[petsc-dev] Use of the DISPLAY environment variable

[Barry Smith]

On Aug 29, 2010, at 2:45 PM, Jed Brown wrote:

> PETSc uses the -display option differently from the DISPLAY environment
> variable.  In particular, it prepends the hostname when DISPLAY starts
> with ":".
> 
>  ierr = PetscOptionsGetString(PETSC_NULL,"-display",PetscDisplay,256,&flag);CHKERRQ(ierr);
>  if (flag) PetscFunctionReturn(0);
> [...]
>    str = getenv("DISPLAY");
>    if (!str || (str[0] == ':' && size > 1)) {
>      ierr = PetscGetHostName(display,255);CHKERRQ(ierr);
>      ierr = PetscStrcat(display,":0.0");CHKERRQ(ierr);
>    } else {
>      ierr = PetscStrncpy(display,str,256);CHKERRQ(ierr);
>    }
> 
> Is this really the right thing (ignoring the small overflow
> possibility)?  Why not concatenate hostname with the value of DISPLAY
> when it starts with ':'?

   I'm not sure why it was set to always use :0.0 there. Yes it sounds reasonable to use DISPLAY there.
> 
> But more importantly, firewalls are frequently set up to block port
> 6000, so DISPLAY=hostname:0.0 won't work, but DISPLAY=:0.0 will work
> (obviously only among local nodes).  I've just gotten used to always
> running with -display :0 in parallel, but how about reducing the
> hostname to check whether it is the same on all hosts.  If so, then just
> use DISPLAY as is.  This would make graphics Just Work on more systems.

    This sounds fine.

   Barry

> 
> Does this sound reasonable?
> 
> Jed