janko.marohnic / gmail.com wrote:
> the memory usage has now doubled to 100MB at the end of the
> program, indicating that some string bytes weren't
> successfully deallocated. So, it seems that String#replace has
> different behaviour compared to String#clear + String#<<.

Yes, this is an unfortunate side effect because of copy-on-write
semantics of String#replace.  If the arg (other_str) for
String#replace is non-frozen, a new frozen string is created with
using the existing malloc-ed pointer.  Both the receiver string
and other_str point to that new, shared string.

> I was *only* able to reproduce this with `IO.copy_stream`

I suspect part of this is because outbuf becomes a long-lived object
with IO.copy_stream (not sure), and the hidden frozen string
becomes long-lived, as well.

So yeah; a combination of well-intentioned optimizations hurt
when combined together.

The other part could be anything using IO#write could create
massive amounts of garbage before 2.5:
https://bugs.ruby-lang.org/issues/13085 And your copy_stream use
hits the IO#write case.  Unfortunately, the "fix" for [Bug
#13085] won't work effectively with shared strings, because we
can't free+recycle the a string which wasn't created internally
by the VM.

Unsubscribe: <mailto:ruby-core-request / ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>