New Compaq Servers (with Closed Source Libs)

pmsyyz sent us a news.com story that talks about the new alpha based Compaq Servers. Lots of interesting tidbits (and hardware specs to drool over) but it reveals that the compiler and libs will not be open source, although they will be cheap. Just read it- its interesting, but frusterating to read about putting Digital's excellence and Compaq's marketing together, and stirring in a PHB decision like a closed source compiler.

Comments

Compaq Compiler Technology

[Caballero]

I was speaking to one of the guys from the Compaq compiler group at Linux World. Compaq realizes that good compilers just make their hardware look better, so they do want to see that happen. The problem is that they can't (for whatever reason) release all the technology open source. They can provide information and some technology to help gcc based compilers get better, but they are afraid that it is a major overhaul and will require a couple years to become functional. The big problem is that the instruction scheduling has to change dramatically particularly for floating point. So, they are taking a two pronged approach. One is to get their compiler out in the short term for a low cost. The second is to add functionality to gcc. Eventually they could get out of selling their compiler. It's a good plan and the right thing happens over time. Alpha's rock running Linux, but our 400 Mhz DEC Unix systems out perfrom 466Mhz Linux systems. I suspect that's mainly because of the compiler and the good math libraries which they have also released for free (proprietary). - |Daryll

And what about bugs?

[raistlinne]

And what about bugs introduced into my code from bugs in the compiler? At least with Open Source compilers, I can fix that myself, if I need to. And if you don't believe me, you obviously haven't read enough changelogs. I've seen plenty of entries scattered all over the place to the effect of "added a work around for the [compiler] [version] bug on [platform]." What do we do with their closed source compilers? Never have them fixed?

So he's off topic then?

[raistlinne]

And my point was that performance isn't the only important issue. If the only response to him is some idea that having source lurking somewhere on the same hard drive is going to magically make a compiler faster, you've got to be kidding me.

Respond to this: The compiler can still be closed source and it won't affect the compiler's performance as a paperweight.

Big deal. Paperweight performance isn't the only issue for compilers. It's legitimate to point that out if I go about saying that it doesn't matter if a program is closed source because paperweight performance isn't affected.

Awesome!

[raistlinne]

Daryll,
This is awesome news. Is the guy that you heard this from a reliable source? I.e. is he high up enough to be getting this info reliably, like as a first-hand source that doesn't change its mind much?

On the math lib note, try out libffm. Sometimes it's faster than the compaq portable math libraries, though it depends greatly on the application.

On a third note, is there any chance that you'll work on glide support for the Alpha?

Thanks for the info, this makes the future of Linux/Alpha look bright indeed.

DEC compiler - open vs. closed source

[raistlinne]

As far as your argument about gcc being abandoned goes, I doubt it. I don't think that Cygnus would let itself fall behind.

As far as "This will lose Compaq the main edge it currently has over its rivals in the Alpha supplier market." What are you talking about? Compaq supplies hardware. You're not going to reverse engineer Alphas and come up with Alpha clones from an optimized compiler. Especially not on any timeframe that matters.

Do you think that Compaq is in the business of selling compilers? They make the compiler to make their hardware faster, and nearly give it away so that they can sell more hardware. Are you going to tell them that 100 different optimized compilers are going to result in compaq selling less hardware?

Not Even a Contender

[raistlinne]

> A source tarball was asked for. A binary is
> available. If I show up to a race with the parts
> to a car and a built car, why would you assume
> the pre-built car was not there?
There is validity to this argument. But let's say that the atmospheric content has changed from what the pre-built and welded-shut car comes with. The parts can be easily adjusted. The welded-shut car can't. It's close enough to have nothing when you have a car that won't count. (please give the analogy some leniency, I'm not a mechanical engineer to give a really dead-on technically correct example).

> > You are correct in that its proprietary nature
> > is unrelated to its performance. However, I
> > don't see how this gives your argument any
> > credence.

> Your first sentence is correct. The second one
> does not make sense. He was talking about
> performance which you agreed with in the first
> sentence.
No, he wasn't making a performance argument, he was making a closed source/open source argument, and using performance as one justification of that argument. Shaw is attacking the open/closed argument, which is the real argument being made.

> He stated facts. You can argue the moon is made
> of green cheese, but it does not change facts.
No one disagreed with the fact that the digital compiler will probably be faster than current egcs/gcc compilers. Many people, myself included, might take issue with the idea that the digital (compaq) compilers are better. That's what the argument is about.

egcs?

[raistlinne]

Isn't egcs currently the market leader in ANSI C++ compilers?

Jon "Maddog" Hall does not have pointy hair

[Bruce Perens]

Let's remember that the pointy-haired boss in this case might be the head of Linux International, who rather than being pointy-haired, bears a strong resemblance to a department-store santa, or Jerry Garcia on a bad day. Or at least he knows the people involved.

Bruce

Just Which libraries won't be Open?

[Bruce Perens]

The Math library (libm)...

Thanks, Jim, that makes sense. I wish the original message had been clearer. Some people got their dander up because they thought some GPL-ed library would be closed.

Of course the usual GCC libm would co-reside on the same machine.

Compiler hackers will take this as a challenge.

Bruce

Just Which libraries won't be Open?

[Bruce Perens]

The libs won't be open-source is a pretty broad statement. Which ones? The C library? Possible, but doubtful, they'd have to use their own, breaking compatibility rather badly. Some run-time library that provides ALPHA low-level facilities for the compiler? Sure.

So, make EGCS better. We've known for a long time that GCC couldn't stand up to DEC's compiler on the ALPHA.

Bruce