[PATCH v2 1/2] libdwfl: use XSI-compliant strerror_r.

[Dmitry V. Levin]

On Thu, Dec 17, 2020 at 01:17:40AM -0300, Érico Nogueira wrote:
> On Wed Dec 16, 2020 at 9:08 PM -03, Dmitry V. Levin wrote:
> > On Wed, Dec 16, 2020 at 07:30:11PM -0300, Érico Nogueira via
> > Elfutils-devel wrote:
> > > From: Érico Rolim <erico.erc@gmail.com>
> > > 
> > > The Linux man pages recommend this version of the function for portable
> > > applications.
> > > 
> > > Signed-off-by: Érico Rolim <erico.erc@gmail.com>
> >
> > I'd rather not do it this way because __xpg_strerror_r in glibc is a
> > wrapper around GNU strerror_r which does this almost always unneeded
> > string copying. Instead of penalizing GNU systems, I'd suggest checking
> > at configure time what kind of strerror_r do we have, and choosing the
> > code appropriately.
> 
> Fair enough. The GNU version of strerror_r does have a nicer API :)
[...]
> > What if we had something like this:
> >
> > static const char *
> > errnomsg(int error)
> > {
> > static const char unknown[] = "unknown error";
> >
> > #ifdef HAVE_STRERROR_R_POSIX_SIGNATURE
> > static __thread char msg[128];
> > return strerror_r (error, msg, sizeof (msg)) ? unknown : msg;
> > #else
> > return strerror_r (error, unknown, 0);
> > #endif
> > }
> 
> Would you prefer this go in lib/ as a "general usage" function or do I
> just leave it in libdwfl/ for now?

Is there any more potential users besides dwfl_errmsg?  If yes, then it
would make sense to have it in lib/, but then you'd have to come up with a
good name for the function. :)

> As a side note, the trick of passing a constant string and buflen=0 is
> quite neat, and isn't obvious from reading the man page :)

I'd suggest to make it explicitly documented in the manual page.


-- 
ldv