expectperl-discuss

Re: [Expectperl-discuss] pipeline question (fwd)

[<gl...@ma...>]

Forwarded message:
> From: "Roland Giersig" <Ro...@a1...>
> To: gl...@ma...
> Cc: exp...@li...
> Date: Mon, 23 Jul 2001 17:07:52 +0200
> 
> > I am trying to do something like this:
> > 
> > $ob=Expect->spawn("rsh machine command arg | localcmd");
> > 
> > I get the impression that spawn cannot handle the pipeline.
> 
> Right, this is a bug, I will fix it in v1.12, which I'm testing
> and preparing for release right now.  But due to my pending vacation
> it will only be released mid to end of august.
> 
> You can of course just patch your copy of Expect.pm, simply
> substitute the "exec(@_)" inside sub spawn with "exec($cmd)".

Thanks :)  I'm about to try it out...

> > I am guessing that I need to do this:
> > 
> > $ob1=Expect->spawn("rsh machine command arg");
> > $ob2=Expect->spawn("localcmd");
> > 
> > and then use interconnect ....???
> 
> This is one possiblity, but probably not the best.  What is that
> localcmd you want to run at the end of the pipeline?  
> Couldn't you just use qx(rsh machine | localcmd) instead of expect and
> then process the output by regular means?  I can't imagine
> a case where a program within a pipe requires user interaction.
> 
> Oh, maybe rsh is asking for a password? ...

Nope.
Since you're interested, this is the guts of what I am trying to do:

spawn("rsh machine dump_cmd various args... mount-point |cat -|dd bs=64k of=tape-device");

then use expect to catch the message:
"End of tape reached.  Press RETURN when new tape loaded ..."
(message varies depending on the dump software - yes, I'm doing this
from multiple machines/OSes/dump commands, each with different 
idiosyncrasies, arguments, etc...)

When I get the expect message, I call a sub to change tapes using the
tape robot.  That's the plan anyway.  :)

Glen

-- 
Glen Davison			gl...@ma...
Computer System Administrator	phone: +61 2 9385 7018
Maths, UNSW			fax:   +61 2 9385 7123

Re: [Expectperl-discuss] pipeline question (fwd)

[Roland G. <r.g...@xs...>]

> Since you're interested, this is the guts of what I am trying to do:
> 
> spawn("rsh machine dump_cmd various args... mount-point |cat -|dd bs=64k of=tape-device");
> 
> then use expect to catch the message:
> "End of tape reached.  Press RETURN when new tape loaded ..."
> (message varies depending on the dump software - yes, I'm doing this
> from multiple machines/OSes/dump commands, each with different
> idiosyncrasies, arguments, etc...)
> 
> When I get the expect message, I call a sub to change tapes using the
> tape robot.  That's the plan anyway.  :)

Oh, OK, that's a perfectly valid use for Expect... :-)

BTW, what's the use of that noop-cat inside that pipe?  'dd' already does
all the buffering you need...

Good luck with that project!

Roland
--
RGi...@cp...

Re: [Expectperl-discuss] pipeline question (fwd)

[<gl...@ma...>]

> 
> > Since you're interested, this is the guts of what I am trying to do:
> > 
> > spawn("rsh machine dump_cmd various args... mount-point |cat -|dd bs=64k of=tape-device");
> > 
> > then use expect to catch the message:
> > "End of tape reached.  Press RETURN when new tape loaded ..."
> > (message varies depending on the dump software - yes, I'm doing this
> > from multiple machines/OSes/dump commands, each with different
> > idiosyncrasies, arguments, etc...)
> > 
> > When I get the expect message, I call a sub to change tapes using the
> > tape robot.  That's the plan anyway.  :)
> 
> Oh, OK, that's a perfectly valid use for Expect... :-)
> 
> BTW, what's the use of that noop-cat inside that pipe?  'dd' already does
> all the buffering you need...

No, unfortunately not.  After a great deal of experimenting, we found
we needed to do this to get the blocking right.  Something to do with
buffering and rsh lags perhaps?

> Good luck with that project!

Thanks.  Believe me, I need it!

I tried the change to the exec in Expect.  Didn't help.  Currently
I am getting this:

Reached end of medium for /dev/nrmt1h; Please wait for file closing
Cannot prompt (can't open /dev/tty): No such device or address

expect is catching the "Press RETURN ..." message, and happily changing
tapes, but the 

print $bakupcmdobj "\r";

which I do after the tape is ready is evidently failing.
The script just hangs - I see this:

# ps -fu root | grep ...
root       3661    198  0.0 -        ??           0:00.00 <defunct>
root        198  12213  0.0 11:14:41 ttyra        0:00.38 perl /usr/local/sbin/bakup_master ....

Any thoughts?


thanks
Glen

-- 
Glen Davison			gl...@ma...
Computer System Administrator	phone: +61 2 9385 7018
Maths, UNSW			fax:   +61 2 9385 7123

Re: [Expectperl-discuss] pipeline question (fwd)

[Roland G. <r.g...@xs...>]

gl...@ma... wrote:

> I tried the change to the exec in Expect.  Didn't help.

Well, it helped insofar as you now can spawn pipe commands, right?

> Currently I am getting this:
> 
> Reached end of medium for /dev/nrmt1h; Please wait for file closing
> Cannot prompt (can't open /dev/tty): No such device or address

Hmm, this sounds bad.  'dd' is trying to read a user response
via /dev/tty, over which Expect doesn't have control.

interconnect() cannot help here, because as long as 'dd' is
reading from stdin, it has to open /dev/tty for a user respone.
[earlier programs tried to read from stderr, and this even worked,
on some systems... :-) ]

Workaround: don't pipe directly into 'dd', use an intermediate
file and spawn 'dd' separately.  Disadvantage: you need enough
harddiskspace.  Workaround for that: a named pipe:
'rsh ... > named_pipe' and 'dd if=named_pipe of=/dev/tape'

That way, 'dd's stdin stays available for user responses.  I do
hope that 'dd' is then using stdin and not stubbornly trying
to open /dev/tty, otherwise you've got a real problem.

> expect is catching the "Press RETURN ..." message, and happily changing
> tapes, but the
> 
> print $bakupcmdobj "\r";
> 
> which I do after the tape is ready is evidently failing.

Yes, because it goes to stdin and not to /dev/tty.

Hope this helps,

Roland
PS: Starting tomorrow, I'm off for summer vacation for two weeks.