On Fri, Jan 15, 2021 at 02:55:41PM +0000, Bruce Richardson wrote:
> On Fri, Jan 15, 2021 at 03:09:25PM +0100, Thomas Monjalon wrote:
> > 15/01/2021 12:59, Bruce Richardson:
> > > On Fri, Jan 15, 2021 at 11:51:49AM +0000, Ferruh Yigit wrote:
> > > > On 1/15/2021 11:10 AM, Bruce Richardson wrote:
> > > > > To verify that all DPDK headers are ok for inclusion directly in a C
> > > > > file, and are not missing any other pre-requisite headers, we can
> > > > > auto-generate for each header an empty C file that includes that
> > > > > header.
> > > > > Compiling these files will throw errors if any header has unmet
> > > > > dependencies.
> > > > >
> > > > > The list of headers to check is based of the "headers" value returned
> > > > > from
> > > > > each library's meson.build file. However, since not all headers are
> > > > > for
> > > > > direct inclusion, add a build variable "headers_no_chkincs" to list
> > > > > those
> > > > > headers and skip checking them.
> > > > >
> > > > > Signed-off-by: Bruce Richardson <bruce.richard...@intel.com>
> > > > > ---
> > > > >
> > > > > v2:
> > > > > * add maintainers entry
> > > > > * distribute exception list among meson.build files.
> > > > >
> > > > > MAINTAINERS | 4 ++++
> > > > > app/chkincs/gen_c_file_for_header.py | 12 ++++++++++
> > > > > app/chkincs/main.c | 4 ++++
> > > > > app/chkincs/meson.build | 28
> > > > > ++++++++++++++++++++++++
> > > >
> > > > +1 to have this kind of tool to check, but it is not an application like
> > > > others in the 'app' folder, what do you think placing it under
> > > > 'devtools' or
> > > > 'buildtools'?
> > >
> > > Couple of reasons why it's placed in app.
> > >
> > > 1. We previously had a "chkincs" app in DPDK which was kept in the app
> > > folder
> > > 2. It allows us to reuse the build infrastructure for building apps,
> > > rather
> > > than reduplicating it.
> > > 3. We don't have any compilable code currently in the devtools folder, and
> > > even in buildtools the pmdinfogen app is going to go away.
> > >
> > > That being said, none of those reasons are major issues that can't be
> > > worked around if the consensus is to move it.
> >
> > It could be easily in devtools if it was a script.
> > By the way, we already have devtools/check-includes.sh
> > If your solution is better, please remove this script.
> >
> I only discovered the script existed when doing the v2 of this patchset,
> since it showed up in some grep calls I did for exception cases. I'm not
> sure that either approach is necessarily better, it's just right now that
> the script is unused (and also unknown) which is why I did this cleanup
> work.
>
> Here is how I see the current comparison between two approaches:
> * Script as advantage in that it performs C++ checks as well as C
> * Script also allows passing arbitrary additional C flags into checks for
> higher levels of compliance, but I'm not sure this is something I like as
> I'd rather have standardisation here across all headers than have some
> headers more pedantic-friendly than others.
> * Main downside of the script is that is works off directories rather than
> a list of files, which means it requires maintenance of the exception
> list in the script, rather than in the build definition files where we call
> out the headers to be installed
>
> I'm honestly fine either way on this (as with directory where
> implementation lives) - main thing is to have the checking done, rather
> than ignored.
>
And I (obviously) forgot to mention that the existing script is not currently
integrated into existing build or build-test scripts. I haven't looked into
how complex this would be, but it would require investigation time.
