naruse / airemix.jp wrote:
> A patch proposed Eric in [ruby-core:35610] is not merged yet.

Won't apply after r43373 (which is way more important).

I don't think the current close(tmpfd) needs to release GVL in the
new code path it is dropping a refcount after rb_cloexec_dup2;
so nothing expensive going on in the kernel.

close performance on inotify descriptors (my main reason for this) is
much improved since 2011, too.  AFAIK the heavy lifting was moved to
asynchronous kernel workqueues.