From eda9fe3d0b9538caa5a1fbf2e9a9e62c13585115 Mon Sep 17 00:00:00 2001
From: IG <omniacinis@gmail.com>
Date: Thu, 10 Sep 2026 04:21:40 +0400
Subject: [PATCH] Transcode non-ASCII-compatible strings before dumping

`visit_String` matches the string against US-ASCII regexp literals to
decide the scalar style. A string whose encoding is not ASCII-compatible
cannot be matched against an ASCII regexp, so `Psych.dump` raised
`Encoding::CompatibilityError` for any UTF-16/UTF-32 string, empty ones
included. The `binary?` guard above those checks only covers ASCII_8BIT,
so these encodings fell through to the regexp.

Transcode to UTF-8 up front when the encoding is not ASCII-compatible, so
every check below operates on a matchable string. Such strings now dump
exactly as their UTF-8 equivalents do, matching how the other non-UTF-8
encodings already behave, and matching `Psych.load`, which already
accepts UTF-16 input.
---
 lib/psych/visitors/yaml_tree.rb | 3 +++
 test/psych/test_encoding.rb     | 9 +++++++++
 2 files changed, 12 insertions(+)

diff --git a/lib/psych/visitors/yaml_tree.rb b/lib/psych/visitors/yaml_tree.rb
index 9c2c4aa4..42382770 100644
--- a/lib/psych/visitors/yaml_tree.rb
+++ b/lib/psych/visitors/yaml_tree.rb
@@ -293,6 +293,9 @@ def visit_String o
         style = Nodes::Scalar::PLAIN
         tag   = nil
 
+        # the checks below match ASCII regexps, which UTF-16/32 strings cannot match against
+        o = o.encode(Encoding::UTF_8) unless o.encoding.ascii_compatible?
+
         if binary?(o)
           o     = [o].pack('m0')
           tag   = '!binary' # FIXME: change to below when syck is removed
diff --git a/test/psych/test_encoding.rb b/test/psych/test_encoding.rb
index 0a31a680..9009fcf6 100644
--- a/test/psych/test_encoding.rb
+++ b/test/psych/test_encoding.rb
@@ -55,6 +55,15 @@ def test_transcode_utf16be
       assert_equal str, loaded
     end
 
+    def test_dump_non_ascii_compatible_encoding
+      ['UTF-16LE', 'UTF-16BE', 'UTF-32LE', 'UTF-32BE'].each do |encoding|
+        ['', 'hello', "multi\nline", 'こんにちは！', 'yes', '<<', '0123'].each do |str|
+          assert_equal Psych.dump(str), Psych.dump(str.encode(encoding)),
+            "#{str.inspect} in #{encoding} should dump as it does in UTF-8"
+        end
+      end
+    end
+
     def test_io_shiftjis
       Tempfile.create(['shiftjis', 'yml'], :encoding => 'SHIFT_JIS') {|t|
         t.write '--- こんにちは！'