Re: [RFC 11/21] test: include headers directly

[Stephen Hemminger <stephen@networkplumber.org>]

On Sun, 2 Aug 2026 19:37:04 +0000
Marat Khalili <marat.khalili@huawei.com> wrote:

> > diff --git a/app/test/test_bpf.c b/app/test/test_bpf.c
> > index 3205afaa63..32389c803a 100644
> > --- a/app/test/test_bpf.c
> > +++ b/app/test/test_bpf.c
> > @@ -15,6 +15,9 @@
> >  #include <rte_random.h>
> >  #include <rte_byteorder.h>
> >  #include <rte_errno.h>
> > +#include <rte_ip.h>
> > +#include <rte_udp.h>
> > +#include <rte_tcp.h>
> > 
> >  #include "test.h"
> >   
> 
> This file has another group of includes guarded by RTE_LIB_BPF several lines
> below, with `#include <rte_ip.h>` already there. It also seems that the guard
> is redundant since `meson.build` already declares same dependency. We probably
> should avoid adding duplicates, and optionally also do a cleanup here.
> 
> With test_pipeline_common.h I could not find what uses the newly added header.
> Other changed files have guards against MSVC which are not redundant. If the
> new includes were added under this guard I could probably ack the change as
> harmless, but as is the opinion of other maintainers is needed. So maybe
> splitting the commit per-file would accelerate the process.
> 
> Obvious question is if we could have some general policy and tools.
> Explicit includes are good, but whack-a-mole approach can only go so far.
> 

Good point. There is a good existing tool 'iwyu' that does this.
Will use that.