bug#14569: 24.3.50; bootstrap fails on Cygwin

[Ken Brown <kbrown@cornell.edu>]

On 6/16/2013 11:01 AM, Christopher Faylor wrote:
> On Sun, Jun 16, 2013 at 09:11:21AM -0400, Ken Brown wrote:
>> [Adding the bug address back to the CC so that this gets archived.]
>>
>> On 6/15/2013 9:54 AM, Angelo Graziosi wrote:
>>> Christopher Faylor wrote
>>>>> On 06/14/2013 11:03 AM, Christopher Faylor wrote:
>>>>>> You pointed to an archived mail messages which implies that was fixed
>>>>>> more than a year ago.  What makes you think it is still a problem?
>>>>>
>>>>> The message I pointed to
>>>>> <http://cygwin.com/ml/cygwin/2012-05/msg00472.html> says this:
>>>>>
>>>>>> Testcase signal/kill: Signals may or may not reach the correct thread
>>>>>> with 1.7.12-1 and newer.
>>>>>
>>>>> Confirmed.  I think the reason is that we only have a single event to
>>>>> signal that a POSIX signal arrived instead of a per-thread event, but
>>>>> I'm not sure.  This is cgf's domain so I leave it at that for now.
>>>>>
>>>>> I interpreted this to mean "the existence of the bug is confirmed,
>>>>> here's why the bug occurs, and I'll let cgf deal with it".  I didn't
>>>>> see any followup message where cgf (is that you?) dealt with it.  My
>>>>> apologies if I misinterpreted the email.
>>>>
>>>> Oops.  I didn't read Corinna's message as thoroughly as I should have.
>>>> Sorry.
>>>>
>>>> That particular issue was supposed to have been fixed in Cygwin 1.7.17,
>>>> released in October 2012.
>>>
>>> Out of curiosity, I tried the test cases I found in that thread, more
>>> precisely here:
>>>
>>>     http://cygwin.com/ml/cygwin/2012-05/msg00434.html
>>>
>>>
>>> and the results are:
>>>
>>> $ gcc otto_test1.c -o otto_test1
>>> $ ./otto_test1
>>> Testing deferred pthread_cancel()
>>>
>>> Thread 0 starting (0x200102c0)
>>> Thread 1 starting (0x20010360)
>>> Thread 2 starting (0x20010400)
>>>
>>> Cancelling thread 2 (0x20010400)
>>> Thread 2 exiting (0x20010400)
>>> Cancelling thread 1 (0x20010360)
>>> Thread 1 exiting (0x20010360)
>>> Cancelling thread 0 (0x200102c0)
>>> Thread 0 exiting (0x200102c0)
>>>
>>> Thread 0 is gone (0x200102c0)
>>> Thread 1 is gone (0x20010360)
>>> Thread 2 is gone (0x20010400)
>>>
>>> $ gcc otto_test2.c -o otto_test2
>>> $ ./otto_test2
>>> Testing asynchronous pthread_cancel()
>>>
>>> Thread 0 starting (0x200102c0)
>>> Changing canceltype from 0 to 1
>>> Thread 1 starting (0x20010360)
>>> Changing canceltype from 0 to 1
>>> Thread 2 starting (0x20010400)
>>> Changing canceltype from 0 to 1
>>>
>>> Cancelling thread 2 (0x20010400)
>>> Thread 2 exiting (0x20010400)
>>> Cancelling thread 1 (0x20010360)
>>> Thread 1 exiting (0x20010360)
>>> Cancelling thread 0 (0x200102c0)
>>> Thread 0 exiting (0x200102c0)
>>>
>>> Thread 0 is gone (0x200102c0)
>>> Thread 1 is gone (0x20010360)
>>> Thread 2 is gone (0x20010400)
>>>
>>> $ gcc otto_test3.c -o otto_test3
>>> $ ./otto_test3
>>> Testing pthread_kill()
>>>
>>> Thread 0 starting (0x200102c0)
>>> Thread 1 starting (0x20010360)
>>> Thread 2 starting (0x20010400)
>>>
>>> Sending SIGUSR1 to thread 2 (0x20010400)
>>> Thread 2 executes signal handler (0x20010400)
>>> Thread 2 encountered an error: Interrupted system call (0x20010400)
>>> Sending SIGUSR1 to thread 1 (0x20010360)
>>> Thread 1 executes signal handler (0x20010360)
>>> Thread 1 encountered an error: Interrupted system call (0x20010360)
>>> Sending SIGUSR1 to thread 0 (0x200102c0)
>>> Thread 0 executes signal handler (0x200102c0)
>>> Thread 0 encountered an error: Interrupted system call (0x200102c0)
>>>
>>> Are the errors in the last test case to be expected under the 20130612
>>> snapshot (CYGWIN_NT-5.1, 1.7.21s 20130612 21:06:59, i686 Cygwin)?
>>
>> I can replicate this on my system, consistently.  There's clearly a
>> problem, but it's not the same as in the original Cygwin bug report.  In
>> the present case, the signal is received by the right thread, but
>> something goes wrong afterwards.
>
> Try it on Linux.  I don't see any difference.  "An error" in this case
> seems to be the script working as designed.
>
> % man sem_wait
>
>      SEM_WAIT(3)   Linux Programmer's Manual      SEM_WAIT(3)
>
>
>
>      NAME
> 	   sem_wait, sem_timedwait, sem_trywait - lock a semaphore
>
>      ...
>
>      ERRORS
> 	   EINTR  The call was interrupted by a signal handler; see signal(7).

Yeah, I missed that.  Sorry for the noise.

Ken