No rb_str_locktmp: IO::Buffer.for(str) { } does not lock str against modification #671

What

rbgo has no equivalent of MRI's rb_str_locktmp / rb_str_unlocktmp — a String flag, distinct from frozen, that makes every mutation raise

RuntimeError: can't modify string; temporarily locked

for as long as something holds the String's bytes.

IO::Buffer.for(string) { … } is the case that needs it. io_buffer_for_yield_instance (io_buffer.c) does

arguments->instance = io_buffer_for_make_instance(arguments->klass, arguments->string, arguments->flags);
rb_str_locktmp(arguments->string);
return rb_yield(arguments->instance);

and io_buffer_for_yield_instance_ensure unlocks it. The buffer aliases the String's bytes, so mutating the String while the block runs could reallocate them under the buffer — hence the lock rather than a copy.

s = +"test"
IO::Buffer.for(s) { s << "x" }
# MRI 4.0.5: RuntimeError: can't modify string; temporarily locked
# rbgo:      succeeds, s == "testx"   (the buffer now aliases freed bytes)

Measured cost

core/io/buffer/for_spec.rb    rbgo = 6    MRI 4.0.5 = 7

The one remaining example is "locks the original string to prevent modification". Every other file under core/io/buffer/ is at its ceiling after #668.

It is one example, but the flag is not IO::Buffer-specific: MRI uses it anywhere C code holds a String's buffer across a Ruby callback, so String#unpack with a block, IO#read into a buffer and the pack/unpack1 family would all want it.

Where

internal/object (a flag beside Frozen on object.String) plus every mutator that checks Frozen today — string_*.go. Both outside the file set of #668, which is why this is filed rather than fixed there. internal/vm/io_buffer.go then takes and releases the lock around the .for and .string block forms.