Linus Torvalds (torvalds.delete@this.linux-foundation.org) on April 6, 2017 2:39 pm wrote:
> RichardC (tich.delete@this.pobox.com) on April 6, 2017 1:49 pm wrote:
> >
> > The evidence, and Occam's Razor, suggests that RISC'y ISAs did offer a compelling advantage
> > in the 1980s.
>
> So here's the counter-argument: looking at end results, the performance and
> simplicity advantages (which I agree with) simply aren't the full picture.
>
> It's just a very narrow view.
>
> The RISC proponents look at one thing (performance), and people continue to pooh-pooh
> (30 years later) the designers that looked at another thing (compatibility).
>
> But if you actually look at what was successful?
>
> It turns out that compatibility simply seems to end up being a hugely important issue.
>
> So here's the deal: if you idolize that RISC designers that you think made the
> right decision, why do you dismiss all those other designers that made another
> decision that we already know to have been at least as important?
>
> That's the BS I'm calling the RISC proponents out on. That's the religion I pointed
> to originally: the incorrect and completely BS idea that "you can just recompile".
>
> See what I'm trying to say? Performance often isn't even the primary issue. The performance
> of a processor that doesn't run your load is entirely immaterial. How easy to was to design
> is also entirely immaterial. How "clean" it is is - wait for it - entirely immaterial.
>
> This is what annoys me in these idiotic "RISC was science" crap. It uses a microscope to look
> at one single side of the real world, and then says that that is all, and that is "science".
>
> That's bullshit. That's not "science". That's just a single small facet of
> reality, and it's actively ignoring all those other facets of reality.
>
> And in many other ways, the RISC people were complete failures. The belief that you can
> just recompile turns out to have been bogus. The belief that register windows are a good
> idea turns out to have been garbage. The belief that you can and should simplify things
> to the point that you expose pipeline details turns out to have been pure garbage.
>
> And then those same people who try to make excuses for all that incredible crap are pooh-pooing the
> people who made architectures that were backwards compatible - very much warts and all - and actually
> made a much bigger difference both commercially and to make computers available to the masses?
>
> Please. Stop putting RISC on a pedestal. Stop calling one particular facet "science",
> when there's a lot of evidence that that other component is at least as important.
>
> Because on the whole, the actual evidence is that RISC matters a hell of a lot less than having lots of software
> infrastructure and actively trying to keep that infrastructure healthy by being compatible with it.
>
> So why aren't you singing the praises of the developers who actually took reality into account, and
> designed those complex monsters that kept running old software even as they introduced new features?
>
> Isn't a big part of "science" that whole thing about how it actually tries to account for "reality"?
>
> Linus

You might be interested in

http://www.computerconservationsociety.org/resurrection/res16.htm#e

which describes how ICT had to react in the 1960's when they discovered IBM was going to come out with a compatible series of machines and how they found out that users wanted to be able to run the same program on different sizes of computer. Yep compatibility is very important.