Re: [RFC 0/2] Eliminate zero length arrays in DPDK

[Stephen Hemminger <stephen@networkplumber.org>]

On Thu, 24 Feb 2022 22:51:31 +0100
Morten Brørup <mb@smartsharesystems.com> wrote:

> > From: Tyler Retzlaff [mailto:roretzla@linux.microsoft.com]
> > Sent: Thursday, 17 February 2022 08.42
> > 
> > On Wed, Feb 16, 2022 at 10:10:01AM +0000, Bruce Richardson wrote:  
> > > On Wed, Feb 16, 2022 at 11:05:09AM +0100, Morten Brørup wrote:  
> > > > > From: Bruce Richardson [mailto:bruce.richardson@intel.com]
> > > > > Sent: Wednesday, 16 February 2022 10.33
> > > > >
> > > > > On Tue, Feb 15, 2022 at 03:00:56PM -0800, Stephen Hemminger  
> > wrote:  
> > > > > > Yet another case of applying Linux kernel best practices
> > > > > > to DPDK. Flexible arrays are supported by Clang, GCC and
> > > > > > Microsoft compilers (part of C99).
> > > > > >  
> > > > > Do we need to start explicitly stating that DPDK uses C99  
> > features, and  
> > > > > adding -std=c99 to our build flags? Are we also requiring that
> > > > > applications
> > > > > are compiled with c99 features to use this (I would hope that  
> > they are,  
> > > > > but
> > > > > I'm not sure we can mandate it).  
> > > >
> > > > No to -std=c99. It's >= C99 for applications; we should not prevent  
> > them from using a newer C standard.  
> > >
> > > Yes. For build flags, I was referring only to having it in the cflags  
> > for the  
> > > build of DPDK itself, not for apps. We definitely need to minimise  
> > the  
> > > build flags we expose to apps.
> > >  
> > > >
> > > > Adding a note about the C standard version to the DPDK requirements
> > > > documentation would be very nice. It only mentions a certain  
> > compiler  
> > > > version required. But I think that documenting the detailed build  
> > and  
> > > > runtime requirements (and why they are that way) is another task.
> > > >  
> > > Sure, we should do that. I am just wanting to be sure that if we  
> > specify a  
> > > minimum of C99, we won't get complaints back from those with legacy
> > > codebasees which only support C89/C90. I am therefore wondering if we  
> > need  
> > > to have our public headers C90-compliant?  
> > 
> > this seems to be the real question. what "minimum" C standard should be
> > documented as required to consume dpdk. we can obviously use any
> > standard
> > we wish to build/provide binaries. similarly we ought to document a
> > minimum C++ standard for consumption.
> > 
> > i would advocate for C99 however before setting that in stone it is
> > fair
> > to ask if there are any consumers using < C99.
> > 
> > we may also want to consider that the minimum required may differ
> > depending on the platform/port. though most differences in public
> > interface
> > i would hope could be trivially abstracted though.
> > 
> > ty  
> 
> Just read that the Linux kernel is moving towards C11, or at minimum C99, for version 5.18:
> https://lwn.net/SubscriberLink/885941/01fdc39df2ecc25f/
> 
> Let's be bold and push for the same for DPDK! :-)

Would be good, but still getting held back by legacy distros (RHEL)
and other compiler environments ICC, etc.