OpenEuphoria

Re: Requests for Eu 2.5

• Posted by Pete Lomax <petelomax at blueyonder.co.uk> Jan 17, 2004
• 487 views

On Fri, 16 Jan 2004 13:00:47 -0500, Robert Craig
<rds at RapidEuphoria.com> wrote:

<snip>
>The Intermediate Language (IL) would take a fair bit of
>effort to document,
Yes. Probably weeks or even months, to do it well.
> and once people start coding to it,
>it would become difficult to change. 

I'd prefer to see a fairly strongly worded warning along the lines of 

"The IR specification is NOT intended to be backwards compatible. Any
hand coded IR may need to be manually updated for future releases. Any
custom front-end generating IR may likewise need to be updated for
future releases. Backwards compatibility is restricted to the Euphoria
programming language, not the Intermediate Representation."...

and maybe:

"Where possible, inefficient backward compatibility may be provided,
however anyone intending to use IR directly is clearly concerned with
performance benefits, and should therefore plan additional effort to
propagate such to future releases, manually."

You should perhaps instead consider releasing 2.6 and later with an
old 2.5 back-end, which is fired up if and only if the version number
hard-wired into the IR indicates you have to. There may be some issues
with shared data and threading/forking style concepts that would need
to be resolved for that to work.


Let's not *plan* to repeat the mistake that is the 80x86 architecture!
(what odds would I have got 20 years ago that the most popular cpu
would run at 3GHz with only eight registers, and that only one of
those could be used for certain opcodes, and that values cannot be
transferred between the alu and the fpu except via memory, etc?)


>We'll have to see
>how many people want to make alternate front-ends or back-ends.
"Want to"? alot, judging by past posts. "Will actually" is the
question 

>In lieu of documentation, people could just examine my source.
>In particular, they could use a back-end written in Euphoria
>(if I get around to it) as a model of how the 160 or so
>IL op codes are supposed to work.
That would be a reasonable compromise. The question is, would that
make the IR specification strictly copyrighted, which I feel would be
counter-productive in the long term?

Pete