This feature depends on bug #57594: the data needed to implement this will only be available after the hyphenation-pattern files are updated.

Since savannah lacks a Blocked state or other mechanism to show a dependency on another bug, I'm following Branden's lead in bug #59738 and marking this Postponed.

comment #1:

> See
>
> https://github.com/hyphenation/tex-hyphen/tree/master/hyph-utf8/tex/generic/hyph-utf8/patterns/tex
>
> for patterns with YAML metadata containing hyphenmin values.

Below I'm inlining what those look like.  Integrating YAML parsing into the troff executable seems like overkill.  We just need a little state machine to read and match the relevant chunk ("hyphenmins") of the comment header and complain if one isn't found.

Design question: since a minimum hyphenation mode can be computed based upon this characteristic of the pattern files, should reading such a pattern file effectively set .hy?

I'm thinking "probably not" and that such a thing can be rolled into bug #55070, if that ever happens, but it's worth considering.

% title: Hyphenation patterns for American English
% copyright: Copyright (C) 1990, 2004, 2005 Gerard D.C. Kuiken
% notice: This file is part of the hyph-utf8 package.
%     See http://www.hyphenation.org/tex for more information.
% language:
%     name: English, American spelling
%     tag: en-us
% version: 2005-05-30
% authors:
%   -
%     name: Gerard D.C. Kuiken
% licence:
%     text: >
%         Copying and distribution of this file, with or without modification,
%         are permitted in any medium without royalty provided the copyright
%         notice and this notice are preserved.
% hyphenmins:
%     typesetting:
%         left: 2
%         right: 3
% changes:
%     March 1, 1990 Initial release
%     May 30, 2005 Added copyright notice, no patterns change.
% texlive:
%     encoding: ascii
%     babelname: usenglishmax
%     legacy_patterns: ushyphmax.tex
%     message: Hyphenation patterns for American English
%     package: english
% known_bugs:
%     de-mo-c-rat: 'instead of dem-o-crat (see GitHub issue #15)'
% ==========================================
%
% ushyphmax.tex -- patterns for more hyphenation pattern memory (12000+).
% Also known as ushyphen.max.
%
% Needs extended pattern memory.
% Hyphenation trie becomes 7283 with 377 ops.
%
% These patterns are based on the Hyphenation Exception Log
% published in TUGboat, Volume 10 (1989), No. 3, pp. 337-341,
% and a large number of incorrectly hyphenated words not yet published.
% If added to Liang's before the closing bracket } of \patterns,
% the patterns run errorfree as far as known at this moment.
%
% These patterns find all admissible hyphens of the words in
% the Exception Log.  ushyph2.tex is a smaller set.
%
% Please send bugs or suggestions to tex-live (at) tug.org.
%
% 2005-05-30 (karl): in the past, ushyphmax.tex was a file containing
% only the additional patterns, without the \patterns command, etc.
% This turned out not to be very useful, since in practice the TeX
% distributions need one self-contained file for a language.  Therefore,
% ushyphmax.tex now contains both the additional patterns from
% Dr. Kuiken, and the original patterns and hyphenations from Knuth's
% hyphen.tex.
%
% The Plain TeX hyphenation tables.
\patterns{ % just type <return> if you're not using INITEX
.ach4
[...]

An additional observation about the posted patch: it has no effect when groff is run in compatibility mode.  I don't know why--I never use compatibility mode in real life, so I'm hardly an expert in debugging problems with it--but this isn't necessarily a showstopper, since the compatibility-mode behavior is the same with or without the patch.  So it's helping some users and not regressing the rest.

(See also bug #53413 comment #7.)

For most supported languages, groff's hyphenation data assumes a minimum number of letters that must appear at the beginning or end of a word before or after a hyphenation point.  The hyphenation algorithm attempting to leave fewer letters than this minimum will result in erroneous hyphenations.  (The groff manual's .hy section contains an example of an erroneous division of the word "splitting" resulting from a .hy value invalid for the English patterns.)

These required minimums vary depending on the language.  Currently, groff has a default .hy value that is used regardless of language, and which is invalid for two of its supported languages (French and English).  By default, groff should set these minimums based on the selected language.

The hyphenation data historically did not include these minimum values -- even though they are essential for using the patterns correctly.  Werner Lemberg points out (http://lists.gnu.org/archive/html/groff/2018-03/msg00002.html) that since the patterns originate in TeX and not groff, that's the proper place for this metadata to be introduced.  Later in the thread he adds, "The TeX hyphenation project is going to [explicitly provide lefthyphenmin and righthyphenmin values] by introducing YAML headers to all pattern files."  (A tex-hyphen thread about this can be found at http://tug.org/pipermail/tex-hyphen/2018-March/001650.html.)

In turn, groff should use these newly supplied values to (1) set .hy to a valid default for the chosen language, and (2) reject any .hy value incompatible with the language's hyphenation data.  (The user is free, of course, to request a value of .hy that specifies greater hyphenation restrictions than the minimums the language requires.)