Tk - is MainWindow robust?

[matt.tovey]

Hi Monks, I have a Tk application that, apart from it's own window, needs to open a MainWindow on a large number of other displays (today: 20+, tomorrow: ??). This is going to be somewhat error-prone, and the application should be able to continue to run on whichever nodes do manage to start up successfully, so I need the opening of the MainWindow's to not cause too many problems in the event that a node is down, a display does not allow connections, etc. This is what I'm thinking so far:

use strict;
use Tk;

my @Displays = qw"node1:0 node2:0";

my $top = MainWindow->new();
my %top;

my $orig_display = $ENV{'DISPLAY'};
foreach (@Displays) {
  $ENV{'DISPLAY'} = $_;

  if (eval {$top{$_} = MainWindow->new()}) {
    print "MainWindow opened on $_\n";
  } else {
    print "MainWindow failed to open on $_\n";
    $top{$_} = 0;
  }
}
$ENV{'DISPLAY'} = $orig_display;

MainLoop;

Can anybody see any problems with this? Or perhaps have any ideas, how this could be done better? I couldn't find any documentation for MainWindow... Cheers, Matt

[rinceWind]

I would use local for tweaking the environment variable:

use strict;
use Tk;

my @Displays = qw"node1:0 node2:0";

my $top = MainWindow->new();
my %top;

foreach (@Displays) {
  local $ENV{'DISPLAY'} = $_;

  if (eval {$top{$_} = MainWindow->new()}) {
    print "MainWindow opened on $_\n";
  } else {
    print "MainWindow failed to open on $_\n";
    $top{$_} = 0;
  }
}

MainLoop;

[matt.tovey]

Good tip - thanks.

[samizdat]

Matt, while I'm amazed at how far you're getting and I don't know what the application you're trying to deploy is, I'd humbly suggest that you think seriously about building a web front end for your application rather than trying to build and maintain a mixed-world (UNIX+WinXP+Exceed+???) application. Apache is designed to be scalable and failure-tolerant, and the web interface is/can be relatively portable. Your talent and time are better spent solving many problems for your org rather than getting caught up in one.

[zentara]

I don't really understand what you are trying to do? It would seem that if you want to open linked displays on different machines, you would get more reliable results using sockets to communicate, rather than relying on the MainWindow's event loop.

[matt.tovey]

My current version does indeed use sockets. Problems with it are:
- All nodes need to have the relevant modules and libs available (I'm using Imager). Node architectures are also not homogeneous.
- The nodes' displays need to be started from the master. I'm currently doing this with rexec and it's not as neat as I would like.
- The new version needs to support WindowsXP, which has Exceed running (X-Server), but no rexecd.

Hence my thought to just open MainWindows direct from the master.

[dk]

how about still use sockets (or pipes), but instead of establishing a communication between different programs on different boxes uning local displays, communicate between different programs on a single local box using remote displays? I mean, if your current version is somewhat like

box1:select(@handles_from_boxes) and do_logic()
box2:tk-interface.pl: connect to box1, create windows and buttons,

then I propose to do this instead:

box1: for ( @boxes) {
     $ENV{DISPLAY} = $_;
     system "tk-interface.pl";
}
select(@handles_from_boxes) and do_logic()