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

[Takashi Yano <takashi.yano@nifty.ne.jp>]

Hi Thomas,

On Thu, 6 Aug 2020 16:33:36 +0200
Thomas Wolff wrote:
> 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...

CYGWIN=disable_pcon needs to be set before opening pty, i.e.
before opening mintty. It works for me with cygwin 3.1.6.

-- 
Takashi Yano <takashi.yano@nifty.ne.jp>