[exprla-devel] Re: [XPL] Instruction Set Architecture in XML?

[reid_spencer <ras...@re...>]

--- In xpl-dev@y..., Jonathan Burns <saski@w...> wrote:
Richard Anthony Hein wrote:
 Ok ... [snivel] ... I WILL then!  In fact I already started!  [snivel
... snort] </insanity> <!-- if you can't find the start tag, it's
because it was inserted into an earlier message on a totally unrelated
newsgroup 4 years ago --><backtowork>


Richard ... lad ... are you all right?

You give the strongest impression of someone as disoriented with 
machine
code
as I am with the Hytime Spec ( yes! the dreaded 10744, and thank you 
for
setting
me onto the Groves path ).

Listen, did you get my long post a few days ago on what and when to
compile in XPL?
I did my best to explain that time and memory savings from compiling 
XML
processing
were likely to be lost in the larger inefficiencies of the system,
unless heroic
measures were taken to optimize the latter.

Despite this, I have a sense that taking XSLT processing down to the
metal might
yield benefits. Whether it's so, would depend on the profile of the
processing.
If we could demonstrate that there are critical loops, in which the 
same
thing
happens to a whole lot of data, and if these loops typically take up a
lot of
the time it takes to do useful things, then yes.

That's not quite what you were talking about, though.

You were talking about making up descriptions of ISAs, in XML, and 
using
these
to generate the XPL compiler for each architecture.

Not a stupid idea. But consider this.

We're talking about compiling the XPL processing - i.e. the operations
on the
parsed data - the semantic stuff - the actual nodes and attributes. In
other
words, the operational semantics.

So what defines the operational semantics?

Well, the XPL specification that we use. Which could be
XPath/XSLT/XPointer/Xcetera.
That's what we want to map to the target ISA.

Big job, eh? Lots of operations in XPath - lots of options for
pattern-matching,
tree-walking, etc. And more of them in XSLT.

OR, the base specification could be a set of Architectural Forms - 
i.e.
property sets.
And the XPath/XPointer/XSLT operations could be written, once and for
all or nearly,
in terms of them.

Then what we would want is a mapping of a sufficient set of property 
set
operations
into the target ISA - plus a mapping of the relatively simple
programming constructions
which express each X-type operation in terms of them.

Little job.

And that is one reason why I am intensely interested in the Groves
stuff, the property
sets. I have not yet discovered any material to the effect: X-whatever
Implemented In
Groves.

If it hasn't been done, it should be. I suspect I could do it, if I 
made
it The Priority.
IT WOULD ELIMINATE A VERITABLE ARMY OF BUGS FROM XPL AT THE OUTSET.

Property Set Architecture -> Instruction Set Architecture

is where I think the value lies.


Jonathan

"Ask anything!"  he muttered, as he spat
a small blue insect whirring into the gauze.
"I would advise stilts for the quagmire
Camels for the snowy hills
And any survivors
Their debts I will certainly pay.
There's always a way."

- The Incredible String Band
--- End forwarded message ---