Boost logo

Boost Testing :

From: Robert Ramey (ramey_at_[hidden])
Date: 2005-07-19 11:28:05


As I write this ,I have no more corrections to make to the serialization
library. There are a number of test failures, but I cannot resolve them for
various reasons.

a) cw- 9_5- darwin - something in this setup prevents the library shared
library version from working with text base archives. The fact that tests
for binary archives DO function suggests that this is fixable but it would
take some digging by someone who has access to this combination of
os/compiler.

b) cw - There are still some problems with code instantiation. I don't
expect these to be resolved for the 1.33 release. Things are better than
the 1.32 release though.

c) acc - DLL/shared library version fails to build -
http://tinyurl.com/9u7kq - with a compile error from
boost/mpl/aux_/integral_wrapper.hpp This is probably fixable as this used
to work before something changed. However the acc traceback doesn't give
enough information to determine where the error originates.

d) gcc- 2.95.3- stlport- 4.6.2- linux - the library won't build and the
tests are skipped. This compiler requires that spirit 1.6x be installed on
the target machine and the environmental variable SPIRIT_ROOT be set to
indicate the directory where it was installed. With these changes, most of
the serialization library will pass with this compiler and the stlport
standard library.

e) gcc- 3.3.6- osf1 - tests that use wide character i/o fail. It seems that
the config file is indicating that wide char i/o is not supported with this
configuration. This is surprising to me. Perhaps its a config file issue.

f) como- 4_3_3- vc7_1 - All tests fail to link. The message suggests that
the Test Library and Serialization Library both instanticiate the same code
in the library. It doesn't seem to create problems with other compilers.
I've been unable to resolve this.

h) tru64cxx65- 042 would be giving 100% passing except for one test -
test_variant. Error message suggests that this should be addressed in the
variant package. In Boost Variant - this compiler is excluded from testing.
I think this should be changed and the tests re-enables. It seems that this
compiler is working very well and that only the smallest of adjustments
would be required to make it pass any tests.

g) tests named test_class_info_load .. fail sometimes due to the fact that
they rely on the existence of files produced by another test -
test_class_info_save. I've been unable to make bjam order the tests without
altering the Jamfile beyond the point of comprehension. I did test mark the
test as - fail sometimes - but that resulted in a display that I considered
confustion - unfortunately, I don't remember the details.

i) vacpp has been excluded from testing. I did this sometime ago as the
library would not build and the problems seemed too numerous to work around.

j) sunpro- 5* has been excluded from testing. The Boost Serialization
library (and only that library) makes extensive use of an idiom which is
believed to be legal but that these compilers fail to support.

Note - I haven't recentl run tests in release mode. I think I will do this.
Last time I checked, the Boost Test library couldn't be used with at least
one compiler I use to test (borland 5.51 (or maybe 5.64?) .

As of now, I'm just making additions to the documentation

Robert Ramey


Boost-testing list run by mbergal at meta-comm.com