On Thursday 30 September 2004 16:37, Ted Zlatanov wrote: > On Thu, 30 Sep 2004, [EMAIL PROTECTED] wrote: > > Another option would be to have an additional input channel to ion which > > is set to a fifo on the machine, assuming the OS has got such a notion. > > (I know some window managers implement something like this.) Writing to > > the fifo is then equivalent to sending input to the window manager, > > except it can be done by any process. > > Why would you want to cripple the network socket concept in favor of a > FIFO? Typically you'd use a FIFO where performance is a concern, and > I don't think any of the operations I proposed would be > resource-intensive. Also, a local network connection is very fast - > depending on the particulars of the OS, it may be as fast as a FIFO.
Security. Security security security. If it's a real network connection, you have to implement some kind of access control or nasty third parties can take over your X session. That's quite a lot of code and is tricky to get right, and none of it belongs in a window manager (IMO). If it's a fifo, it's just a matter of setting it to be writable only by the user who started ion and the kernel does it all for you. If you need it, you can get network transparency by using an external process to handle the authentication, or you could just do "ssh [EMAIL PROTECTED] 'cat > /path/to/fifo'". -- +- David Given --McQ-+ "SCSI is *not* magic. There are *fundamental | [EMAIL PROTECTED] | technical reasons* why it is necessary to sacriface | ([EMAIL PROTECTED]) | a young goat to your SCSI chain now and then." --- +- www.cowlark.com --+ John woods
