Internals::SvREADONLY loses UTF-8 metadata for wide scalars #1970

Summary

Internals::SvREADONLY($scalar, 1) loses the scalar's UTF-8-on metadata for wide strings. A subsequent interpolation or string operation therefore treats the unchanged Unicode value as byte-backed. This breaks Readonly::Scalar users: the pure-Perl Readonly module marks values by calling Internals::SvREADONLY.

Minimal reproducer

my $wide = "\x{2122}";
Internals::SvREADONLY($wide, 1);
my $copy = "$wide";
printf "ord=%X utf8=%d\n", ord($copy), utf8::is_utf8($copy) ? 1 : 0;

System Perl 5.42.2 prints ord=2122 utf8=1. PerlOnJava prints ord=2122 utf8=0 on both the JVM and interpreter backends. The mismatch is already observable on $wide immediately after SvREADONLY, before interpolation.

CPAN evidence

• Distribution: Encode-Wide 0.07
• Run: 20261104-151051-20617
• System-Perl oracle: all 343 tests pass.
• PerlOnJava: t/integration.t fails the “No shared state: interleaved calls produce correct independent results” subtest; eight conversions for U+017E, U+017D, U+010D and U+2122 return the original wide character instead of an entity. The failure repeats after dependency retry.
• Environment: macOS 26.6 ARM64, OpenJDK 24.0.2, system Perl 5.42.2.

The test's calls are sequential; this is not a threading failure. The common cause is that read-only scalar state obscures the underlying string type/UTF-8 flag, which string interpolation and utf8::is_utf8 use to choose byte versus Unicode behavior.