[dpdk-dev] [PATCH] lpm6: make IPv6 addresses immutable

[Stephen Hemminger]

On Mon, 9 Mar 2020 18:02:21 +0000
"Medvedkin, Vladimir" <vladimir.medvedkin at intel.com> wrote:

> On 09/03/2020 15:52, Stephen Hemminger wrote:
> > On Mon, 9 Mar 2020 13:38:53 +0000
> > "Medvedkin, Vladimir" <vladimir.medvedkin at intel.com> wrote:
> >  
> >> Hi Andrze,
> >>
> >> Adding const qualifier for bulk lookup leads to compilation problems
> >> (see http://c-faq.com/ansi/constmismatch.html)
> >> Also I don't think it would be good for usability to make users
> >> explicitly cast to (const uint8_t **) when passing 'ips' argument. I'd
> >> suggest leaving lookup_bulk as is.  
> >
> > Please give a more concrete example. There is no need for explicit cast
> > in current C.  
> 
> Hi Stephen,
> 
> In theory, i agree - all modern versions of GCC don't give warnings for 
> this code. The problem is, we don't only support modern versions of GCC. 
> For a "more concrete example", see the following error in gcc-4.8 on 
> CentOS 7:
> 
> http://mails.dpdk.org/archives/test-report/2020-March/119455.html
> 
> I have reproduced this on my non-CentOS machine with gcc4.8, so it 
> appears to be gcc4.8 specific rather than CentOS-specific. That said, 
> the same email contains a compile check from RHEL7 with the same 
> compiler version, and it didn't trigger the warning, and also it seems 
> to be that only meson build has triggered the warning (while i have 
> reproduced it with make on my machine).
> 
> So, i think it is best to leave lookup_bulk as is, and add const 
> qualifiers in other places (for const uint8_t *).
> 

Agree, just drop it from the bulk functions.
The array of array semantics is confusing compiler (and users).