From maintainers-request at octave dot org Sun Jan 30 16:12:31 2005
Subject: Re: Reprise: external pager quitting kills standard error
From: Daniel J Sebald <daniel dot sebald at ieee dot org>
To: "Dmitri A. Sergatskov" <dmitri at unm dot edu>
Cc: maintainers at octave dot org
Date: Sun, 30 Jan 2005 16:18:47 -0600

Dmitri A. Sergatskov wrote:

> I am on FC3 and I cannot reproduce this problem:
>
> octave:1> bogus
> error: `bogus' undefined near line 1 column 1
> octave:1> junk = ones(100)
> warning: connection to external pager (pid = 26581) lost --
> warning: attempting to finish pending computations...
> warning: broken pipe
> octave:2> bogus
> error: `bogus' undefined near line 2 column 1
> octave:2> junk = ones(100)
> octave:3>
> octave:3> bogus
> error: `bogus' undefined near line 3 column 1
> octave:3> junk = ones(1000)
> warning: connection to external pager (pid = 26583) lost --
> warning: attempting to finish pending computations...
> warning: broken pipe
>
> (I am hitting "q" after first page is displayed)


Yes "q", right.  But thanks for the results because it illustrates 
something.  I am not getting the "warning: broken pipe" complaint before 
the error messages are lost.  This could be a sign of where the problem 
lies.

I'm logically sort of narrowing this down.  I think it does have 
something to do with buffer behavior.  Below is some code in pager.cc, 
which may be the origin of the problem (not the problem 'though).  I 
suspect for some reason the "external_pager->good()" fails so the  "// 
We had a pager, but it must have died" section gets run.  This command 
"octave_set_interrupt_handler (saved_interrupt_handler);" is probably 
what messes things up (something to do with interrupts and 
synchronization, tricky stuff).  However, even if it were, probably 
"external_pager->good()" should not have failed.  [Remember, I've not 
tested this, just thinking it through.]

Furthermore, the external pager is a stream.  And it wouldn't surprise 
me that when large amounts of data are put into the stream and the other 
end quits while there is still stuff to be pulled out, that the stream 
might think the external pager has died.  (Does "less" pull everything 
out of the stream at the beginning?  Or only as the cursor is advanced?)

// Our actual connection to the external pager.
static oprocstream *external_pager = 0;

Dan






      if (external_pager)
        {
          if (octave_pager_pid > 0 && external_pager->good ())
        {
          external_pager->write (msg, len);

          // These checks are needed if a signal handler
          // invoked since the last set of checks attempts
          // to flush output and then returns

          if (octave_pager_pid > 0
              && external_pager
              && external_pager->good ())
            external_pager->flush ();
        }
          else
        {
          // We had a pager, but it must have died.  Restore
          // the interrupt state so we can escape back to the
          // prompt if there are lots of computations pending.

          if (interrupt_handler_saved)
            {
              octave_set_interrupt_handler (saved_interrupt_handler);
              interrupt_handler_saved = false;
            }
        }
        }
      else
        {
          std::cout.write (msg, len);
          std::cout.flush ();
        }
    }