Re: [PATCH bpf-next v3 1/2] Add subdir support to Documentation makefile

[Donald Hunter <donald.hunter@gmail.com>]

Daniel Müller <deso@posteo.net> writes:

> On Fri, Sep 09, 2022 at 11:12:22AM +0100, Donald Hunter wrote:
>> Andrii Nakryiko <andrii.nakryiko@gmail.com> writes:
>> 
>> > On Tue, Sep 6, 2022 at 3:50 AM Donald Hunter <donald.hunter@gmail.com> wrote:
>> >>
>> >> Daniel Borkmann <daniel@iogearbox.net> writes:
>> >>
>> >> > On 8/29/22 11:14 AM, Donald Hunter wrote:
>> >> >> Run make in list of subdirs to build generated sources and migrate
>> >> >> userspace-api/media to use this instead of being a special case.
>> >> >> Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
>> >> >
>> >> > Jonathan, given this touches Documentation/Makefile, could you ACK if
>> >> > it looks good to you? Noticed both patches don't have doc: $subj prefix,
>> >> > but that's something we could fix up.
>> >> >
>> >> > Maybe one small request, would be nice to build Documentation/bpf/libbpf/
>> >> > also with every BPF CI run to avoid breakage of program_types.csv. Donald
>> >> > could you check if feasible? Follow-up might be ok too, but up to Andrii.
>> >>
>> >> Sure, I can look at what is needed for the BPF CI run.
>> >>
>> >
>> > Daniel (Mueller, not Borkmann), is this something that can be added to BPF CI?
>
> I think as long as all required packages are available in the CI distribution
> (which I believe is currently a Ubuntu image, but may in the future become Arch
> Linux) it should not be a problem to perform checking in CI. It seems as if
> generating the documentation may take a while, so we should likely try to have
> it run in a separate job. I can't tell what hidden dependencies there may be,
> though.
>
>> It looks to me like it can be added to BPF CI if we change docs/conf.py
>> to call a new make target in docs/sphinx/Makefile. Hopefully Daniel can
>> confirm whether this is the case.
>
> I am not familiar with the documentation generation, but my quick search seems
> to suggest that this is done by a 3rd party service and is decoupled from BPF
> CI. Specifically, what you suggest may be reflected in the generated docs at
> https://libbpf.readthedocs.io/, but I believe they are created from the libbpf
> GitHub repository, which is only infrequently synced from bpf tree sources. I
> didn't find any indication that CI triggers documentation creation, but it's
> possible I missed something.

Apologies, I was referring to the libbpf GitHub repository. Anything
committed to the kernel Documentation/bpf/libbpf risks breaking the
documentation generation in the libbpf GitHub repository which, as you
say, uses readthedocs.