In article <[EMAIL PROTECTED]>,
Pavel Machek <[EMAIL PROTECTED]> wrote:
>Hi!
>
>> The Linux answer is the same as the MS answer (and I think in both cases
>> probably for techncially sound reasons). We get cleaner maintainable code
>> by chosing to do things right. Microkernel and isolation glue API's have a
>> very bad historical record (eg Mach) for general purpose systems.
>
>Well -- usb is not performace critical (at 12MBps at least), so many
>of arguments just don't apply here. I'd like to see drivers like acm.c
>in userland...
480Mb/s. Next year. Let's not design for the current crop, and then be
stuck with a bad design.
Besides, note Alan's comment on "cleaner" and "maintainable". Black-box
isolated interfaces _always_ end up being crap.
I wonder why they still teach the isolationism ideas at computer science
classes. It doesn't work, never has, and probably never will. There's
not a _single_ successful case of isolating API's that I've ever heard
of that works with more than one vendor.
Linus
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
