Boost regression notification (2005-10-13 [RC_1_33_0])
Boost regression test failures ------------------------------ Report time: 2005-10-13T14:34:13Z This report lists all regression test failures on release platforms. Detailed report: http://engineering.meta-comm.com/boost-regression/CVS-RC_1_33_0/developer/is... 75 failures in 6 libraries: graph (4) parameter (2) python (2) range (3) serialization (63) variant (1) |graph| copy: mingw-3_4_2 graph_concepts: mingw-3_4_2 graphviz_test: cw-9_4 vector_graph_cc: mingw-3_4_2 |parameter| tutorial: gcc-2.95.3-linux unnamed: vc-7_1 |python| pointer_vector: cw-9_4 vector_indexing_suite: cw-9_4 |range| array: vc-6_5 vc-6_5-stlport vc-7_0 |serialization| test_demo_xml_load: gcc-3_3-darwin gcc-3_4_3-sunos gcc-4_0-darwin intel-win32-8_1 test_demo_xml_load_dll: gcc-3_3-darwin gcc-3_4_3-sunos gcc-4_0-darwin test_exported_binary_archive: vc-6_5 test_exported_binary_archive_dll: vc-6_5 test_exported_text_archive: vc-6_5 test_exported_text_archive_dll: vc-6_5 test_exported_text_warchive: vc-6_5 test_exported_text_warchive_dll: vc-6_5 test_exported_xml_archive: vc-6_5 test_exported_xml_archive_dll: vc-6_5 test_exported_xml_warchive: vc-6_5 test_exported_xml_warchive_dll: vc-6_5 test_no_rtti_binary_archive: vc-6_5 test_no_rtti_binary_archive_dll: vc-6_5 test_no_rtti_text_archive: vc-6_5 test_no_rtti_text_archive_dll: vc-6_5 test_no_rtti_text_warchive: vc-6_5 test_no_rtti_text_warchive_dll: vc-6_5 test_no_rtti_xml_archive: vc-6_5 test_no_rtti_xml_archive_dll: vc-6_5 test_no_rtti_xml_warchive: vc-6_5 test_no_rtti_xml_warchive_dll: vc-6_5 test_shared_ptr_132_binary_archive: vc-6_5 vc-6_5-stlport test_shared_ptr_132_binary_archive_dll: vc-6_5 test_shared_ptr_132_text_archive: vc-6_5 vc-6_5-stlport test_shared_ptr_132_text_archive_dll: vc-6_5 test_shared_ptr_132_text_warchive: vc-6_5 vc-6_5-stlport test_shared_ptr_132_text_warchive_dll: intel-win32-8_1 vc-6_5 test_shared_ptr_132_xml_archive: vc-6_5 vc-6_5-stlport test_shared_ptr_132_xml_archive_dll: vc-6_5 test_shared_ptr_132_xml_warchive: vc-6_5 vc-6_5-stlport test_shared_ptr_132_xml_warchive_dll: vc-6_5 test_shared_ptr_binary_archive: borland-5_6_4 vc-6_5 test_shared_ptr_binary_archive_dll: borland-5_6_4 vc-6_5 test_shared_ptr_text_archive: borland-5_6_4 vc-6_5 test_shared_ptr_text_archive_dll: borland-5_6_4 vc-6_5 test_shared_ptr_text_warchive: borland-5_6_4 vc-6_5 test_shared_ptr_text_warchive_dll: borland-5_6_4 vc-6_5 test_shared_ptr_xml_archive: borland-5_6_4 vc-6_5 test_shared_ptr_xml_archive_dll: borland-5_6_4 vc-6_5 test_shared_ptr_xml_warchive: borland-5_6_4 vc-6_5 test_shared_ptr_xml_warchive_dll: borland-5_6_4 vc-6_5 |variant| variant_reference_test: mingw-3_4_2
Re the remaining serialization test failures. Incorporating Rene's fix and rolling back a change that looked innocuous pretty much addressed all the serialization problems that I think I can fix. There are a couple of remaining ones. MSVC 6.5 - text_exported was broken by fixes included to remove unnecessary header dependencies - it now provokes a compiler ICE. Borland Test_shared_ptr now fails. This is annoying because on my borland tests this passes. The message is Error E2451 C:\Users\Administrator\boost\main\boost\boost/mpl/aux_/integral_wrapper.hpp 45: Undefined symbol 'N' in function archive::detail::iserializer<archive::binary_iarchive,shared_ptr<A>
::tracking(const unsigned int) const Error E2230 C:\Users\Administrator\boost\main\boost\boost/mpl/aux_/integral_wrapper.hpp 45: In-line data member initialization requires an integral constant expression in function archive::detail::iserializer<archive::binary_iarchive,shared_ptr<A> ::tracking(const unsigned int) const
which suggests to me that this could be easily addressed by someone who has an interest and can reproduce the problem on his machine. I also notice that my borland 5.64 uses a different STL version than the one in the test matrix. So the test matrix fails on test_set while my own tests pass here. Other failures are not new and have failed to yield to efforts in the past. So we're down to a couple of regressions on less important platforms. If it's any consolation, there is a comparable number of new "passes". In anycase, I feel that that's the best I can do for now. Robert Ramey
Robert, I can feel the pain you're going through, but ... Robert Ramey wrote:
Re the remaining serialization test failures.
Other failures are not new and have failed to yield to efforts in the past.
So we're down to a couple of regressions on less important platforms.
I think known regressions in a minor release are absolutely unacceptable, even on less important platforms. The whole point of the minor release is to improve on what we got and not to break user code. Thomas -- Thomas Witt witt@acm.org
I think known regressions in a minor release are absolutely unacceptable, even on less important platforms. The whole point of the minor release is to improve on what we got and not to break user code.
These regressions are new ICEs that result from making the follwing changes: a) Remove header dependencies which many people complained about. b) Correct error which crashed some applications (though none of the tests) when serialization is used from dynamically loaded DLLS. Also times this error crashed applications with some compilers (also none of the tests) So if the indicated regressions are "absolutely unacceptable", then would leaving in the above bugs be acceptable? Problem is, you can't have it both ways with this particular compiler (msvc 6.5) Robert Ramey
Thomas, Robert Ramey wrote:
I think known regressions in a minor release are absolutely unacceptable, even on less important platforms. The whole point of the minor release is to improve on what we got and not to break user code.
These regressions are new ICEs that result from making the follwing changes:
a) Remove header dependencies which many people complained about.
In general I think that's something that should be delayed to the next major release. That being said I don't know the details and in the end it's Doug's call.
b) Correct error which crashed some applications (though none of the tests) when serialization is used from dynamically loaded DLLS. Also times this error crashed applications with some compilers (also none of the tests)
I can see this being important.
So if the indicated regressions are "absolutely unacceptable", then would leaving in the above bugs be acceptable?
AFAICS b is a bug and a is not.
Problem is, you can't have it both ways with this particular compiler (msvc 6.5)
Yes, it is difficult. Thomas -- Thomas Witt witt@acm.org
On Oct 13, 2005, at 6:04 PM, Thomas Witt wrote:
Thomas,
Robert Ramey wrote:
I think known regressions in a minor release are absolutely unacceptable, even on less important platforms. The whole point of the minor release is to improve on what we got and not to break user code.
These regressions are new ICEs that result from making the follwing changes:
a) Remove header dependencies which many people complained about.
In general I think that's something that should be delayed to the next major release. That being said I don't know the details and in the end it's Doug's call.
I think the removal of header dependencies is important enough to make the point release. It's hurting people now that upgraded to 1.33.0, and we should fix it as soon as possible... 1.34.0 is too far off to make them wait for this. Doug
Robert Ramey wrote:
Borland Test_shared_ptr now fails. This is annoying because on my borland tests this passes. The message is Error E2451 C:\Users\Administrator\boost\main\boost\boost/mpl/aux_/integral_wrapper.hpp 45: Undefined symbol 'N' in function archive::detail::iserializer<archive::binary_iarchive,shared_ptr<A>
::tracking(const unsigned int) const Error E2230 C:\Users\Administrator\boost\main\boost\boost/mpl/aux_/integral_wrapper.hpp 45: In-line data member initialization requires an integral constant expression in function archive::detail::iserializer<archive::binary_iarchive,shared_ptr<A> ::tracking(const unsigned int) const
which suggests to me that this could be easily addressed by someone who has an interest and can reproduce the problem on his machine.
Robert Ramey
[sniped part of the message above] Something very much like this error message shows up for me, I fixed it by changing in the following lines: typedef mpl::int_<1> type_; BOOST_STATIC_CONSTANT(unsigned int, value = type_::value); The word type to type_. I figured that the use of the word "type" as a macro parameter name in BOOST_STATIC_CONSTANT might be giving BCB trouble. And I was either right, or lucky enough that it worked around the problem. HTH Pablo
participants (6)
-
Doug Gregor -
Douglas Gregor
-
Douglas Gregor -
Pablo Aguilar -
Robert Ramey -
Thomas Witt