World: r3wp

volker: broadcasting to everything... do you think that scales well?

Gabriele - that is still nice. I did NOT expect View to be released 
at all, so ....

I thought Core comes first (but more complete), then View ...

the compositing engine is almost ready... so why not release it.

we may not be able to emulate r2 faces in july... so no compatibility... 
but other than that it should be functional

Gabriele - but, what I would not like to see is - to start with system, 
which is not flexible enough - e.g. old VID - you CAN'T add some 
things later, unless you count on them from the very beginning! Then 
docs appear, ppl start to produce scripts, and then we complain that 
we can't change it, because there is lots of dependency ;-) Please 
bear in mind, that NOW is the time for the change ...

Gabriele, did i notice i used Oberon? :)

Fastest thing i have seen on a P100. Maybe muchmore on Amiga could 
compete :)

that may mean that they were smart, it doesn't mean that broadcasting 
scales well. ;)

eg. imagine if every face got every event.

also... we have network events, system events, you could have usb 
events and many more... do you broadcast everything to everything? 
when an event generates another event, is it broadcast to everything? 
it does not seem a great model to me... :)

Gabriele - is there new event system in View? IIRC we were thinking 
along the lines of plugging some external library into rebol - libevent, 
liboop, etc.

It works for the gui. I dont step into theory about slow, if i have 
a real life example which is fast :) about everything to everything, 
that would be in a bad case: each event to 100 receivers, 100 events/sec, 
10k dispatches/sec. cpu can do 1 billion instructions. 10k instructions/event. 
most of them: i am interested? no. ~100.

100k instruction/event.

petr, yes and no. yes in the sense that events come from devices, 
no in the sense you mean (libevent). i don't see why we would need 
libevent or anything like that.

volker... oberon may be doing tricks, such as not really broadcasting 
to everything... so the fact that it is fast does not make the approach 
a good approach.

No multithreading on Oberon (back in the day - Active Oberon has 
it now). They managed multitasking through shared memory structures 
and other interesting tricks.

Because of this it was able to manage a full GUI and still be as 
fast as DOS.

Single process, no virtual memory - through other tricks, they managed 
to avoid needing it. Because of this the event broadcast, like everything 
else it did, ran by itself on the bare metal. Hence the speed.

Gabriele, liquid (dataflow) adds a level of stability to any project. 
 the fact that its lazy pays off very well so far.

the system in its current form is not meant to be easy to code  (for 
maintenance reasons).  I know many ways to change this, but before 
going to far, I had to have as stable an architecture as possible.

I have gone through 3 rewrites before hitting this sweetspot.  other 
versions might have been simpler in a way, but left the system, non 
generic and unscalable in one of many different ways.

liquid's strength lies in the fact that it is generic.  we all write 
dataflow within our applications, without realising it.  but we then 
recode each little system in its own isolated architecture.  this 
means we just loose a lot of time.

I was able to write elixir in about 40-50 hours of time and the only 
bugs it has are within the parts of code which has no dataflow.  
everytime I trust liquid and switch part of the code to it, I end 
up forgetting about that part, because it gets to be so stable.

I am as anxious as all of you and understand ALL of your qualms about 
using dataflow and liquid. 
 *complexity?
 *speed?
 *scalability?
 *difference in programming model
 * ....

The truth is, I do not have the reflex of using liquid for most of 
my coding, still, but actuall exposure and use, is forcing me to 
value its effect on my code.  this is empiric use, not advocacy. 
 If you could see just how easy it was for me to build fully bug-free 
AGG gadgets in so little time, you'd understand.


its not about just sharing data between gadgets, its about allowing 
your code to know what's going on.

and in fact, if we do add a measure of reflexivity to VID, we will 
just be redoing most of liquid, or run in the same issues, I had 
in my other prototypes, which led to this design.   ;-)

but we will not gain the advantage of having generic dataflow!

max: liquidgl is push based, isn't it? so "lazy" does not count.

nope, its all using the same core plug.  so its really pull based.

the only difference is that some (switcheable in real time, even 
set to a function) which will always want to refresh, when they are 
aware of data changes.

when you move a slider that is attached to something, is that something 
that has to pull? does it pull x times a second?

its also nice to know that your GUI is actually capable of reflecting 
data.  not just gui.  change the data:

fill data value


and don't even have to wonder IF and HOW any gadget should change.

that is push, isn't it?

 does it pull x times a second?

that is up to you to decide, really.

my decision is not to pull at all, if something changes the change 
is pushed to everything that needs to know.

at a high level yes.  but in reality, no.  if you have some gadgets 
or other parts of dependent systems which do not "need" the value, 
it might not cause any refresh.

that doesn't scale well.

(tested)

that really is just like an action.

exactly, just an action, or to say it differently, just an event.

but you are only seeing all of this as a VID thing.

dataflow has nothing to do with GUI.

we only need it for the gui - that's exactly my point

its something the whole language should have.

nope.

noone stops people from including liquid

cause you do not see its value if it where built in.

;-)

especially since its such a small kernel !  (not the current implementation, 
which has a few prototypical stuff left)

liquid is small, but the functionality we needs comes free with my 
vid design, so you can't win ;)

you see, if we had a dataflow datatype, we would not even need to 
talk about "do we add this to VID"  people could just set values 
to attributes which are DF based.