Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> writes:
> On Tue, Jan 14, 2020 at 11:07 AM Andrii Nakryiko
> <andrii.nakryiko@xxxxxxxxx> wrote:
>>
>> On Tue, Jan 14, 2020 at 8:43 AM Toke Høiland-Jørgensen <toke@xxxxxxxxxx> wrote:
>> >
>> > The change to use angled includes for bpf_helper_defs.h breaks compilation
>> > against libbpf when it is installed in the include path, since the file is
>> > installed in the bpf/ subdirectory of $INCLUDE_PATH. Fix this by adding the
>> > bpf/ prefix to the #include directive.
>> >
>> > Fixes: 6910d7d3867a ("selftests/bpf: Ensure bpf_helper_defs.h are taken from selftests dir")
>> > Signed-off-by: Toke Høiland-Jørgensen <toke@xxxxxxxxxx>
>> > ---
>> > Not actually sure this fix works for all the cases you originally tried to
>>
>> This does break selftests/bpf. Have you tried building selftests, does
>> it work for you? We need to fix selftests simultaneously with this
>> change.
>>
>> > fix with the referred commit; please check. Also, could we please stop breaking
>> > libbpf builds? :)
>>
>> Which libbpf build is failing right now? Both github and in-kernel
>> libbpf builds are fine. You must be referring to something else. What
>> exactly?
>
> I think it's better to just ensure that when compiling BPF programs,
> they have -I/usr/include/bpf specified, so that all BPF-side headers
> can be simply included as #include <bpf_helpers.h>, #include
> <bpf_tracing.h>, etc
And break all programs that don't have that already? Just to make the
kernel build env slightly more convenient? Hardly friendly to the
library users, is it? :)
As far as selftests are concerned, I finally managed to get an LLVM
version that will build them all; so I'll test that tomorrow and send a
v2 that doesn't break them...
-Toke
