Re: name pipe problem: 1 writer, multiple concurrent readers

[Ken Brown <kbrown@cornell.edu>]

On 5/19/2020 11:20 AM, Morten Kjærulff wrote:
> On Tue, May 19, 2020 at 3:55 PM Ken Brown via Cygwin <cygwin@cygwin.com> wrote:
>>
>> On 5/19/2020 7:26 AM, Morten Kjærulff via Cygwin wrote:
>>> Hi,
>>>
>>> The following script:
>>>
>>> #!/bin/sh
>>>
>>> rm -f t.pip t.txt
>>>
>>> mkfifo t.pip
>>> printf "line1\nline2\n" >t.txt
>>>
>>> ps
>>>
>>> {
>>>     while true ; do
>>>       cp t.txt t.pip
>>>     done
>>> } &
>>>
>>> rm -f t.rc.*
>>> for rc in 0 1 2 3 4 5 6 7 8 9 ; do
>>>     {
>>>       diff t.pip t.txt
>>>       echo $? >t.rc.$rc
>>>     }
>>> done
>>> echo result1 start
>>> ls  t.rc.* | wc -l
>>> cat t.rc.*
>>> echo result1 end
>>>
>>> rm -f t.rc.*
>>> for rc in 0 1 2 3 4 5 6 7 8 9 ; do
>>>     {
>>>       diff t.pip t.txt
>>>       echo $? >t.rc.$rc
>>>     } & # run the readers in parallel
>>> done
>>> sleep 10
>>> echo result2 start
>>> ls  t.rc.* | wc -l
>>> cat t.rc.*
>>> echo result2 end
>>>
>>> ps
>>>
>>> Give me output like this:
>>>
>>> $ ./tpip.sh
>>>         PID    PPID    PGID     WINPID   TTY         UID    STIME COMMAND
>>>        1642    1600    1642      65264  cons1    1058872 13:18:58 /usr/bin/sh
>>>        1600       1    1600      72728  cons1    1058872 13:18:33 /usr/bin/bash
>>>        1645    1642    1642      42088  cons1    1058872 13:18:58 /usr/bin/ps
>>> result1 start
>>> 10
>>> 0
>>> 0
>>> 0
>>> 0
>>> 0
>>> 0
>>> 0
>>> 0
>>> 0
>>> 0
>>> result1 end
>>> 0a1,2
>>>> line1
>>>> line2
>>> diff: t.pip
>>> result2 start
>>> 1
>>> 2
>>> result2 end
>>>         PID    PPID    PGID     WINPID   TTY         UID    STIME COMMAND
>>>        1690    1688    1642     104032  cons1    1058872 13:19:01 /usr/bin/diff
>>>        1681    1642    1642      95012  cons1    1058872 13:19:00 /usr/bin/sh
>>>        1642    1600    1642      65264  cons1    1058872 13:18:58 /usr/bin/sh
>>>        1684    1681    1642      99624  cons1    1058872 13:19:00 /usr/bin/diff
>>>        1678    1676    1642      94532  cons1    1058872 13:19:00 /usr/bin/diff
>>>        1688    1642    1642      88864  cons1    1058872 13:19:01 /usr/bin/sh
>>>        1698    1642    1642     104820  cons1    1058872 13:19:11 /usr/bin/ps
>>>        1692    1642    1642      66572  cons1    1058872 13:19:01 /usr/bin/sh
>>>        1677    1674    1642      86692  cons1    1058872 13:19:00 /usr/bin/diff
>>>        1646    1642    1642      30888  cons1    1058872 13:18:58 /usr/bin/sh
>>>        1600       1    1600      72728  cons1    1058872 13:18:33 /usr/bin/bash
>>>        1686    1685    1642      14320  cons1    1058872 13:19:01 /usr/bin/diff
>>>        1685    1642    1642      25608  cons1    1058872 13:19:00 /usr/bin/sh
>>>        1676    1642    1642     104212  cons1    1058872 13:19:00 /usr/bin/sh
>>>        1689    1642    1642      98004  cons1    1058872 13:19:01 /usr/bin/sh
>>>        1674    1642    1642      44152  cons1    1058872 13:19:00 /usr/bin/sh
>>>        1680    1646    1642      28224  cons1    1058872 13:19:00 /usr/bin/cp
>>>        1682    1679    1642      43612  cons1    1058872 13:19:00 /usr/bin/diff
>>>        1694    1692    1642      67736  cons1    1058872 13:19:01 /usr/bin/diff
>>>        1683    1642    1642      93544  cons1    1058872 13:19:00 /usr/bin/sh
>>>        1679    1642    1642      90188  cons1    1058872 13:19:00 /usr/bin/sh
>>>        1691    1689    1642      68560  cons1    1058872 13:19:01 /usr/bin/diff
>>>        1687    1683    1642      83952  cons1    1058872 13:19:01 /usr/bin/diff
>>>
>>> $
>>>
>>> That is, when I run the readers in sequence, no problem, but when I
>>> run the in parallel, they either hang or t.pip seems empty.
>>
>> The current cygwin release doesn't support multiple concurrent readers of a
>> FIFO.  I have recently added that support, which should appear in the next
>> release, although there are still bugs to be fixed.  See
>>
>>     https://cygwin.com/pipermail/cygwin-patches/2020q2/010195.html
>>
>> Ken
> Great!

Let's keep the discussion on the mailing list
> Then we will have a read-only file (pipe), which is the output of an
> abitrary command, right?

Yes.

> Can we have multiple concurrent writers and one reader, which would
> become a write-only file, which becomes the input of an abitrary
> command?

That already exists, although I found and fixed some bugs when I was developing 
the multiple reader support.

> Anywhere I can see the plans for next release, if any?

That's up to Corinna.

Ken