Pulling my last comment #4092 (comment) here:

My intention with the first sentence was to mandate the fallback to English, and would advocate heavily for upgrading "SHOULD" to "MUST" (I realize I used should in my original draft, apologies).

I think there are already too many optional bits to the language spec, and I think this one should be mandatory as it provides a consistent behavior across clients that do language (namely that they fallback to English).

@sbrl I edited two minor things. Other than that, the proposal is all good.

I was suggested to engage in this discussion. Since this PR is about the specific formulation of the specification I added my suggestion to the #4092 (comment).

This needs review by org owners, i.e. @agnivade @owenvoke @mebeim.

@MasterOdin I completely agree about the fallback there. I'll edit.

I haven't been following this discussion closely. I'll settle for whatever is decided here.

Updates complete!

Re-reading the env var paragraph, I don't think that explicitly requiring LANG to be set is needed here. The linked GNU gettext reference does not actually say that LANGUAGE is to be ignored if LANG is not set.

This paragraph is quite complex. I'll try to reword it when I have time to make it as simple and clear as possible so that you guys can tell me what you think, but in a nutshell I think we should just stick to the priority list of this documentation page, simply giving LANGUAGE priority over LANG. Ignoring LANGUAGE if LANG is not set is just a missed chance to follow the user's preference.

Re-reading the env var paragraph, I don't think that explicitly requiring LANG to be set is needed here. The linked GNU gettext reference does not actually say that LANGUAGE is to be ignored if LANG is not set.

As i read the Note:-Part LANG must be present to enable the LANGUAGE variable. I tested it with gettext --usage and found exactly this behaviour:

% echo $LANG                          
de_DE.UTF-8
% gettext --usage
gettext: Unbekannte Option »--usage«
»gettext --help« gibt weitere Informationen.
% LANGUAGE=it_IT:de_DE gettext --usage
gettext: Unbekannte Option »--usage«
Usare 'gettext --help' per maggiori informazioni.
% env -u LANG LANGUAGE=it_IT:de_DE gettext --usage
gettext: unrecognized option '--usage'
Try 'gettext --help' for more information.
% LANG=en gettext --version
gettext (GNU gettext-runtime) 0.20.2
Copyright (C) 1995-2020 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <https://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.
Written by Ulrich Drepper.

You may try this on your system.
Edit: Fail on my side, i didn't noticed, that gettext is missing a usage option, but help prints much more text and for demonstration purpose this should be enough.

@columbarius Ok I see. Then the bottom line is: respect $LANG when available, followed by the list in $LANGUAGE if present, otherwise use English. I still believe the current paragraph is quite complex so I'll reword it.

Thanks for spotting the typo @mebeim! I guess I'll hold off on merging until you've re-worded it?

@mebeim Yes, respect $LANG when available, followed by the list in $LANGUAGE which prioritizes the value in $LANG, which is the confusing part of this standard. If $LANG and $LANGUAGE is set, respect$LANGUAGE before $LANG, but if $LANG is not set, ignore $LANGUAGE.

My example with the gettext command, shows, that this is used for stdout, but stderr just uses $LANG and ignores $LANGUAGE, but I don't think, we have to use that behaviour.

Since virtually every team member agreed with this, I guess we can merge it now?

Sounds good to me @zdroid - go ahead :D