[ruby-core:88382] [Ruby trunk Bug#14968] [PATCH] io.c: make all pipes nonblocking by default

Issue #14968 has been updated by usa (Usaku NAKAMURA).


> Uh oh, really? I saw rb_w32_set_nonblock() has is_pipe() check
> and {Get,Set}NamedPipeHandleState calls, already, so that's
> something else?

1. A pipe may be a named pipe, or an anonymous pipe :)
2. The code is not tested.
3. Microsoft says that "Note that nonblocking mode ... should not be used to achieve asynchronous input and output (I/O) with named pipes."

> Does win32 have TOCTTOU race in signal handling?

Maybe.  But I should note that Windows does not have real signals :P

----------------------------------------
Bug #14968: [PATCH] io.c: make all pipes nonblocking by default
https://bugs.ruby-lang.org/issues/14968#change-73428

* Author: normalperson (Eric Wong)
* Status: Open
* Priority: Normal
* Assignee: 
* Target version: 
* ruby -v: 
* Backport: 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: UNKNOWN
----------------------------------------
Crap, I always planned to have something like this for [Feature #13618]
(auto-Fiber); but introducing a race condition for Timer-thread
elimination in [Misc #14937] forced me to introduce this early.

Anyways, I might have to revert and reintroduce timer-thread if this
change is unnacceptable :<  (I HATE timer thread)

```
io.c: make all pipes nonblocking by default

All normal Ruby IO methods (IO#read, IO#gets, IO#write, ...) are
all capable of appearing to be "blocking" when presented with a
file description with the O_NONBLOCK flag set; so there is
little risk of incompatibility within Ruby-using programs.

The biggest compatibility risk is when spawning external
programs.  As a result, stdin, stdout, and stderr are now always
made blocking before exec-family calls.

Timer-thread elimination in https://bugs.ruby-lang.org/issues/14937
introduced a race condition in signal handling.  It is possible
to receive a signal inside BLOCKING_REGION right before read/write
syscalls.  If this patch cannot be accepted, I will have to revert
to reintroduce timer-thread and increase resource use (which
led to other failures in the past).  The race condition
introduced for [Misc #14937] led to rare CI failures on a few
tests:

- test/ruby/test_thread.rb (test_thread_timer_and_interrupt):
  http://www.rubyist.net/~akr/chkbuild/debian/ruby-trunk/log/20180805T080500Z.fail.html.gz

- test/ruby/test_io.rb (test_race_gets_and_close):
  http://ci.rvm.jp/results/trunk@P895/1190369

This change is ALSO necessary to take advantage of (proposed
lightweight concurrency (aka "auto-Fiber") or any similar
proposal: https://bugs.ruby-lang.org/issues/13618

TODO: all sockets and FIFOs non-blocking by default, too
```


---Files--------------------------------
0001-io.c-make-all-pipes-nonblocking-by-default.patch (12.1 KB)


-- 
https://bugs.ruby-lang.org/

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