Bug#864840: error: cannot overload functions distinguished by return type alone version graph

[Jorge Moraleda]

Hello Lumin,

Thank you for looking into it. Personally, I have no problem if you decide
not to fix this and I can see that patching the package instead of waiting
for upstream support could open a can of worms.

I have patched my own system with the fix I sent you and cuda is working
smoothly for me with the default gcc 6 tool chain after that. The fact that
I had a patch and it was minimal is the reason I sent the bug report: I
thought other people would benefit from the simplicity of being able to use
the default tool chain.

Also, even though cuda 8 does not officially support gcc 6, in practice
recent releases have been moving into compliance in preparation for the 8.5
release (in fact the two lines the patch comments out were introduced
explicitly to address an ideosyncracy of gcc 6.1 as the comments in that
file reveal).

Thank you again. Best regards,

Jorge

On 15 June 2017 at 20:03, Lumin <cdluminate at gmail.com> wrote:

> Control: tags -1 +wontfix
>
> Hi,
>
> > The header file math_functions.h is not compatible with gcc 6.3 shipped
> with stretch.
>
> Actually CUDA 8.0.44 does not support GCC-6 at all. Lookup this document
> for
> working compiler combinations:
>
>   /usr/share/doc/nvidia-cuda-toolkit/README.Debian
>
> I'd suggest that (1) compile the application with clang-3.8 (shipped
> by stretch).
> OR (2) temporarily add the sid/unstable apt source, pull GCC-5 then remove
> that apt source and compile with GCC-5. OR (3) Get GCC-5/clang-3.8 by other
> approaches.
>
> We can do nothing to change the fact, hence marking this bug as wontfix.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-nvidia-devel/attachments/20170616/37c866e5/attachment.html>