So Microsoft is becoming even more a lawyer company, and a bit less a technical one. If this is true, I won't bet one Microsoft share's raising again.... Did they hire SCO attorneys or laywers? :-)

whatever you'll do, you will have to work, to learn more about your Linux system, or to subcontract someone to (or buy support) to run your cluster and help your users. The point is then: do you have a budget (time and money) for that? Are you interested yourself to learn more about Linux systems (hence to spend less time on numerical codes or science)? If not, you'll need to pay someone to do the work. If yes, you need to learn a lot.

I was serious. Tail recursive call is one of the major feature lacking in the JVM. See http://blogs.sun.com/jrose/entry/tail_calls_in_the_vm http://openjdk.java.net/projects/mlvm/ And there is no way to implemnent tail-recursive call in its full generality (including tail-call optimisation to statically unknown code, e.g. thru virtual methods...) without a proper support from the JVM. This has nothing to do with addressing modes (which you don't need in a VM offering garbage collected values). And the Ocaml virtual machine bytecode does indeed show that other kind of bytecodes can be useful. JVM is not the perfect VM; it could and should be improved.

I believe what would be significant is to improve the JVM bytecode, to add some additional instructions. Tail-recursive calls is an example. There are some others. If the JVM specification was improved (and implemented by Oracle & other major JVM sources), several new languages (or language feature) could flourish. And the important part is not the Java language; it is the JVM. Better languages (Clojure, Scala) can be implemented for the JVM, and if the JVM was improved, even better languages could be experimented, all able to use the legacy of Java. Regards.

One of the main reasons to go to GCC not LLVM is that Google has several top GCC experts (nice people like Ian Taylor, Diego Novillo, Rafael Espindola and many others). On the contrary, few Google employees seem to post on LLVM mailing lists. And I see a simple reason for Google to pay leading experts on GCC. Google has a big lot of servers running Linux, and most of Linux is GCC compiled (in particular the kernel). Put it in a simpler way: if their experts are able to improve GCC just by 0.03% (or perhaps 0.05%) the saving in energy bills alone is paying the entire team of GCC experts at Google. In GCC summits some Google guys mention a single binary executable of more than 700 megabytes. I don't think that migrating such a monster program to Clang is easy, and an easy decision to take. In addition, Google has several Linux kernel experts (and they have their own patched kernel). These guys need GCC to compile that kernel. Perhaps Google also invest effort in LLVM. I don't know.

This is almost true, but I see one important exception: the exception machinery in C++ (that is the compilation of throw and catch C++ statements) is not exactly a goto (and neither is longjmp in C). And of course, any (method or function) call and return is not exactly a goto neither. Exceptions,calls and returns also change the stack pointer.

I would also notice that computed goto (i.e. the goto *p; GNU extension of C) is compiled as an indirect jump.

A more interesting concept is continuation Regards

But that would have made a better world... Microsoft succeeded not on technical grounds, but because of good lawyers.

I was suggesting a real operating system and that would have meant something better than QDOS (ie the first MSDOS).

I believe the biggest mistake was IBM using an Intel8088, instead of a Motorola68000.

Imagine for a moment what would have happened if IBM choose in the early 1980s a 32 bits processor for the first successful Personal Computer!

• no infamous 640k memory limit
• probably no MSDOS (or QDOS), and a real operating system instead
• 32 bits computing would have become mainstream a decade earlier at least!
• much less assembly written software

What is scary (or at least very sad) today is that very probably no manager would let a few brilliant programmers to develop their own system during a couple of years: in academia, publishing is much more important that working on a big software system, and in industrial R&D, one could no more work for a couple of years on a brand new software.

Current managers would look with scare at their spreadsheet and would not let that kind of things happen anymore in 2009, and I still think it is really a pity, and we could get some really innovative systems if R&D was managed differently today.

J.Pitrat does know (as I do) Godel's theorem, and did wrote some interesting pages on the relation between Godel's theorem and his view of AI.

J.Pitrat also explains how his CAIA system is in practice able to detect most of the looping situations, when it is stuck.

Regards.

If I understood correctly, the Slashdot editor is probably Bob Roberts.

He explained me that Amazon is a partner, and that he had to link to their site. I do understand that (obviously, Slashdot have some business model).

I still think the biggest mistake is at Amazon. I suggested Bob to double check the link in further submission.

I do hope that, even we that mistake, some people would read J.Pitrat's book. I really think it has interesting and provocative ideas (especially about AI and software).

Regards.

AFAIK the price is not decided by the author, but only by the publisher.

The exact link to Google's cache is this. Regards