[boost] serialization - test/compiler status

From: Robert Ramey (ramey_at_[hidden])
Date: 2004-07-29 12:04:46

I've been working to make changes in the serialization system to support
more platforms and pass more tests.  I'm approaching the end of what I can
do.  The following comments reflect the state of the library as I recently
uploaded and so may not be reflected by the current test tables.

Windows platforms from
http://www.meta-comm.com/engineering/boost-regression/developer/serializatio
n.html

Metroworks CodeWarrior 9.2. - all test failures are related to one of the
following three pending issues:

a) slist test doesn't compile. The error is related to the internals of the
slist implementation which I've been unable to decipher.

b) link errors/warnings of replicated symbols.  A number of tests - all
those that use export and perhaps others generate duplicate symbol warnings
for i/o streams.  The generated executables also fail with unexpected
messages.  I suspect an issue with the compiler configuration files
cwpro9.hpp for this platform.

c) tests using non-intrusive serialization fail to find overload for
serialization functions.  (Commeau also has this problem).  I suspect a
compiler bug.

So far I've been unable to resolve these issues and would be very
appreciative if someone who has and interest and experience with this
compiler would like to investigate this issues.

MSVC-STLPORT

a) In some environments - has a problem compiling spirit 1.6.1.  I believe
this has to be resolved by tweaking spirit.

Borland

Passes most tests - certain test failure indicate features which are not
ever going to be usable on this platform - basically automatic serialization
of arrays.

MSVC 6.0/7.0

Pass mosts tests - certain test failure indicate features which are not ever
going to be usable on this platform - basically automatic serialization of
arrays.  These are documented in the manual.

VC 7.1

Passing 100 %

VC 8.0

Small problem with compile error in library build - I'm hoping that this
will build and pass 100 %

Gcc 3.x

Passing 100 %.  Tests on platforms which don't support wide char i/o (e.g.
cygwin and mingw).  show failure for the related tests This is a feature of
the build/bjam system and are not to be considered failures.

Intel 7.1

This fails when export is used with invalid dynamic cast.  This is
unresolvable by me as I don't have this compiler on my system.  Anyone with
and interest in this compiler is invited to jump in.

Intel 8.0

Passing 100 %

MACOS - Darwin from
http://www.meta-comm.com/engineering/boost-regression/developer/serializatio
n.html

Previous test results show failure with wide-char i/o.  I believe I have
addressed this and hoped that next results will show 100 %.  Current test
results suggest an issue outside my influence.  The program links with
messages which suggest some sort of issue related to the test_exec_monitor.
I would be very appreciative if anyone else want's to take a look at this.

Linux - from
http://boost.sourceforge.net/regression-logs/cs-Linux/developer/serializatio
n.html

Gcc 3.2.3, 3.3.2, 3.3.3

Pass 100%

Gcc 2.95

This system doesn't support new style streams.  This might be made to work
but I can't justify the effort.

Gcc 2.95 - stlport

This probably could be made to work without tooo much problem - but as
before its hard to justify the effort given the availability of a free
upgrade to gcc 3.3+.  As in the case of all the stlport tests, its not clear
whether or not the version of stlport is built with streams and in some case
which version is being used.

GCC 3.4.1

This always passed 100 % - I made and corrected an obscure error while
working on other platforms.  I'm hoping for 100% next time

Intel 7.1 
Same as above for gcc 3.4.1

Intel 8.0

Hmmmm this passes 100% in windows and I believe has passed before.  This
situation is much confused by the policy of frequent releases of compiler
updates.  We label all of these intel 8.0 even though they are not the same.
This os/compiler combination exhibits problems with the array test and
export. I see it as unresolvable by me.  Interested parties are invited to
jump in.

Other platforms - from http://boost.sourceforge.net/regression-logs/

AIX

This is hung up in the compilation of large mpl expression.  Working around
this would require access to such a compiler platform which I don't have.
This will only get resolved if someone has sufficient interest/necessity to
invest the required effort.

HP-UX/PARISC - aCC

These compiles fail to build the library.  The error messages suggest that
these compilers don't support enough C++ template syntax to be useable
without undertaking heroic efforts and having direct access to the
compiler/platrom.

IRIX64

This libraries aren't being tested frequently enough to be useful for
resolving problems of library implementation.

I am marking these platforms unusable for the serialization library.

Robert Ramey