Re: Synchronous vs. Asynchronous

[Peter Stephenson <p.w.stephenson@xxxxxxxxxxxx>]

On Fri, 20 Aug 2010 10:36:38 -0700
Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx> wrote:
> On Aug 20,  4:45pm, Peter Stephenson wrote:
> } My big remaining worry is whether the >(...) could think it's in the
> } foreground when it's actually in the background after the patch in the
> } second subthread.
> 
> Yes, mine too.

I think that's not a problem in the case of redirection, and I should
have known... as already described in the manual, the reason
"external_command > >(blah)" apparently executes "blah" asynchronously
is because it forks "external_command" before starting the process
substitution.  (Casual visitors should remember that the difference
between starting a process in the background and the foreground is
essentially that the shell doesn't wait for the former; the sequence of
starting processes is otherwise the same, though the jiggery pokery with
terminal handling is different.)  So there's no chance of >(blah)
grabbing the terminal when it shouldn't because it's already associated
with a job that's disconnected from the terminal; it's basically the
same case as starting a new command in a list of commands that's been
forked to the background.  (If it tried to talk to the terminal, having
already been forked, the job would be stopped by SIGTTIN, so you'd find
out straight away --- I think that would indicate in this case that the
logic was screwy rather than what we were trying to do was wrong,
however.)  The change I made simply means it grabs the terminal when it
should, which is when it's in the foreground and there was no fork,
which is exactly what we need to make it get ^C.

People who read the zsh source code for fun(*) may be confused because
prefork() documents process substitutions as happening at that point.
That's the case of >(...) as an ordinary(**) command line argument, not
redirection.  It turns out this is different enough that the change
doesn't have any effect:

cat >(sleep 100)

is uninterruptible, while

cat <(sleep 100)

is interruptible but the sleep doesn't get killed.  Both these are true
with and without the ESUB_ASYNC change (unless I'm even more confused
than I think I am).

I think the difference must be because with >(...) we add the PID to our
job table of things to wait for, whereas with <(...) we don't (see
getproc() in exec.c).  (My best guess as to why is that in the case of
<(...) it will get SIGPIPE if the reader of the file goes away, so it
will go away of its own accord, though I'm not at all sure that's
guaranteed in all circumstances; although it may just be a hack to make
it interruptible for all i know.)  Then presumably with >(...) we have
the same problem that the process isn't attached to a suitable job to
allow it to get the ^C but the shell still waits for it to finish (as it
was created before forking the main command it's there in the job
table).

I'm tempted to back off the change for non-redirection process
substitution and attack that separately, and commit the remaining
shebang.


Another piece of unfinished business is the reference to "asynchronous"
process substitutions.  This is replaced by some circumlocution below.


(*) "people who read the zsh source code for fun": This reminds me of
those scenes where Krusty the Clown, standing up performing before an
audience, makes a joke, and the only thing you can hear on the soundtrack
in response is someone coughing at the back of the hall.

(**) "the case of >(...) as an ordinary command line argument": This
reminds of a sketch I can vaguely remember of a gorilla travelling on
the Tube (London Underground to overseas visitors) and everyone ignoring
it.  It might have been Monty Python.  Or I might just have invented it.


(The book of the Collected Zsh Mailing List Posts On Process Management
will soon be published by Fantasy Press, price several thousand dead
brain cells.)


Index: Doc/Zsh/expn.yo
===================================================================
RCS file: /cvsroot/zsh/zsh/Doc/Zsh/expn.yo,v
retrieving revision 1.117
diff -p -u -r1.117 expn.yo
--- Doc/Zsh/expn.yo	13 Jun 2010 15:59:59 -0000	1.117
+++ Doc/Zsh/expn.yo	21 Aug 2010 17:45:19 -0000
@@ -390,7 +390,8 @@ Process substitutions may be used follow
 case, the substitution must appear with no trailing string.
 
 In the case of the tt(<) or tt(>) forms, the shell runs the commands in
-var(list) asynchronously.  If the system supports the tt(/dev/fd)
+var(list) as a subprocess of the job executing the shell command line.
+If the system supports the tt(/dev/fd)
 mechanism, the command argument is the name of the device file
 corresponding to a file descriptor; otherwise, if the system supports named
 pipes (FIFOs), the command argument will be a named pipe.  If the form with
@@ -456,7 +457,7 @@ version of the example above:
 example(tt(paste <LPAR()cut -f1) var(file1)tt(RPAR() <LPAR()cut -f3) var(file2)tt(RPAR()) tt(> >LPAR())var(process)tt(RPAR()))
 
 (note that no tt(MULTIOS) are involved), var(process) will be run
-asynchronously.  The workaround is:
+asynchronously as far as the parent shell is concerned.  The workaround is:
 
 example(tt({ paste <LPAR()cut -f1) var(file1)tt(RPAR() <LPAR()cut -f3) var(file2)tt(RPAR() }) tt(> >LPAR())var(process)tt(RPAR()))
 

-- 
Peter Stephenson <p.w.stephenson@xxxxxxxxxxxx>
Web page now at http://homepage.ntlworld.com/p.w.stephenson/