On Tue, 16 Feb 2016 08:58:11 -0500 Clem Cole <cl...@ccc.com> wrote: > As for what started this thread. I think it is interesting that the long > term successful architectures in the market did have a excellent > compatibility stories. IBM with system 360 certainly set a high bar, and > DEC has nothing to be ashamed of, the different DEC lines, particularly > the Vax, did a great job here. In truth, probably the best of pure > compatibility story has to be Intel.
No offense, but to even suggest Intel has the best compatibility or that pure and Intel go together in the same sentence is completely over the top. From every angle I can see, Intel has about the worst track record of upward or backward compatibility and the least amount of design integrity or purity of any vendor still in business and probably in the history of modern computing. The basic architecture of S/360 from 1964 is still being used and developed today, and as Dave pointed out the very latest OS and hardware still run object code from that period. It is not uncommon for thirty plus year old software systems to be running today on the latest platform. Where is there anything even remotely comparable to that on Intel? Backwards compatibility for IBM? No. That was not a priority, quite the contrary. IBM was always a hardware company until the later years. They sold hardware that preserved your software and development investment. New OS won't run on old hardware and that is by design. > The H/L registers of the 4004 are still there ;-) All the registers from OS/360 are still there too. The difference is IBM code from those days still runs today. IBM preserved the investment by making sure all the applications and most of the systems software that was ever written will run as long as you want to run it. How many apps from the 4004 or 8080 days will even run on an old copy of Windows XP from Year 2000? Intel broke compatability at almost every step of the way, real mode to protected mode, the list goes on. All the investment people had in DOS realmode apps and development and everything prior to that basically went into the trash bin. Why the API bork between x86 and x86_64? The linkage conventions in IBM land haven't changed since OS/360. Sure, there are new and better ways to do things but the old ways still work too. Why not on Intel? Why can't I link a 32 bit executable with a 64 bit executable on Intel? Yes, I know about X32 but there isn't any OS using it and it was too little too late. Even with their billions of dollars in R&D Intel couldn't get a 64 bit chip out the door, they had to rely on AMD to do it. Why do I need two copies of libraries if I want to run 32 bit and 64 bit code in the same OS? We don't with IBM. Why does a piece of Intel code that runs 64 bit need twice the footprint of a 32 bit piece of code? It doesn't on an IBM machine. Why were the BCD instructions removed in 64 bit mode? Especially since you can't change modes in one program on Intel, and since you can't invoke a 32 bit piece of code from a 64 bit piece of code on Intel, or vice versa, how could they remove instructions that used to work? That's compatibility? I have stayed away from Intel in my career as much as I could but even I am aware of these problems. I'm sure people who know could point out a lot more. These are all problems that never even happened with IBM. IBM boxes can run 24 bit, 31 bit, and 64 bit mode apps all in one OS, with one set of libraries, no relinking. We can link 24 bit, 31 bit, and 64 bit addressing mode code in one executable. We can switch modes within one application, from one line of code to the next. Instructions didn't become obsolete or disappear as the architecture evolved and generally every instruction is available in every mode. There is no automatic footprint increase between executables running in 24 bit mode to those running in 64 bit mode. Every new IBM machine and OS was designed to preserve the investment in all the software and development skills the customer already had. In terms of architectural and implementation purity with compatability as a fundamental principle and a 52 year track record of success, IBM wrote the book. And they're still going strong. You'll need a big glue gun and a room full of Intel's best chips to come close to getting as much actual work done as you can on one IBM z12 or z13. _______________________________________________ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh
