[fpc-pascal] Interresting discussion

[Michael Van Canneyt]

On Wed, 22 Feb 2006, Jonas Maebe wrote:

>
> On 22 feb 2006, at 11:26, Michael Van Canneyt wrote:
>
>>> I don't think there are many host OS'es out there which e.g. only run 
>>> signed applications. The fact is that for some purposes, none of the 
>>> current OS'es "does its job correctly" in that sense. Just use the right 
>>> tool for the job, I personally don't understand all this hostility against 
>>> managed environments (except as a reaction against claims that they are 
>>> always much better than unmanaged environments, but overdoing it in the 
>>> other direction isn't going to get the discussion anywhere).
>> 
>> Well, the very idea of a program running in a managed environment which by
>> itself is also a managed environment seems like a waste of resources.
>
> The environments have different management capabilities. The OS manages the 
> hardware and provides an interface between the hardware and the software. 
> These managed environments pure manage software. They don't have a 
> compatibility scourge to deal with, and therefore can break paradigms which 
> are widely in use but which are inherently insecure (such as pointers) or 
> which make programs unverifiable (self-modifying code, data-in-code and 
> code-in data, code flow which is impossible to determine etc).

This is all true, but none of these cannot be handled by the OS.
No-one said that an OS has to apply the same rules to user-space
programs and to device drivers. In fact, most don't...

>
>> Secondly, too many links in the chain makes it easier for the chain to 
>> break...
>
> I don't think a managed environment is easier to break than in case you try 
> to stuff all those things in an already existing environments.

It's not about breaking in.
But 2 cooperating programs break down easier than 1.

>
>> The advantage of running 'signed' applications also eludes me. Even so,
>> provided you really want that, you could easily integrate that in the OS,
>> without having a new managed environment...
>
> Then you have to integrate it in every OS, bolting it on on structures which 
> were never designed for such things.

Now you must rewrite your engine on each OS. To do that, you must build some
kind of OS-asbtraction layer if you don't want to re-implement from scratch,
so, it's the same thing as far as I can see...

Anyway, it's not really worth the discussion. It exists, people use it...

Michael.