Re: [tcpdump] About HAVE_NO_PRINTF_Z

[Denis Ovsienko via tcpdump-workers <tcpdump-workers () lists tcpdump org>]

> --- Begin Message ---
>
>
> From: Denis Ovsienko <denis () ovsienko info>
>
> Date: Sat, 14 Jan 2023 16:57:45 +0000
>
> On Thu, 12 Jan 2023 10:40:24 -0500
> Michael Richardson <mcr () sandelman ca> wrote:
>
> > Denis Ovsienko via tcpdump-workers
> > <tcpdump-workers () lists tcpdump org> wrote:
> >     > Thank you for the detailed analysis, Guy.  I did not realize
> >     > the support is in the libc space rather than C99 space.  Do you
> >     > think it would be appropriate to use [again] %lu instead of %zu
> >     > for size_t and %ld instead of %z for ssize_t on systems that
> >     > don't have %zu and %z?  
> >
> > We could... but...
> > It's a significant hassle to maintain this and validate this for
> > future patches as well... for a product that basically nobody out
> > there uses anymore.
>
> The tests below confirm it works almost as I thought: when %z is not
> available, using %l still prints the correct value, even when long is
> 64-bit and sizeof is 32-bit (with a compile-time warning, of course).
> In other words, this is the old behaviour from several years ago:
>
> -------------------------------------------------------------------
> int main (void)
> {
>         int i;
>
>         printf ("sizeof (size_t) == %u\n", sizeof (size_t));
>         printf ("sizeof (ssize_t) == %u\n", sizeof (ssize_t));
>         printf ("sizeof() as %%u: %u\n", sizeof (i));
>         printf ("sizeof() as %%d: %d\n", sizeof (i));
>         printf ("sizeof() as %%lu: %lu\n", sizeof (i));
>         printf ("sizeof() as %%ld: %ld\n", sizeof (i));
>         printf ("sizeof() as %%zu: %zu\n", sizeof (i));
>         printf ("sizeof() as %%zd: %zd\n", sizeof (i));
>         return 0;
> }
> -------------------------------------------------------------------
> # 32-bit ARM Linux
> sizeof (size_t) == 4
> sizeof (ssize_t) == 4
> sizeof() as %u: 4
> sizeof() as %d: 4
> sizeof() as %lu: 4
> sizeof() as %ld: 4
> sizeof() as %zu: 4
> sizeof() as %zd: 4
> -------------------------------------------------------------------
> # 64-bit ARM Linux
> sizeof (size_t) == 8
> sizeof (ssize_t) == 8
> sizeof() as %u: 4
> sizeof() as %d: 4
> sizeof() as %lu: 4
> sizeof() as %ld: 4
> sizeof() as %zu: 4
> sizeof() as %zd: 4
> -------------------------------------------------------------------
> # SPARC and x86 Solaris 9
> sizeof (size_t) == 4
> sizeof (ssize_t) == 4
> sizeof() as %u: 4
> sizeof() as %d: 4
> sizeof() as %lu: 4
> sizeof() as %ld: 4
> sizeof() as %zu: zu
> sizeof() as %zd: zd
> -------------------------------------------------------------------
>
> So it would be practicable to restore Solaris 9 support in the trailing
> edge department of tcpdump using compile-time-conditional length
> modifier.  If anybody volunteers to implement that, it would be wrong
> to get in the way, I guess.
>
> >     > Whatever is the eventual solution, in any case from today's
> >     > perspective it makes sense to stop masking these test failures.
> >     >  
> >
> > Agreed.
> > Do you mean: Maybe just let the tests fail on Solaris, and document
> > that this happens?
>
> As the first step in any case, yes.  And make the build fail early if
> the required length modifier is not supported.
>
> -- 
>     Denis Ovsienko
>
> --- End Message ---
_______________________________________________
tcpdump-workers mailing list
tcpdump-workers () lists tcpdump org
https://lists.sandelman.ca/mailman/listinfo/tcpdump-workers