[iostreams] Progress on regressions
The iostreams regression tests are looking a LOT better. Great! I see that everything is failing on gcc-2.95.3-linux. I'm going to mark this toolset as unusable: nobody expects iostreams-anything to work with GCC 2.95.x's non-templated iostreams, and the maintenance burden far outweighs any advantage. The bzip2, gzip, and zlib tests are failing on Windows. I can't tell if this is a configuration error on the testers' machines or if it's a build issue in the iostreams library. I'll take a shot at reproducing the problem this evening. Doug
Doug Gregor wrote:
The iostreams regression tests are looking a LOT better. Great!
I see that everything is failing on gcc-2.95.3-linux. I'm going to mark this toolset as unusable: nobody expects iostreams-anything to work with GCC 2.95.x's non-templated iostreams, and the maintenance burden far outweighs any advantage.
Actually, it took only a few hours several months ago to make the whole library work on gcc 2-95. The current problem are a result of two issues: - iterator_range doesn't work on gcc 2.95. I made it work on this platform some time ago, but subsequent changes have broken it again. I haven't got around to fixing it but it should be fairly trivial. - Apparently gcc 2.95 with old iostreams is confused by boost::iostreams::char_traits, which may need to be explicitly qualified. At any rate, I don't much care if gcc-2.95 is marked unusable, but infact it probably will be usable.
The bzip2, gzip, and zlib tests are failing on Windows. I can't tell if this is a configuration error on the testers' machines or if it's a build issue in the iostreams library. I'll take a shot at reproducing the problem this evening.
I've been delaying posting configuration information on the testing list, since I didn't want to have to correct it later. But since we've run out of time, I'll do it now. Best, Jonathan
FWIW, I get these two errors when building from CVS with VC 8, beta 2: boost\libs\iostreams\build\..\src\bzip2.cpp(14) : fatal error C1083: Cannot open include file: 'bzlib.h': No such file or directory boost\libs\iostreams\build\..\src\zlib.cpp(14) : fatal error C1083: Cannot open include file: 'zlib.h': No such file or directory (Building with "bjam -sTOOLS=vc-8_0".) Presumably iostreams will either have to provide those headers, or conditionally exclude any of its sources that reference them. Keith MacDonald "Doug Gregor" <dgregor@cs.indiana.edu> wrote in message news:15896a5acc600c7a284adba2c19db51c@cs.indiana.edu...
The iostreams regression tests are looking a LOT better. Great!
I see that everything is failing on gcc-2.95.3-linux. I'm going to mark this toolset as unusable: nobody expects iostreams-anything to work with GCC 2.95.x's non-templated iostreams, and the maintenance burden far outweighs any advantage.
The bzip2, gzip, and zlib tests are failing on Windows. I can't tell if this is a configuration error on the testers' machines or if it's a build issue in the iostreams library. I'll take a shot at reproducing the problem this evening.
Doug
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
Keith MacDonald wrote:
FWIW, I get these two errors when building from CVS with VC 8, beta 2:
boost\libs\iostreams\build\..\src\bzip2.cpp(14) : fatal error C1083: Cannot open include file: 'bzlib.h': No such file or directory boost\libs\iostreams\build\..\src\zlib.cpp(14) : fatal error C1083: Cannot open include file: 'zlib.h': No such file or directory
(Building with "bjam -sTOOLS=vc-8_0".) Presumably iostreams will either have to provide those headers, or conditionally exclude any of its sources that reference them.
Right. You can define NO_ZLIB, NO_BZIP2 or NO_COMPRESSION if you don't want to configure Iostreams to use these thrid party libraries. For more info see my message with subject line "[iostreams] ATTENTION all Windows testers" on the testing list. Jonathan
On 5/24/05, Jonathan Turkanis <technews@kangaroologic.com> wrote:
Keith MacDonald wrote:
(Building with "bjam -sTOOLS=vc-8_0".) Presumably iostreams will either have to provide those headers, or conditionally exclude any of its sources that reference them.
Right. You can define NO_ZLIB, NO_BZIP2 or NO_COMPRESSION if you don't want to configure Iostreams to use these thrid party libraries.
Might it make sense to have these values be the default, or to use some sort of auto-detection for the dependent libraries as Boost.Python does with Python and Boost.Regex does with ICU? -- Caleb Epstein caleb dot epstein at gmail dot com
Caleb Epstein wrote:
On 5/24/05, Jonathan Turkanis <technews@kangaroologic.com> wrote:
Keith MacDonald wrote:
(Building with "bjam -sTOOLS=vc-8_0".) Presumably iostreams will either have to provide those headers, or conditionally exclude any of its sources that reference them.
Right. You can define NO_ZLIB, NO_BZIP2 or NO_COMPRESSION if you don't want to configure Iostreams to use these thrid party libraries.
Might it make sense to have these values be the default, or to use some sort of auto-detection for the dependent libraries as Boost.Python does with Python and Boost.Regex does with ICU?
I'm going to make windows users explicitly request compression support, as suggested by Rene and Doug. Things seem to work pretty well on Linux/Unix. I'm sort of glad I didn't have time to add OpenSSL support in 1.33. ;-) Jonathan
Doug Gregor wrote:
The iostreams regression tests are looking a LOT better. Great!
I see that everything is failing on gcc-2.95.3-linux. I'm going to mark this toolset as unusable: nobody expects iostreams-anything to work with GCC 2.95.x's non-templated iostreams, and the maintenance burden far outweighs any advantage.
gcc-2.95 should work now. I've commented out your markup. The tests failed today because of a CRAY workaround with a space after a line-continuation character. If they don't pass tomorrow, I'll mark gcc-2.95 as unusable again. Jonathan
participants (4)
-
Caleb Epstein -
Doug Gregor -
Jonathan Turkanis -
Keith MacDonald