as well. 

Does this include running all the tests?

The bjam tools for VC release build don't seem to include the /Gy switch
(enable function level linking).  This is important to me for making keeping
executables small.  Unfortunately I've had problems in the past where this
through away "too much" code resulting in runtime problems.

>Although I'm not able to directly archive shared_ptr<>'s. I get the following compilation error
>attempting to compile demo_shared_ptr:

Same thing happens on my 7.0 system.  Investigation reveals that code has been
broken.   Its not hard to fix and I will included a fixed version in draft # 13.  In
fact, I plan to add the demos to the test set via a wrapper.

>I've found a problem with binary_archive missing archive header information.
>The same text_archive starts with: "22 serialization::archive 2 0 0 0 0 24"
>where the last 24 is the count of std::strings for a std::set<std::string>
>with 24 entries. In the binary_archive there are two NULL bytes before the
>24(0x18).

well binary files render 0 values as NULL bytes so that is not surpising.
was there any indication that there was a problem? was the archvie readable?

>Ease of implementation and performance are outstanding.
don't forget to show up if this thing ever gets reviewed again.

>I'm successfully text-archiving in one of my projects. 

That's very encouraging.   Especially since I have recently been struggling
with the following.

Comeau - 
a) The library seems to build - though it emits a wierd message
relating to creation of aout.exe
b) I've been unable to link an exe that uses any boost library
c)compiing, linking and executing small tests programs that don't
use boost libraries seems to work fine.
d)I have a problem with compiliation of programs that use
non-intrusive serialization - its torturosly difficult to make a 
test case small enough to demonstrate or resolve the
problem
e)I have found Comeau very useful for picking through the
C++ nits that seem to show up regardless of how much
one tries to avoid it - I really want to be successful with
this in making the tests build and run.

GCC 3.2 / cygwin
This platform doesn't support wide char i/o out of the box.
library compiles and all tests link.  XML tests fail
when test_main is entered - I havn't had time to 
look into this but would be curious to know if its
a GCC 3.2 issue - That is, what happens with GCC 3.3?

VC 7.0 This is my main system so everything builds and runs
VC 7.1 You seem to be doing great here
VC 6.5 I've gotten much help. I hope to have this
running in the course of next week.  This might not include
XML due to some issues with mulitpass and codecvt on this platform.
Borland - lots of help has given.  But it seems more problematic
than the others.  I think this should be left for last.

Thanks to all  for all your help.

Robert Ramey