Summary: | media-libs/opencv-2.3.0 fails to build with cuda useflag | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | optiluca <optiluca> |
Component: | Current packages | Assignee: | Andreas K. Hüttel <dilfridge> |
Status: | RESOLVED FIXED | ||
Severity: | normal | CC: | kde, navid.zamani |
Priority: | Normal | ||
Version: | unspecified | ||
Hardware: | All | ||
OS: | Linux | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Attachments: | build.log |
Description
optiluca@gmail.com
2011-09-20 17:13:45 UTC
Created attachment 287185 [details]
build.log
/opt/cuda/include/host_config.h:82:2: error: #error -- unsupported GNU version! gcc 4.5 and up are not supported! ...well, doesn't look libpng related in any reasonable way. I'd say INVALID. Disabling cuda use flag makes the ebuild compile successfully, as you noticed. Unrelated issue to the libpng bug then... (In reply to comment #2) > /opt/cuda/include/host_config.h:82:2: error: #error -- unsupported GNU version! > gcc 4.5 and up are not supported! > I'd say INVALID. I think you agree that if there is a USE flag, then enabling it must also be supported. Otherwise wouldn’t it be better to just remove the flag and functionality? But the normal Gentoo way certainly is to make functionality that the upstream package contains available on Gentoo too. So just shoving it away with a quick “INVALID” is a bit out of line, I guess. ;) Maybe I misunderstood… Same problem here. The problem actually lies with nvidia-cuda-toolkit, not opencv or the ebuild. Even the latest 4.0 is failing. So I'm not sure there's anything the gentoo devs can do about it until Nvidia finally release a newer version that works. This GCC is simply not supported upstream. To make matters worse, gcc-config downgrade does *NOT* work because symbol versioning will break a lot of executables, including cmake: cmake: /usr/lib/gcc/x86_64-pc-linux-gnu/4.4.6/libstdc++.so.6: version `GLIBCXX_3.4.14' not found (required by cmake) I actually managed to recompile cmake using the older gcc and then compile opencv and then go back to the new gcc. But turning off the cuda flag is probably the easiest way out for most people. (In reply to comment #5) > The problem actually lies with nvidia-cuda-toolkit, not > opencv or the ebuild. Even the latest 4.0 is failing. So I'm not sure there's > anything the gentoo devs can do about it until Nvidia finally release a newer > version that works. This GCC is simply not supported upstream. Agreed. If nVidia has a bug tracker, we should put it in there, link this bug to it, and put it in some “Waiting for upstream” mode. > To make matters worse, gcc-config downgrade does *NOT* work because symbol > versioning will break a lot of executables […]. Interesting. Thanks for that info. That’s bad news. > But turning off the cuda flag is > probably the easiest way out for most people. The problem is, that it will still cause the ebuild to fail for people. How about we force the flag to off for all systems where gcc ≥4.5 is installed? Or some kind of conditional masking? I have the additional problem, that any form of # echo 'dev-libs/opencv -cuda' >> /etc/portage/package.use/bugs seems to be ignored. (In reply to comment #6) > I have the additional problem, that any form of > # echo 'dev-libs/opencv -cuda' >> /etc/portage/package.use/bugs > seems to be ignored. media-libs/opencv, not dev-libs/opencv. The ebuild now prints an ewarn and disables cuda if cuda is requested and >=gcc-4.5 is used. |