2009/12/18 RedShift :

Hi all

It dawned on my that lots of industries have standards and companies generally keep to them. For example slabs of aluminium have standard sizes, building materials have well defined specifications, or take electrical components: there's a huge list of standardized components. You can expect between 220 and 240 VAC from your wall socket, fuses have standard formats and ratings, 1 meter here is exactly the same as 1 meter in another country, etc... Even CD's, which have been around for decades by now, have always been created using the same format (albeit extended somewhat, over time, but a normal CD pressed now should still play in a CD player that's 20 years old).

It allows for a very competitive market where choices are made based on price, quality, availability, etc...

Why doesn't the computer business have something similar? Sure processors are interchangeable in a limited way, we use standardized RAM, standard interfaces for accessing our peripherals, etc... But not when it comes to software. Why don't we have one universal API that works on every operating system? Yes there is libc, the language C is defined "in some way", but I'm talking about stuff that would make applications 100% portable. Things like enumerating all hardware devices, configuring a network interface, drawing a window, ejecting the CD-ROM drive, getting notified about new hardware plugged in, etc... It's different on every operating system. You cannot write a driver for Linux and expect it to work on FreeBSD. You cannot write an application for windows and expect it to work on Linux. When you buy a piece of hardware you usually hope for the best that it'll work out-of-the-box including all "extra" features.

So why is that? Why hasn't someone stepped up and even try and create a universal operating system API? Is it because the computer business is still a "child" in some way, compared to other industries?

Just a thought.

Glenn

Isn't this what POSIX was, albeit quite old now, but still a standard?

For example slabs of aluminium have standard sizes

I would guess there are at least 50 different standards available for alu plates. But the difference to the computer world is, you can take any of that plates, drill a hole and mount stuff.

building materials have well defined specifications

With an emphasis on specification_s_. Sometimes it's troublesome to keep up with the changes (or complete overhauls).

So why is that? Why hasn't someone stepped up and even try and create a universal operating system API? Is it because the computer business is still a "child" in some way, compared to other industries?

Computers are still in it's childhood, i would say, too. I would say, if you create such unified API now, you would delimit the functionalty to the intersection of all hardware that is now availbale. If everybody would develop on this base, it would lead to restricted (and slow) software. (I'm no programmer, so i mighty be wrong here.) But i think we already have a big abstraction level in software nowadays. Think of Java or Python. Nobody has to write everything in assembler. I think things have to iron out themselves in the computer bussines. I read today about the computer simulation of the construction of the roof from the munich Olypmic stadium. The result of that calculations is on 600.000 punched cards. That was only 25 years ago. Now think of what long way metallurgy has gone in our history. So i think it is too early for hat. Most people even don't know, what they want from computers. Cheers, Fabian

Le vendredi 18 à 10:24, RedShift a écrit :

Things like enumerating all hardware devices, configuring a network interface, drawing a window, ejecting the CD-ROM drive, getting notified about new hardware plugged in, etc... It's different on every operating system.

Isn't it one of the goals of hal ? It does exist outside of Linux (in FreeBSD for instance: http://www.freebsd.org/gnome/docs/halfaq.html).

You cannot write a driver for Linux and expect it to work on FreeBSD. You cannot write an application for windows and expect it to work on Linux. When you buy a piece of hardware you usually hope for the best that it'll work out-of-the-box including all "extra" features.

If OS internals are to be so similar that they expose the same API to hardware drivers, where is there room for differentiation between two OSes? FreeBSD, to continue with "the unkwown giant", prides itself with writing a very well designed OS, whereas Linux (kernel & userland) does not have the same quality standards. As a result, things evolve less quickly in FreeBSD, but are usually more stable. I'm talking about the architecture of the code; when was the last time a major subsystem of FreeBSD was rewritten? (I mean, except the USB stack in 8). -- Fred

On Fri, Dec 18, 2009 at 4:24 AM, RedShift wrote:

Hi all

It dawned on my that lots of industries have standards and companies generally keep to them. For example slabs of aluminium have standard sizes, building materials have well defined specifications, or take electrical components: there's a huge list of standardized components. You can expect between 220 and 240 VAC from your wall socket, fuses have standard formats and ratings, 1 meter here is exactly the same as 1 meter in another country, etc... Even CD's, which have been around for decades by now, have always been created using the same format (albeit extended somewhat, over time, but a normal CD pressed now should still play in a CD player that's 20 years old).

It allows for a very competitive market where choices are made based on price, quality, availability, etc...

I look at it this way: an OS is a *tool,* whereas electricity, CDs and such are commodities, and need to be fungible. Tools are *not* fungible; the way you interface with a tool is very tightly coupled with the purpose of that tool, which is why you should never use a hammer to pull a screw. The abstractions OSs (and also programming languages) present represent what they're designed to do, so making a one-size-fits-all tool is worse than useless. The "desktop wars" and such arguments all commit the fallacy that OSs are a pretty shell over computer hardware, whereas they are (or should be) tools targeted at (more or less) specific solutions. -- Ryan W Sims