OK, I got some time today to look into it and I found and fixed
the problem. It was where I expected, and the elegant OpenVMS
afforded an elegant solution.
There is this nice IO$M_READERCHECK qualifier you can put on
writes to mailboxes which cause the write to complete immediately
if no read channels are assigned to the mailbox. This is the case
here. The child process dies, leaving a mailbox with no readers
and perl hangs trying to write an EOF into the mailbox.
Robert, if you are able to rebuild Perl and are eager for this fix,
just change the line that reads:
retsts = sys$qiow(0,chan,IO$_WRITEOF,iosb,0,0,0,0,0,0,0,0);
to
retsts =
sys$qiow(0,chan,IO$WRITEOF|IO$M_READERCHECK,iosb,0,0,0,0,0,0,0,0);
in [.VMS]VMS.C from the top level of your Perl hierarchy and rebuild
and you should be OK.
Now, if I can find any more time to look into this, maybe I'll prepare
a diff and join the rarified club of Perl Maintainters! Yaah!
> -----Original Message-----
> From: Jordan Henderson [mailto:[EMAIL PROTECTED]]
> Sent: Friday, February 18, 2000 1:30 PM
> To: 'Robert Sundstrom'; [EMAIL PROTECTED]
> Subject: RE: Executing DCL-commands from Perl programs
>
>
> That's a bug, allright.
>
> Looks to me like it's my_pclose() in VMS.C gets hung up
> trying to write an EOF
> to the pipe. I haven't really debugged it, but I have some
> inductive reasons to
> believe this.
>
> When try Robert's test the subprocess created by the pipe
> open is dead, but it
> appears that there's a write active on a MBA device (noted
> with ANAL/SYS). You
> can read that mailbox (just a COPY MBANNNN: TT:) and the
> process completes. I
> see that there's an WRITE EOF in my_pclose() that does a
> SYS$QIOW (wait) with no
> timeout or anything.
>
> I don't know why an ERROR exit status like 44 would be
> different than a success
> status here. The error exit must force closing of SYS$INPUT
> or something which
> causes the process to die before the EOF can be written (just
> a wild guess).
>
> If I get a chance this weekend, I'll look into this in more detail. A
> workaround would just to make that EOF write a SYS$QIO and
> enforce a timeout.
> Any number of cleaner fixes could be imagined. I would bet
> that a SYS$GETDVI()
> could be done on the MBA device being written to to determine
> the owner and if
> the owner is gone (Owner = "", which it is in this case) just
> forgo the writing
> of the EOF.
>
> It could easily be the case that I won't get a chance to look
> at it this
> weekend, but I thought someone might want to hear my
> speculations anyway.
>
> Robert,
>
> You probably want a workaround that doesn't involve debugging
> and rebuilding
> Perl...
>
> I can't come up with anything other than creating a temporary
> file for your
> commands. Anybody have other ideas?
>
> > -----Original Message-----
> > From: Robert Sundstrom [mailto:[EMAIL PROTECTED]]
> > Sent: Friday, February 18, 2000 11:59 AM
> > To: [EMAIL PROTECTED]
> > Subject: Executing DCL-commands from Perl programs
> >
> >
> > Hello all.
> > I have been trying to integrate DCL-scripts with
> > Perl-programs. It works
> > fine when one is using temporary files to build the sequence of
> > DCL-commands you want. On the VMS Perl FAQ I learned about
> a method to
> > execute several DCL-commands in a row without using
> temporary files.
> >
> > Unfortunately, it does not work well if you have a DCL-script
> > that ends
> > with exit 44, that is forcing an abort. The Perl program
> never resumes
> > execution after the print statement. Something probably
> > aborts (just as it
> > is supposed to... :-), and control is never returned to Perl.
> >
> > I don't have enough VMS-diplomas to determine if this is a
> feature, a
> > misfeature or a bug. I simply either want my Perl script to
> > abort as well,
> > or to be able to detect the EXIT parameter of the DCL-script
> > in my Perl
> > program. Does anyone have any advice on this matter?
> >
> > This is my source code:
> >
> > A.COM
> >
> > $ EXIT 44
> >
> > A.PL
> >
> > open(CMD,"|\@sys\$input");
> > print CMD "\@a.com\n";
> > close(CMD);
> > exit;
> >
> > A little transcript of my output:
> >
> > ROBERT>type a.com
> > $ exit 44
> >
> > ROBERT>perl a.pl
> > %SYSTEM-F-ABORT, abort
> > Interrupt
> >
> > ROBERT>stop
> >
>
