[PATCH 1/3] Fix call to DL::Function.new in tests

[PATCH 1/3] Fix call to DL::Function.new in tests

[Hleb Valoshka]

The 3rd argument of DL::Function.new is ABI type, not a return type.
---
 test/test_posix_mq.rb |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/test/test_posix_mq.rb b/test/test_posix_mq.rb
index d48238c..5c7cc11 100644
--- a/test/test_posix_mq.rb
+++ b/test/test_posix_mq.rb
@@ -110,7 +110,7 @@ class Test_POSIX_MQ < Test::Unit::TestCase
       if defined?(DL::Function)
         alarm = libc["alarm"]
         alarm = DL::CFunc.new(alarm, DL::TYPE_INT, "alarm")
-        alarm = DL::Function.new(alarm, [DL::TYPE_INT], DL::TYPE_INT)
+        alarm = DL::Function.new(alarm, [DL::TYPE_INT])
       else
         alarm = libc["alarm", "II"]
       end
-- 
1.7.10

[PATCH 2/3] Add libc names for Debian GNU/kFreeBSD and FreeBSD in tests.

[Hleb Valoshka]

---
 test/test_posix_mq.rb |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/test/test_posix_mq.rb b/test/test_posix_mq.rb
index 5c7cc11..5b78c22 100644
--- a/test/test_posix_mq.rb
+++ b/test/test_posix_mq.rb
@@ -104,7 +104,7 @@ class Test_POSIX_MQ < Test::Unit::TestCase
 
   def test_alarm_signal_safe
     libc = alarm = nil
-    libcs = %w(libc.so.6 /usr/lib/libc.sl)
+    libcs = %w(libc.so.6 libc.so.0.1 libc.so.7 /usr/lib/libc.sl)
     libcs.each do |name|
       libc = DL::Handle.new(name) rescue next
       if defined?(DL::Function)
-- 
1.7.10

[PATCH 3/3] Add a sleep in alarm test in current Ruby.

[Hleb Valoshka]

Without that sleep alarms will be 0. Looks like Ruby VM issue.
---
 test/test_posix_mq.rb |    1 +
 1 file changed, 1 insertion(+)

diff --git a/test/test_posix_mq.rb b/test/test_posix_mq.rb
index 5b78c22..98feff2 100644
--- a/test/test_posix_mq.rb
+++ b/test/test_posix_mq.rb
@@ -127,6 +127,7 @@ class Test_POSIX_MQ < Test::Unit::TestCase
     a = nil
     assert_raises(Errno::EINTR) { a = @mq.receive }
     elapsed = Time.now - t0
+    sleep 0.01 if RUBY_VERSION >= '1.9'
     assert_nil a
     assert elapsed >= interval, elapsed.inspect
     assert elapsed < 1.10, elapsed.inspect
-- 
1.7.10

Re: [PATCH 3/3] Add a sleep in alarm test in current Ruby.

[Eric Wong]

Hleb Valoshka <375gnu@gmail.com> wrote:
> Without that sleep alarms will be 0. Looks like Ruby VM issue.

Thanks, your other two patches look good and I've pushed them.

Can you elaborate on this issue?

I would like to see a bug report/analysis to Ruby upstream
(and or EGLIBC) maintainers.
I assume this only happens on GNU/kFreeBSD?

> --- a/test/test_posix_mq.rb
> +++ b/test/test_posix_mq.rb
> @@ -127,6 +127,7 @@ class Test_POSIX_MQ < Test::Unit::TestCase
>      a = nil
>      assert_raises(Errno::EINTR) { a = @mq.receive }
>      elapsed = Time.now - t0
> +    sleep 0.01 if RUBY_VERSION >= '1.9'
>      assert_nil a
>      assert elapsed >= interval, elapsed.inspect
>      assert elapsed < 1.10, elapsed.inspect

Fwiw, I'm not sure if this is a good test, either.  Since I
wrote this code, I've come to believe normal API methods should
not raise Errno::EINTR.  I've changed a few of my other libraries
(sleepy_penguin, kgio) to avoid EINTR but missed this one.

Re: [PATCH 3/3] Add a sleep in alarm test in current Ruby.

[Eric Wong]

Eric Wong <normalperson@yhbt.net> wrote:
> Fwiw, I'm not sure if this is a good test, either.  Since I
> wrote this code, I've come to believe normal API methods should
> not raise Errno::EINTR.  I've changed a few of my other libraries
> (sleepy_penguin, kgio) to avoid EINTR but missed this one.

Something like this

>From 535c9541fdba10a9572b63efc0d0502d009da94c Mon Sep 17 00:00:00 2001
From: Eric Wong <normalperson@yhbt.net>
Date: Tue, 3 Jul 2012 17:57:44 -0700
Subject: [PATCH] change blocking functions to not return EINTR

This is an API change.
---
 ext/posix_mq/posix_mq.c | 12 +++++++++++-
 test/test_posix_mq.rb   | 10 ++++++----
 2 files changed, 17 insertions(+), 5 deletions(-)

diff --git a/ext/posix_mq/posix_mq.c b/ext/posix_mq/posix_mq.c
index 2958352..df667e3 100644
--- a/ext/posix_mq/posix_mq.c
+++ b/ext/posix_mq/posix_mq.c
@@ -542,8 +542,11 @@ static VALUE _send(int sflags, int argc, VALUE *argv, VALUE self)
 	x.timeout = convert_timeout(&expire, timeout);
 	x.msg_prio = NIL_P(prio) ? 0 : NUM2UINT(prio);
 
+retry:
 	rv = (mqd_t)rb_thread_blocking_region(xsend, &x, RUBY_UBF_IO, 0);
 	if (rv == MQD_INVALID) {
+		if (errno == EINTR)
+			goto retry;
 		if (errno == EAGAIN && (sflags & PMQ_TRY))
 			return Qfalse;
 		rb_sys_fail("mq_send");
@@ -574,9 +577,13 @@ static VALUE send0(VALUE self, VALUE buffer)
 	x.timeout = NULL;
 	x.msg_prio = 0;
 
+retry:
 	rv = (mqd_t)rb_thread_blocking_region(xsend, &x, RUBY_UBF_IO, 0);
-	if (rv == MQD_INVALID)
+	if (rv == MQD_INVALID) {
+		if (errno == EINTR)
+			goto retry;
 		rb_sys_fail("mq_send");
+	}
 
 	return self;
 }
@@ -678,8 +685,11 @@ static VALUE _receive(int rflags, int argc, VALUE *argv, VALUE self)
 	x.msg_len = (size_t)mq->attr.mq_msgsize;
 	x.des = mq->des;
 
+retry:
 	r = (ssize_t)rb_thread_blocking_region(xrecv, &x, RUBY_UBF_IO, 0);
 	if (r < 0) {
+		if (errno == EINTR)
+			goto retry;
 		if (errno == EAGAIN && (rflags & PMQ_TRY))
 			return Qnil;
 		rb_sys_fail("mq_receive");
diff --git a/test/test_posix_mq.rb b/test/test_posix_mq.rb
index 5b78c22..8d6b014 100644
--- a/test/test_posix_mq.rb
+++ b/test/test_posix_mq.rb
@@ -118,16 +118,18 @@ class Test_POSIX_MQ < Test::Unit::TestCase
     end
     alarm or return warn "alarm() not found in #{libcs.inspect}"
     alarms = 0
-    trap("ALRM") { alarms += 1 }
+    trap("ALRM") do
+      alarms += 1
+      Thread.new { @mq.send("HI") }
+    end
     interval = 1
     alarm.call interval
     @mq = POSIX_MQ.new(@path, :rw)
     assert ! @mq.nonblock?
     t0 = Time.now
-    a = nil
-    assert_raises(Errno::EINTR) { a = @mq.receive }
+    a = @mq.receive
     elapsed = Time.now - t0
-    assert_nil a
+    assert_equal(["HI", 0], a)
     assert elapsed >= interval, elapsed.inspect
     assert elapsed < 1.10, elapsed.inspect
     assert_equal 1, alarms

Re: [PATCH 3/3] Add a sleep in alarm test in current Ruby.

[Hleb Valoshka]

[-- Attachment #1: Type: text/plain, Size: 489 bytes --]

On 7/4/12, Eric Wong <normalperson@yhbt.net> wrote:
> Can you elaborate on this issue?

I'll try, but I don't promise a result, cause I don't have enough knowledge :)

> I assume this only happens on GNU/kFreeBSD?

No, It's on GNU/Linux. On GNU/kFreeBSD I have "Errno::ENOSYS: Function
not implemented - mq_open" (yes, I've made "kldload mqueuefs).

On ordinal FreeBSD 9.0 this assert is successful but there other
problems. You may look attachments, out.1 is stdout and out.2 is
stderr.

Re: [PATCH 3/3] Add a sleep in alarm test in current Ruby.

[Hleb Valoshka]

This issue can be reproduced with 3 simple LOC

mkfifo myfifo
ruby -e "a=0;trap(:USR2) {a+=1}; IO.readlines('myfifo') rescue nil; puts a"
killall -USR2 ruby

And it affects only GNU/Linux and OpenBSD but not FreeBSD or GNU/kFreeBSD.

I've filed an issue: https://bugs.ruby-lang.org/issues/6725

Re: [PATCH 3/3] Add a sleep in alarm test in current Ruby.

[Eric Wong]

Hleb Valoshka <375gnu@gmail.com> wrote:
> This issue can be reproduced with 3 simple LOC
> 
> mkfifo myfifo
> ruby -e "a=0;trap(:USR2) {a+=1}; IO.readlines('myfifo') rescue nil; puts a"
> killall -USR2 ruby
> 
> And it affects only GNU/Linux and OpenBSD but not FreeBSD or GNU/kFreeBSD.
> 
> I've filed an issue: https://bugs.ruby-lang.org/issues/6725

Odd, thanks!

Btw, I've also pushed out the Errno::EINTR-avoidance patch.

I'm still hoping to have more time to look at the FreeBSD9 issues
you've reported.

Re: [PATCH 3/3] Add a sleep in alarm test in current Ruby.

[Hleb Valoshka]

On 7/12/12, Eric Wong <normalperson@yhbt.net> wrote:

>> I've filed an issue: https://bugs.ruby-lang.org/issues/6725

Bug report is rejected with the following explanation: "Asynchronous signals may
not dealt with in the order you expect."

Re: [PATCH 3/3] Add a sleep in alarm test in current Ruby.

[Eric Wong]

Hleb Valoshka <375gnu@gmail.com> wrote:
> On 7/12/12, Eric Wong <normalperson@yhbt.net> wrote:
> 
> >> I've filed an issue: https://bugs.ruby-lang.org/issues/6725
> 
> Bug report is rejected with the following explanation: "Asynchronous signals may
> not dealt with in the order you expect."

True, I don't disagree with upstream... Signal handlers are tricky :<

Re: [PATCH 3/3] Add a sleep in alarm test in current Ruby.

[Eric Wong]

Hleb Valoshka <375gnu@gmail.com> wrote:
> No, It's on GNU/Linux. On GNU/kFreeBSD I have "Errno::ENOSYS: Function
> not implemented - mq_open" (yes, I've made "kldload mqueuefs).

OK, I've reproduced this, too.  Could be an EGLIBC issue...

Re: [PATCH 3/3] Add a sleep in alarm test in current Ruby.

[Hleb Valoshka]

On 7/13/12, Eric Wong <normalperson@yhbt.net> wrote:
> Hleb Valoshka <375gnu@gmail.com> wrote:
>> No, It's on GNU/Linux. On GNU/kFreeBSD I have "Errno::ENOSYS: Function
>> not implemented - mq_open" (yes, I've made "kldload mqueuefs).
>
> OK, I've reproduced this, too.  Could be an EGLIBC issue...

Or kernel. The following is from irc:

<KiBi> kibi@asdfasdf:~$ sudo ipcs
<KiBi> kernel not configured for shared memory
<KiBi> kernel not configured for semaphores
<KiBi> kernel not configured for message queues
<KiBi> but maybe that's just some missing bits in the kernel.

I dunno what version they run on asdfasdf, but on sid the report from
ipcs is the same.