Robert Ramey schrieb:
Markus Schöpflin wrote:
Robert Ramey wrote:
c)I've noticed that bjam doesn't detect failure of test_exported which is an imported test. I've got it passing on my branch but I don't know what the true status is on the trunk. The test_exported tests are failing for a wide range of testers. (gcc 4.1.1, gcc 4.2.1, gcc 4.3.0, hp_cxx) What do you mean by saying that this failure isn't detected by bjam?
When I tested on my own system, i was getting a test that trapped on an assert(false) being marked "pass". This was when compiled for debug. When I compiled for release, the test failed with "unregistered cast". I wasn't particularly confident that I wasn't making some sort of mistake or drawing some false conclusion, but I believe that it was subsequently confirmed and a bjam fix was checked in - (at least for *nix).
From looking at http://beta.boost.org/development/tests/trunk/developer/serialization.html I'm don't know for a fact which variant is being run.
From looking at the compiler options of the failing tests, I would think that the gcc tests are run in the debug variant. And I'm sure that hp_cxx is running the debug variant, as I'm the one running the tests.
So it seems that here are couple of real possibilities:
a) export functionality fails on gcc 4.1+ compilers. Seems quite likely from looking at the table and the fact the implementation of export is compiler dependent.
There is hopefully someone around with access to one of these gcc versions which is able to help. I could look into the failures on hp_cxx tomorrow, perhaps the failures are related.
b) It concievable that its failing on all platforms but not being detected.
In any case, I don't have gcc 4.1+ available so its hard for me to look at.
Robert Ramey
Markus