Re: pre-release announcement - Linux AMD64

[Bruno H. <br...@cl...>]

Sam wrote:
> I have a syspicion that the cause of this problem is the
> "symbol_without_flags" and related things (I am just assuming that the
> problem lies in the area which I know nothing about! :-)

Exactly, that was it. It took me a look at the pushSTACK_symbolflags
macro in lispbibl.d and at make_variable_frame in control.d to see what is
wrong.

The committed patch fixes the testsuite failure on Linux/alpha, Linux/amd64,
Linux/ia64, and OSF/1 alpha.

> BTW, why do you use fake_gcv_object to keep the number of bindings?
> why not fixnum?

fixnum(x) is one more instruction than fake_gcv_object(x), so fake_gcv_object
is faster. And since the number of bindings is stored in a STACK slot that is
not scanned by the GC, it does not need to be a Lisp object.

> > You need lispbibl.h when interpreting the hexadecimal numbers as
> > objects.
>
> this is far too cryptic.
> sorry.
> this is just beyond me

You can also generate control.s (without debug info, probably) and look
there, if you happen to understand x86 assembly language.

Bruno