Re: [PATCH v2] man/man7/environ.7: Fix underspecification of "name=value" strings

"G. Branden Robinson" <[email protected]> Tue, 7 Jul 2026 08:39:23 -0500

Newsgroups	org.kernel.vger.linux-man
Message-ID	<20260707133923.lu46dj4p5dhew65v@illithid>

Hi Alex,

At 2026-07-07T13:31:31+0200, Alejandro Colomar wrote:
> I think these programs don't prove what happens on a system with an
> encoding incompatible with ASCII.  I think if you compile glibc on an
> EBDIC-based system (if that is possible at all, which I ignore),
> you'll get the behavior based on '=' and not 0x3D.  I think this is
> required by POSIX.  0x3D is not special.
> 
> Of course, if your system is a mix of ASCII and EBDIC, you have a
> problem: glibc might have been compiled as ASCII (and thus use 0x3D)
> and your program might be using something else.

A few years ago, Mike Fulton of IBM explained to the groff list how
modern z/OS copes with mixed EBCDIC/UTF-8 environments.  Essentially
they use an extended file system attribute that tags each file such that
system libraries know which encoding to use.

https://lists.gnu.org/archive/html/groff/2023-04/msg00003.html

His explanation gave me the courage to delete GNU troff's internal
support for CCSID ("code page") 1047 in time for the groff 1.24.0
release.

To date, no one has yet complained about that.

At the risk of offending someone, I venture that no OS vendor but IBM
cares about EBCDIC encodings, and no application developer who targets
any OS other than z/OS cares about it either.

Regards,
Branden

signature.asc (application/pgp-signature, 833 B)