World: r3wp

That would only work for literal float values.

If you use a variable, you need to pass thru the registers.

The FPU registers?

In the float case, yes FPU registers.

IA32 32-bit registers can't hold C double floats (64-bit).

I've started working on a cURL binding. It can print the cURL version 
now :-)

I could call it a port from the R3 binding, but it will be more like 
a rewrite

The dynamics in Red are quite different

You mean in Red/System. Red does not exist yet.

I will improve the web site that make that difference more clear.

Red/System can hardly be compared to REBOL, they don't live at the 
same level of abstraction.

Great to know you are working on a new binding. Let me know when 
you will put it online, so I can add a link from red-lang.org.

As long as Red proper doesn't exist, I use Red to refer to Red/System 
:-)

I'm careful not to do that in publications, though

http://rebol.esperconsultancy.nl/Red-ZeroMQ-binding/doc/trunk/COPYRIGHT.txt
refers to Red :)

Ah, yes, because it will be a binding for Red, even though the binding 
is written in Red/System

Hmm, form me the AltME large font swallowed the "refers to Red :)" 
remark.

Not on my screen size

Since R3 bindings are written in C, their level is comparable to 
a binding in Red/System. But the dynamics are quite different, that's 
what I meant

looking forward to more weird bugs discovered by the curl binding 
:)

I hope they were all shaken out with the other bindings :-)

Hm, a cURL progress callback needs floats. It's pretty limiting if 
you can't get progress feedback

You should be able to pass 2 integers (if it's a 'double or just 
one if it's a 'single), then do the float->integer (by doing a * 
100 e.g.). Probably fun to code. :-)

But you said they must be passed via the FPU?

Right, if you want them in a variable. What I was proposing here 
is to split them from the stack by declaring 2 integers, so that 
the float gets stored in 2 arguments (supposing it is a double).

If cdecl functions pass them on the stack, they may be shuffled between 
them

The stack order for cdecl and stdcall (used by Red/System) is the 
same.

Almost everything needs floats... better to wait for proper implementation.

Well, it depends what are the boundaries of your "everything". From 
my own experience, except for 2D/3D and scientific calculations, 
they are rarely needed.

Well, you need it for music and graphics. That's almost everything 
for me. I was checking the FMOD binding, and it's not so bad... at 
least if you don't need to change volume or some more advanced effects:)

Is this useful? http://en.wikibooks.org/wiki/X86_Disassembly/Floating_Point_Numbers

It is another source to confirm that cdecl/stdcall both pass floating 
point arguments via the normal stack :)

Got cURL to read a website and print it:

curl-begin curl-global-all
session: curl-new-session
curl-set session curl-set-upload?  as variant! no
curl-set session curl-set-url  as variant! "http://rebol.com"
curl-do session
curl-end-session session
curl-end

Not as concise as READ "http://rebol.com"but we'll get there

Added a higher level interface that reduces that to

curl-begin
session: curl-new-session
curl-read session
curl-do session "http://rebol.com"
curl-end-session session
curl-end

Next: wxWidgets :)

They say it's a man year work to port it

In any case, I'm not going to do it

But I can upload a file to FTP now

That's direct from disk, so large files can be uploaded without using 
much memory

Can you do multiple uploads concurrently?

No, I can upload one file from disk exclusively. The binding is two 
days old

For multiple concurrent uploads before next week, use my R3 binding

Of course, if you use the 0MQ binding in combination with the cURL 
binding, you can do multiple concurrent uploads from disk only

I've implemented a callback framework and used that to implement 
sending a data block from memory, instead of direct from a file

So callbacks are working fine? No more issues?

A callback used internally by cURL is fine, yes, on Linux. We'll 
see what a progress callback does with trying to print floats

Are you using a version of Red/System that includes my deeper fix 
for callbacks from 2011-06-30 (commit log: Merge branch 'callbacks-cconv')?