[boost] serialization and cwpro status

From: Robert Ramey (ramey_at_[hidden])
Date: 2004-08-10 14:33:54

Rene,

Here is my assessment of the status of the serialization library as it
relates to the Metrowerks compiler.

all versions of CodeWarrior C++

CW two-phase lookup threw me for a loop.  It took me long time to really
understand this.  Now that I do, I'm strongly convinced that it's a really
bad idea.  I don't know if its true - but I think I heard that it was
implemented in order to make template export work - which looks like another
bad idea to me. Oh well.

CWPro 9.2 - my home copy

All tests using utf8-codecvt fail.  This includes the test for utf8-codecvt
itself.  I have come to believe that its an error somewhere in the MSL 9.2
implementation of the standard library.

CWPro 8.3 - your machine

The tests which serialize arrays fail - I haven't looked at this too
carefully.  There might be an issue regarding references to arrays decaying
into pointers.  Borland also has this problem.

test_diamond fails with this compiler - I haven't figured out why it fails
only here.

CWPro 8.3 and 9.2

Hashed set serialization fails to produce the correct result at runtime.  I
haven't looked real closely at this. Since its particular to this version of
CWpro I suspect it might be a library problem.  I don't know how many
standard libraries include support for hashed sets so I don't know for a
fact that other compilers are in fact testing this.

serialization of polymorphic derived pointers.  

Forward declaration method. I had reorganize code in
archive_pointer_oserialize etc and use MSVC specific key words (used by
BOOST_FORCE_INCLUDE) in order to get certain things to instantiate static
variables whose pre-runtime construction is necessary for this to function.
This seems like a bad idea as CW supports lots of other platforms.  I tried
every alternative I could find but wasn't able to find a better one that
worked.

"Export" functionality suffers from a variation of the above which I have
not been able to overcome.  So tests which use export fail.  I've spent even
more time on this - without success.  It seems to me that CW is too smart in
eliminating code that it "knows" it won't need.

Actually I'm not even sure it's a CW issue.  Comeau has the same symptom.
That would suggest it's another manifestation of the problem I'm having
comprehending two-phase lookup.  

CW < 8

I believe that these versions will pass 100 %

Robert Ramey