Hi,
[Replying to multiple mails at once due to laziness.]
On Fri, Jun 17, 2016 at 06:09:20PM +0200, Daniel Borkmann wrote:
> Hmm, seems like a lot of stuff ...
It is. At some point I thought about maybe hack something in cocci
instead, but that would probably have taken longer given the code
diversity. :/
I know this is crap to review, but splitting it up into a 100 patches
doesn't make much sense, either.
> Please have a look at commit 8f80d450c3cb ("tc: fix compilation with old gcc
> (< 4.6)") ...
>
> Your changes effectively revert them again. Here, and some other parts of the
> bpf frontend
> code bits.
Oh, good catch! Thanks a lot for pointing this out. It definitely needs
to be sorted prior to applying my mess.
On Fri, Jun 17, 2016 at 11:57:53AM -0700, Stephen Hemminger wrote:
> It makes sense that if you can build a kernel with old toolchain, that
> iproute2 needs to be buildable as well.
>
> The current kernels are documented to require 3.2 or later.
So in your opinion we should stay compatible to gcc-3.2? Clarifying
requirements like this one would make sense in order to know what to
check against.
On Fri, Jun 17, 2016 at 02:47:51PM -0600, David Ahern wrote:
> On 6/17/16 2:36 PM, Daniel Borkmann wrote:
[...]
> > I just pointed to the fact that this would basically undo their changes
> > that they've submitted some time ago to the BPF frontend, reintroducing
> > the issue for them. Unfortunately, the anonymous struct cannot be named
> > due to uapi reasons. It should have been named from the very beginning,
> > but unfortunately too late now. So I would suggest to just leave those
> > affected parts as is.
>
> I was referring to Phil's patch. All of the struct {} req;
> initializations should be fine if you name the structs, but then need to
> run it through whatever compiler version the 6wind folks care about to
> verify that is true.
I'm not so sure about that. What I did regarding the anonymous struct
req is not new in iproute2 code base: There is the GENL_REQUEST macro
which expands to an identical construct and it's there since end of
2012.
Commit 8f80d450c3cb changes only the initializers of union bpf_attr, so
maybe the problem is limited to anonymous structs in unions? Anyway, I
guess defining which minimum gcc version to depend on and testing
against it is the only real solution here.
Thanks, Phil
