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

Paul Eggert <[email protected]> Mon, 13 Jul 2026 12:42:35 -0600
Newsgroups gmane.comp.lib.gnulib.bugs
Message-ID <[email protected]>
On 7/13/26 03:20, Luca Boccassi wrote:
> coreutils_already_ uses dlopen (correctly!) to optionally load
> openssl for an optional feature in sort:

Sure, but that's a different case. It is designed for improving 
performance in the normal case and for causing link failure to occur 
only for unusual cases, whereas the proposed change is for a much bigger 
deal: it's for default functionality. Default functionality is why 
programs ordinarily link dynamically before 'main' starts up.


> It just crashes on startup, and you
> can't even run --help or so, which is very much not what we want.

So the main point of the proposed change is to improve diagnostics, 
right? If so, how about this idea instead? Let's improve the diagnostics 
of the dynamic linker. That would improve this issue for all programs, 
not just for 'who'. And it would mean we needn't worry about this sort 
of thing for coreutils, either for libsystemd or for other dynamically 
linked libraries that are essential for default operation.