From bug-octave-request  Fri Jul  2 15:01:01 1993
Subject: ^c problem on Dells
From: A. Scottedward Hodel <Dr dot A dot Scottedward dot Hodel at eng dot auburn dot edu>
To: bug-octave
Cc: Dr dot A dot Scottedward dot Hodel at eng dot auburn dot edu, tiller@galois.msfc.nasa.gov
Date: Fri, 2 Jul 93 15:00:56 CDT


We have compiled octave-0.72 on a silicon graphics machine at MSFC and
have encountered similar behavior to what I described on the DELLS:

^C -> some type of fault on the next command.

Here's what we've got:

> Date: Fri, 2 Jul 1993 13:39:04 -0500
> From: tilleja at galois dot msfc dot nasa dot gov (John Tiller)
> To: hodelas at controls dot msfc dot nasa dot gov
> Subject: 1/0
> 
> Looking at the 1/0 problem on the SGI, it appears that the problem is that
> without a xisinf function, octave is trying to format infinity and never
> completing.
> 
1/0 yields Inf on suns, runs out of virtual memory on both Dell's 
and on SGI.

> Date: Fri, 2 Jul 1993 13:41:42 -0500
> From: tilleja at galois dot msfc dot nasa dot gov (John Tiller)
> To: hodelas at controls dot msfc dot nasa dot gov
> Subject: xisnan and xisinf
> 
> Looks like both xisnan and xisinf are problem areas.  Eaton has a "work
> around" for xisinf in mappers.cc that is not completely right.  We might
> just put these on the list of problems and go on.
> 
> Date: Fri, 2 Jul 1993 13:43:18 -0500
> From: tilleja at galois dot msfc dot nasa dot gov (John Tiller)
> To: hodelas at controls dot msfc dot nasa dot gov
> Subject: ``I can't stop myself...''
> 
> Guess what, on the SGI, CNTL-C generates a SIGILL on the next command.
> Another data point.
> 
> Date: Fri, 2 Jul 1993 14:08:17 -0500
> From: tilleja at galois dot msfc dot nasa dot gov (John Tiller)
> To: hodelas at controls dot msfc dot nasa dot gov
> Subject: first clue
> 
> See line 266 in symtab.cc.  This is where the debugger is telling me I'm
> getting a segmentation fault after an interrupt.  Could be that something
> is being deleted that should not be deleted?
> 
> From tilleja at galois dot msfc dot nasa dot gov Fri Jul  2 14:41:35 1993
> To: Dr dot A dot Scottedward dot Hodel at eng dot auburn dot edu
> Subject: Re:  first clue
> 
> >From gdb:
> 
>   0  0  0  0  0  0  0  0  1  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
>   0  0  0  0  0  0  0  0  0  1  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
>   0  0  0  0  0  0  0  0  0  0  1  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
>   0  0  0  0  0  0  0  0  0  0  0  1  0  0  0  0  0  0  0  0  0  0  0  0  0  0
>   0  0  0  0  0  0  0  0  0  0  0  0  1  0  0  0  0  0  0  0  0  0  0  0  0  0
>   0  0  0  0  0  0  0  0  0  0  0  0  0  1  0  0  0  0  0  0  0  0  0  0  0  0
>   0  0  0  0  0  0  0  0  0  0  0  0  0  0  1  0  0  0  0  0  0  0  0  0  0  0
>   0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  1  0  0  0  0  0  0  0  0  0  0
>   0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  1  0  0  0  0  0  0  0  0  0
>   0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  1  0  0  0  0  0  0  0  0
> 
> Program received signal 2, Interrupt
> _syssgi () at sys/_syssgi.s:16
> sys/_syssgi.s:16: No such file or directory.
> (gdb) signal 2
> Continuing with signal 2.
> 
> octave:2> 3/2
> 
> Program received signal 11, Segmentation fault
> 0x42ad4c in define__13symbol_recordP4treeQ210symbol_def11symbol_type (
>     this=0x10052490, t=0x1007b3f8, st=variable) at symtab.cc:266
> 266               delete saved_var_def;
> (gdb)
> 
> 
> Date: Fri, 2 Jul 1993 14:44:36 -0500
> From: tilleja at galois dot msfc dot nasa dot gov (John Tiller)
> To: hodelas at controls dot msfc dot nasa dot gov
> Subject: gnu oststream bug
> 
> Here is something else you might want to pass along to Eaton.  He ran across
> the same bug in GNU libg++ that I did with ostrstream::str.  It *should* put
> a null byte at the end but doesn't.  I "fixed" the problem by adding these
> lines:
> at strstream.C line 132:
>     memset(buf,0,initial_size);
> and at strstream.C line 81:
>     memset(new_buf,0,new_size);
> I haven't passed these along to the GNU people yet, but they work for me.
> 
> Date: Fri, 2 Jul 1993 14:23:47 -0500
> From: tilleja at galois dot msfc dot nasa dot gov (John Tiller)
> To: hodelas at controls dot msfc dot nasa dot gov
> Subject: conjecture
> 
> Since standard C++ does not have exception handling in it yet, my guess is
> that Eaton is having to do some funny stuff to handle destructors during an
> exception and something is not getting reintialized properly when he does his
> longjmp.