nvidia-cuda-toolkit 6.5 transition - pycuda bits

[Graham Inggs]

On 22 September 2015 at 02:36, Andreas Beckmann <anbe at debian.org> wrote:
> On 2015-09-21 11:41, Graham Inggs wrote:
>> On 21/09/2015 10:50, Graham Inggs wrote:
>>>
>>> Do we need to revert to using the gcc/g++ wrappers?
>>
>> This seems to do the trick.  I'll let the cuda-samples continue building
>> then commit my changes to SVN.
>
> So should we revert the compiler check back to strict <= 4.9, too ?

I don't think that is needed, and we'll probably need that line in its
current form in future versions.


I sent another two messages to the list, but found they were rejected,
I'll repeat below:


I committed r5525 and uploaded to Ubuntu, and I have already received
negative feedback. :(

I inadvertently had hardening-wrapper installed on the 3 different
machines I used for testing.
This caused symlinks for gcc-4.4, gcc-4.6, gcc-4.7, gcc-4.8 and
gcc-4.9, etc. all to exist, breaking Andreas' wrappers.
I thought I could rely on gcc-4.9.real, but this turned out to be
installed by hardening-wrapper as well.

I'll try to find a more robust solution.


The most reliable way I have found is to run 'g++-4.9 --version'.
i.e. debian/g++ would become:


#!/bin/sh

if g++-4.9 --version >/dev/null 2>&1; then
    prog=g++-4.9
else
    prog=g++-4.8
fi

exec $prog "$@"


Any other suggestions?