On Sun, Aug 24, 2014 at 12:07 AM, Richard Stallman <rms-mXXj517/zsQ@public.gmane.org> wrote:
 
    The research is all done. We just need to implement it.

I would describe this as a domain for research, because it is
far from clear whether it could possibly work in practice;
if it could, it is not obvious how to make it work.

It's not like writing an assembler, where we know it can work.

We don't actually know the assembler _does_ work. That's what I tried to say. I didn't want to over-emphasise it, because I thought people might already be bored of me banging on about it But apparently not! The problem is the semantics of the whole GNU toolchain are inaccessible because they depend 100% on the proprietary non-free tools that were used to build them.

This is important, so let me make a statement:

     There is not one single line of GNU source code, the meaning
      of which does not depend entirely on non-free proprietary
      software.

The source code will not tell you what the software actually does. See Thompson's 1984 Turing Award Speech http://www.ece.cmu.edu/~ganger/712.fall02/papers/p761-thompson.pdf for the details. I quote:

     "The moral is obvious. You can't trust code that you did not
       totally create yourself. (Especially code from companies
       that employ people like me.) "

He works for Google now, doesn't he :-)

The trick Thompson demonstrates is what I call a _syntactic fixed-point_. He basically makes a Y combinator of the C compiler, so that it takes in its own source and outputs it again, but that does not mean that the source it outputs is the source you read when you look at the source-distribution.

The solution is to establish a _semantic_ fixedpoint. But to do that we need to be able to generate arbitrarily many essentially different working implementations of the compiler, and make sure that the code it generates when it compiles itself is the same, whatever the particular implementation. So if we see that the version that we compiled into MS Word BASIC produces the same binary executable as the JavaScript and COBOL versions then we can have some more confidence that the semantics of the GNU toolchain haven't been hijacked, ... yet.

Ian