Re: [PATCH 3/3] readutmp: load libsystemd with dlopen instead of linking

Luca Boccassi <[email protected]> Sun, 19 Jul 2026 18:26:21 +0100
Newsgroups gmane.comp.lib.gnulib.bugs
Message-ID <CAMw=ZnTXz6vYVjL5OW2J0+yRqYLgEhONuPb9B3Zf_G4Epw1fXA@mail.gmail.com>
On Sat, 18 Jul 2026 at 07:55, Paul Eggert <[email protected]> wrote:
>
> For this one, many of the same questions as for patch 2/3.
>
> Also, for coreutils, we'll want to stick with the current way of using
> readutmp, in which everything is dynamically linked before 'main' starts
> up. I realize this runs afoul of some of Debian's static checking for
> library usage, but we routinely run afoul of static checking all the
> time with GCC, and simply install exceptions when GCC's static checking
> issues a false positive. There needs to be something similar for the
> library usage static checkers; that is better than altering the source
> code to deal with a checker gone awry.

I'm not sure what static checking you are talking about? There's
static analyzer involved that I know of

> > +#include "gettext.h"
> > +#define _(msgid) dgettext (GNULIB_TEXT_DOMAIN, msgid)
>
> Let's not do that. Instead, the functions should fail and their callers
> should issue diagnostics. These functions are too low-level to be put in
> charge of user diagnostics.

Ok