[tannevaled]

Ruby 4.0's deprecated_rs_setter (io.c v4.0.9:8734) stops storing the String it was handed, unless that String is already frozen and bare:

rb_deprecated_str_setter(val, id, &val);       // TypeError, and the warning
if (!NIL_P(val)) {
    if (rb_str_equal(val, rb_default_rs)) val = rb_default_rs;
    else                                  val = rb_str_frozen_bare_string(val);
}

Four of the 27 examples from the 4.0 move (#799).

⛔ $\ and $, do NOT take this setter

Init_IO hooks them to rb_deprecated_str_setter, and only $/ and $-0 to deprecated_rs_setter:

rb_define_hooked_variable("$,",   …, rb_deprecated_str_setter);
rb_define_hooked_variable("$/",   …, deprecated_rs_setter);
rb_define_hooked_variable("$-0",  …, deprecated_rs_setter);
rb_define_hooked_variable("$\\",  …, rb_deprecated_str_setter);

rbgo handled all four through one checkStrGvar. The obvious edit — the one that does not read Init_IO — would have frozen four variables where MRI freezes two, silently. Both are now controls, and I checked them against a real MRI rather than against my reading of the C:

            MRI 4.0.7                     rbgo
ors   equal?=true frozen?=false      equal?=true frozen?=false
ofs   equal?=true frozen?=false      equal?=true frozen?=false
ret   true                           true
dash0 true                           true

⛔ The equality branch runs first

A frozen, plain "\n" satisfies "frozen and has no instance variables" word for word and is still replaced — by the canonical separator object. Every example in language/predefined_spec.rb uses "abc", which can only reach the other branch, so the suite passes either way and the identity is wrong.

Measured on ruby 4.0.7 before being written here, and proposed upstream as ruby/spec#1410 so the next implementation does not have to find it the same way.

BARE_STRING_P (string.c:372) is both halves — exact class String and no instance variables — which is exactly what makes the three existing examples distinguish each other.

A limit named rather than papered over

predefined_spec.rb's "makes a new frozen String from the assigned String" assigns a String subclass, and rbgo refuses that one step earlier: checkStrGvar decides with a Go type assertion where MRI tests T_STRING. So that example still fails, for a reason that has nothing to do with freezing.

Filed as #884 rather than fixed here — widening the assertion touches every "$x must be String" global, which is a different change and should not ride along inside this one.

Tests

Eleven cases, including the two $\/$, controls, the canonicalisation in both frozen and unfrozen form, $-0 sharing the slot, nil, the TypeError without #to_str, and the fact that assignment returns the value given rather than the value stored.

Refs #799, #884

🤖 Generated with Claude Code