Help debugging Re: bug report

[Mathew Yeates]

Hi-
I'm tryin to debug groff running on SGI Irix 6.5.

When I do a system trace on man I see that file descriptors 0,1,2 are
closed by make immediately prior to gnroff being forked. Is this correct
behavior? I tried adding the line "echo HELLO" near the beginning of
gnroff and, as expected, this fails with a "bad file descriptor" message
in my trace.

Does that happen with other OS's?

Mathew

> Sorry, this doesn't ring any bells with me.  Can you send me the test
> case?  I'll try building a debug version of groff (2.5?) and see if I
> can find the close().
> 
>       David
> 
> Werner LEMBERG wrote:
> > 
> > [David, maybe you can help here.  I've noted that recently a similar
> > problem has been discussed on the emacs-devel list, and maybe this
> > problem is related.
> > 
> > Here the original report; Mathew is using an SGI Octane with Irix 6.5:
> > 
> >   Error: geqn fails when it tries to write output to stdout.  This
> >   happens at line 378 of main.cc.  Doing a system call trace, it
> >   appears that stdout has been closed.  I added a line of code to
> >   print out the errno when the printf fails.  I get "stdout': Bad file
> >   number"
> > 
> >   It appears that stdout was closed long before the error.  In the
> >   trace I see "open" calls returning the value 1.
> > 
> > Does this sound familiar?]
> > 
> > > Something really strange.... my system trace shows that man closes
> > > file descriptors 0,1, and 2 just prior to execv-ing nroff.  Maybe
> > > the problem here is SGI's "man"??  I notice that when I run using
> > > the sgi nroff, eqn does not write to stdout.  Instead, nothing is
> > > written to stdout again until man does a "write" after calling
> > > ioctl(1, __OLD_TCGETA, <bad indirect __old_termio param - len 30>)
> > > OK
> > >
> > > Maybe it isnt safe to call printf from with eqn without checking
> > > first that the IO file handles are still open??
> > 
> > AFAIK, stderr, stdin, and stdout are opened and closed automatically
> > at the start and end of the program.  If nothing explicitly closes the
> > handles, they should be always available.  There is some trickery to
> > duplicate streams in groff.cc IIRC, but none is really closed.
> > Anyway, this shouldn't affect eqn.
> > 
> >     Werner
> 
> -- 
> David KAELBLING <address@hidden>          Silicon Graphics Computer Systems
> 1 Cabot Rd, suite 250; Hudson, MA 01749           781.839.2157, fax ...2357