Re: Race condition hangs on multiple mintty/tcsh? Brad Wetmore

[Thomas Wolff <towo@towo.net>]

Am 06.08.2020 um 16:26 schrieb Corinna Vinschen:
> On Aug  6 15:31, Thomas Wolff wrote:
>> Am 06.08.2020 um 13:46 schrieb Thomas Wolff:
>>> Am 06.08.2020 um 01:23 schrieb Kevin Schnitzius via Cygwin:
>>>> On Wednesday, August 5, 2020, 06:56:48 PM EDT, Thomas Wolff
>>>> <towo@towo.net> wrote:
>>>>> Am 04.08.2020 um 12:02 schrieb Thomas Wolff:
>>>>>> Am 04.08.2020 um 00:13 schrieb Brad Wetmore via Cygwin:
>>>>>>> Hi,
>>>>>>>
>>>>>>> I generally kick off multiple (10) mintty sessions, and place them
>>>>>>> around the screen.
>>>>>>>
>>>>>>> C:/cygwin64/bin/mintty.exe -i /Cygwin-Terminal.ico \
>>>>>> --position @3 --position 120,0 --size 80x71 /bin/tcsh &
>>>>>>> C:/cygwin64/bin/mintty.exe -i /Cygwin-Terminal.ico \
>>>>>>> --position @3 --position 715,0 --size 80x45 /bin/tcsh &
>>>>>>> C:/cygwin64/bin/mintty.exe -i /Cygwin-Terminal.ico \
>>>>>>> --position @3 --position 715,660 --size 80x24 /bin/tcsh &
>>>>>>>
>>>>>>> Within the last 6 months or so, about 2-3 of them would hang and
>>>>>>> either mintty/tcsh would not start. I put a "sleep 1" in between each
>>>>>>> invocation and that seemed to take care of it.
>>>>>>>
>>>>>>> With the latest cygwin update, about 8 of them just hang even with
>>>>>>> the sleep 1. I put in a "sleep 2", and now everything is coming up
>>>>>>> again.
>>>>>>>
>>>>>>> Not sure if this is a mintty or tcsh issue, but just wondering if
>>>>>>> others are seeing this before I start trying to debug this.
>>>>>> I can reproduce such behaviour with /bin/bash (easy cross-check), and
>>>>>> in fact the shell is running in that case (easy test via `echo >
>>>>>> .log`), so I have a vague and unpleasant suspicion it might in fact be
>>>>>> related to mintty although I have no idea how that would happen. To be
>>>>>> analysed.
>>>>> When in this state, on the pty from which mintty receives child process
>>>>> output, select() does not report a ready for reading condition;
>>>>> could it
>>>>> be related to the recent poll/select patch? I could not reproduce it in
>>>>> cygwin 3.0.7.
>>>> for i in 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 ; do
>>>> mintty  -i /Cygwin-Terminal.ico  --position @3 --position 120,$i
>>>> --size 80x24 /bin/bash & done
>>>>
>>>> This does not fail for me.  Nor does it fail from cmd or powershell
>>>> using a script.
>>>>
>>>> I tried 3.1.4 and 3.1.6 on Windows 10.  I tried up to 100 instances
>>>> of of mintty...
>>> Thanks for testing. It is in fact hard to reproduce, maybe also
>>> depending on system load (speculating).
>>> I could reproduce one case of one of three terminals being unresponsive
>>> also with xterm.
>> And it also happens if I drop select() from mintty (and use just
>> non-blocking read()).
>>> Does anybody familiar with pty/select or recent changes have any idea?
>> ?
> Something with the pseudo console?  Can you test w/o pseudo console code, just to check?
Hmm, CYGWIN=disable_pcon does not seem to disable ConPTY support for me 
any more...
>
> Other than that we probably need some bisecting to find the cause.