[dpdk-dev] [PATCH] net/mlx5: fix compilation issue with gcc pragma

[Stephen Hemminger]

On Tue, 1 Oct 2019 17:15:46 +0000
Slava Ovsiienko <viacheslavo at mellanox.com> wrote:

> > -----Original Message-----
> > From: Stephen Hemminger <stephen at networkplumber.org>
> > Sent: Tuesday, October 1, 2019 17:54
> > To: Slava Ovsiienko <viacheslavo at mellanox.com>
> > Cc: dev at dpdk.org; Matan Azrad <matan at mellanox.com>; Raslan
> > Darawsheh <rasland at mellanox.com>; ferruh.yigit at intel.com
> > Subject: Re: [dpdk-dev] [PATCH] net/mlx5: fix compilation issue with gcc
> > pragma
> > 
> > On Tue,  1 Oct 2019 11:10:23 +0000
> > Viacheslav Ovsiienko <viacheslavo at mellanox.com> wrote:
> >   
> > > +#if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION >= 40600)  
> > #pragma GCC  
> > > +diagnostic push
> > >  #pragma GCC diagnostic ignored "-Wformat-nonliteral"
> > > +#endif
> > > +	/* Use safe format to check maximal buffer length. */
> > >  	while (fscanf(file, format, ifname) == 1) { -#pragma GCC diagnostic
> > > error "-Wformat-nonliteral"
> > > +#if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION >= 40600)  
> > #pragma GCC  
> > > +diagnostic pop #endif  
> > 
> > This is messy, is there not a better way to do this?  
> 
> At least I did not find one.
> 
> The GCC compile-time format checking feature is nice in general and it worth 
> to be engaged. The legitimate fscanf() usage with variable format parameter
> causes GCC to emit error/warning, so we should suppress these ones for this
> single line. ICC does not emit warning and does not recognize GCC pragmas.
> Clang just does not recognize fscanf().
> 
> Should we use "#ifndef __INTEL_COMPILER" (typical workaround for
> GCC diagnostic pragma in DPDK)? I'm not sure, It is not completely correct.
> 
> The alternative I see is to implement dedicated routine to read words from the file,
> but it means more code and more run-time resources. It seems not to be 
> the right way to push compile-time issues resolving to the run-time.
> 
> Defining the macro is not relevant here because this is a single case.
> 
> WBR, Slava
> 
> 

You are going to a lot of effort to solve a problem of interface name length
which can not happen.  The maximum interface name in linux and bsd is always
15 characters plus null. Therefore there is no need to build a dynamic
format string at all here. Or you could use the assignment allocation modifier
so that the resulting string from fscanf was allocated.

Could you try one of these instead.