Bug #10193: TestIO#test_readpartial_locktmp fails randomly

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: ruby 2.2.0dev (2014-08-26 trunk 47288) [x86_64-linux]
• Backport: 2.0.0: UNKNOWN, 2.1: UNKNOWN

[vo.x (Vit Ondruch)]

  4) Error:
TestIO#test_readpartial_locktmp:
IOError: stream closed
    /builddir/build/BUILD/ruby-2.2.0-r47288/test/ruby/test_io.rb:2861:in `readpartial'
    /builddir/build/BUILD/ruby-2.2.0-r47288/test/ruby/test_io.rb:2861:in `block (2 levels) in test_readpartial_locktmp'

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Closed
• % Done changed from 0 to 100

Applied in changeset r47353.

---

test_io.rb: ignore stream closed IOError

• test/ruby/test_io.rb (test_readpartial_locktmp): stream closed
  IOError while reading is not a matter. [Bug #10193]

[normalperson (Eric Wong)]

[e-mail] wrote:

Applied in changeset r47353.

---

test_io.rb: ignore stream closed IOError

• test/ruby/test_io.rb (test_readpartial_locktmp): stream closed
  IOError while reading is not a matter. [Bug #10193]

I get EBADF now:

TestIO#test_readpartial_locktmp:
Errno::EBADF: Bad file descriptor
/home/ew/ruby/test/ruby/test_io.rb:2870:in readpartial' /home/ew/ruby/test/ruby/test_io.rb:2870:in block (2 levels) in test_readpartial_locktmp'

Testing the following:

--- a/test/ruby/test_io.rb
+++ b/test/ruby/test_io.rb
@@ -2874,6 +2874,7 @@ End
assert_predicate(th, :alive?)
w.write(data)
Thread.pass while th.alive?

•  th.join
  

  end
  assert_equal(data, buf, bug6099)
  rescue RuntimeError # can't modify string; temporarily locked

I'm not sure if setting th->waiting_fd for rb_wait_for_single_fd
is a good idea. I think any poll/select wrappers should reload fileno
from IO objects on EBADF, because th->waiting_fd does not scale.

[normalperson (Eric Wong)]

Eric Wong [e-mail] wrote:

Testing the following:

Sorry, bad patch. The big rescue/ensure clauses was hiding bugs.

The problem is th.stop? happens twice:

1) release GVL for read() => EAGAIN
2) release GVL for rb_io_wait_readable()

We only accounted for one case in the previous code (and it is
unpredictable which).

I'm confident about the following patch:

--- a/test/ruby/test_io.rb
+++ b/test/ruby/test_io.rb
@@ -2866,24 +2866,29 @@ End
data = "a" * 100
th = nil
with_pipe do |r,w|

•  r.fcntl(Fcntl::F_SETFL, Fcntl::O_NONBLOCK)
  

•  th = Thread.new {r.readpartial(100, buf)}
  

•  Thread.pass until th.stop?
  

•  buf.replace("")
  

•  assert_empty(buf, bug6099)
  

•  assert_predicate(th, :alive?)
  

•  w.write(data)
  

•  Thread.pass while th.alive?
  

• end
• assert_equal(data, buf, bug6099)
• rescue RuntimeError # can't modify string; temporarily locked
• ensure
• if th
  begin

•    r.fcntl(Fcntl::F_SETFL, Fcntl::O_NONBLOCK)
  

•    th = Thread.new {r.readpartial(100, buf)}
  

•    Thread.pass until th.stop?
  

•    assert_equal 100, buf.bytesize
  

•    begin
  

•      buf.replace("")
  

•    rescue RuntimeError => e
  

•      assert_match(/can't modify string; temporarily locked/, e.message)
  

•      Thread.pass
  

•    end until buf.empty?
  

•    assert_empty(buf, bug6099)
  

•    assert_predicate(th, :alive?)
  

•    w.write(data)
  

•    Thread.pass while th.alive?
     th.join
  

•  rescue IOError
   end
  

  end

• assert_equal(data, buf, bug6099)
  end

def test_advise_pipe