Linux announcement may mark the end of Palm OS (Sunrise author)

[Dianne Hackborn]

Originally Posted by Morpheus

I wonder if the Cobalt was we've seen it today through the PalmSource emulator will look the same once it is used in actual PDA and Smartphone devices.

It will look exactly the same. The current simulator we have available is almost the entire Palm OS Cobalt, recompiled for Windows. Even most of the kernel is simulated, though some of the lower-level parts are different -- for example Palm OS threads are simulated using native Windows threads.

[Dianne Hackborn]

Originally Posted by doctorow

I don't agree with the absolute seperation of kernel (Linux) vs. user land (Cobalt). The ones goes along with the other. Due to the new features introduced by having a Linux kernel (for instance the superb TCPIP stack), Cobalt would most likely also introduce new applications such as an SSH Client, etc.

This is actually a good example against what you are saying -- the Palm OS Cobalt APIs for dealing with networking are basically the traditional socket APIs, which would easily sit on top of Linux with no changes. Switching from one kernel to another would basically have no impact on applications using them.

There is nothing special about Linux that makes it magic to implement an SSH client. Cobalt has a good, modern IO system (using Streams), which can easily support such a thing. The main advantage you'd get in this area moving to Linux is that it would be a little easier to port some existing SSH client... except you still have to deal with UI code, which is really where most of the work is.

Originally Posted by doctorow

Also, hacker, you quoted FreeBSD. Remember that kernel and user land are very tightly integrated in this case. Unlike it is the case with Linux, everytime you update the kernel sources and recompile them, you should also do the same with the user land files.

I am not all that familiar with FreeBSD, but I find this statement surprising. When a new version of Mac OS X comes out (with a new version of its kernel), you don't have to recompile your existing applications, do you?

In fact, user space never directly calls into the kernel. Instead you supply shared libraries (libc being the prime example) that do the kernel calls for you, which give to user space an API that can be maintained across changes to the low-level kernel traps.

If your kernel and user land are "very tightly integrated" like this, I would say you've got a very basic flaw in your system design. It isn't a problem in Windows, Mac OS, the current Palm OS, etc., etc.

Originally Posted by doctorow

Hence I am *convinced* Cobalt, if it ever returns from vaporware, would look different if it was built on top of Linux.

I guess this bears repeating: "Palm OS for Linux" does not replace the current version of Cobalt. We expect devices to ship with Cobalt 6.1 this year.

As for things looking different, there is nothing about changing the kernel that says this. For example, you can run X Windows on Linux and FreeBSD, and switching between them doesn't change how it looks or operates. The "looK" is really one of the most trivial aspects in this.

[doctorow]

Mrs Hackborn, first let me thank you for taking your time to answer our speculations. This is highly appreciated, and I am sure not just by me!

On FreeBSD: FreeBSD is developed as an integrated system, with a kernel matching userland tools. For instance, one should not run a kernel compiled for FreeBSD 5.3 RELEASE on a CURRENT machine. The kernel and all userland utilities are meant to be upgraded simultaneously, and must be kept synchronized. While Linux users are usually forced to acknowledge this good system administration practice when they upgrade major versions of their kernel (e.g., 2.2 to 2.4, or 2.4 to 2.6), they often maintain the same userland across minor kernel versions. FreeBSD strongly encourages users to always keep the userland and kernel in sync. So I wouldn't consider this a "basic flaw", as you referred to it, but actually an instrinsic feature of the underlying OS system.

Again, I appreciate your elaborations!

[hacker]

Originally Posted by doctorow

The ones goes along with the other. Due to the new features introduced by having a Linux kernel (for instance the superb TCPIP stack), Cobalt would most likely also introduce new applications such as an SSH Client, etc.

There are already two excellent ssh clients for Palm (and I use them both heavily, for remote administration of boxes, when I'm offsite and a client demands my attention):

pssh



TuSSH

[hacker]

Originally Posted by doctorow

On FreeBSD: FreeBSD is developed as an integrated system, with a kernel matching userland tools. For instance, one should not run a kernel compiled for FreeBSD 5.3 RELEASE on a CURRENT machine. The kernel and all userland utilities are meant to be upgraded simultaneously, and must be kept synchronized.

This isn't exactly true. You can (and many do), upgrade their kernels only without building the world target every time to rebuild userland. You should keep them in lockstep, but there are very specific cases where you never want to upgrade userland when you upgrade kernels, lest you risk stability of your existing userland.

[Alexander Turcic]

Looks nice indeed. I haven't come up with a decent SSH client for PPC yet. I tried Putty for Pocket PC, but it doesn't work very well and is not VGA-aware ;(