Re: [PATCH] tools/resolve_btfids: Include linux/types.h

[Khazhy Kumykov]

On Thu, Mar 21, 2024 at 12:51 PM Dmitrii Bundin
<dmitrii.bundin.a@xxxxxxxxx> wrote:
>
> On Mon, Mar 18, 2024 at 7:56 PM Khazhy Kumykov <khazhy@xxxxxxxxxxxx> wrote:
> > I'm also seeing this, on clang.
>
> I think the error is more related to the libc version. I updated the
> libc6 to 2.35 and noticed that the <linux/types.h> header is included
> indirectly through <sys/stat.h>. The relevant sample of the include
> hierarchy for <sys/stat.h> with libc6 v2.35 looks as follows:
>
> . /usr/include/x86_64-linux-gnu/sys/stat.h
> .. /usr/include/x86_64-linux-gnu/bits/stat.h
> ... /usr/include/x86_64-linux-gnu/bits/struct_stat.h
> .. /usr/include/x86_64-linux-gnu/bits/statx.h
> ... /linux/tools/include/uapi/linux/stat.h
> .... /linux/tools/include/linux/types.h
>
> The <linux/types.h> is included on the latest line of the sample.
> Starting the version 2.28 there's an inclusion of <bits/statx.h> which
> was not presented in 2.27.
>
> When building the resolve_btfids with the libc6 version 2.27
> <linux/types.h> is not included through <sys/stat.h>. The include
> hierarchy for <sys/stat.h> with libc6 v2.27 looks as follows:
>
> . /usr/include/x86_64-linux-gnu/sys/stat.h
> .. /usr/include/x86_64-linux-gnu/bits/stat.h
> . /usr/include/fcntl.h
>
> To avoid being dependent on the inclusion of <linux/types.h> at some
> other place it looks reasonable to include it explicitly to bring all
> the necessary declarations before their usage.
I would agree - include what you use. The u32 type is defined in
linux/types.h, relying on indirect includes seems fragile (and in this
case, does seem to break folks).
>
> What do you think?