World: r3wp

maybe you can get what you want with iptables or ipfw (on os x it's 
ipfw)

So you think that using something like:
...
uniserve/boot/no-wait
forever [
	wait 0:0:1
	foreach-listener-write-enough-data-to-play-1s
]
...
is not good approach?

I think I cannot do it from 'on-write as this would require to have 
separate input for each listener, which I don't want.

What happens is I write on port which does not finished previous 
write?

Will... I want to make a private mp3 stream server to play music 
on a local network. So it must be solved on the server side... read 
only enough sound data from disk to play and distribute it to listeners. 
I have already the mp3 parser to get for example enough data to play 
during specified interval of time. Now it's just how to distribute 
it and don't send more data than is necessary  for the listeners.

I don't remember if WAIT is processing network events when called 
with a time! value. If it's not processing events, your writes would 
have to be in blocking mode, so sending data one after another.

I wouldn't think that wait would be working like that.  I think you 
can only use wait [] to process the wait-list.

UniServe would be then useless, because it couldn't work in async 
mode.

it should be wait [ 0:0:1]

I will test it anyway

Wait [ 0:0:1] won't work as you expect. You don't have any guaranty 
that it will exit from the loop every seconds. It depends on network 
activity. The time! value indicates a timeout duration from last 
network event.

Which events? Even the on-write events? Not being called in precise 
interval would not be a problem. I think.

For example, if you have, on average, one network event (port opening, 
packet coming, packet to be sent (async write),...) every 0.5sec, 
it's enough to delay the exit from your WAIT [ 0:0:1 ] for several 
seconds or even minutes.

As long as you get any event happening before a timeout occurs, you'll 
stay in the WAIT event loop.

what the prob ? if you do a wait [0.001] it should process one event 
at a time.

maybe wait [0] works too

Ah.. I see Doc, that is a problem.. so it looks I will have to write 
the data from 'on-write. Just will have to find out, how to make 
it synced, because I would like to have all listeners play the sound 
in the same moment. If it's possible.

WAIT [0] should return just after the first event is processed, but 
that's not the point. It's about being able to send the same amount 
of data every seconds.

how can you do it in terms of single process?

As long as you get any event happening before a timeout occurs, you'll 
stay in the WAIT event loop.

 - Doc - is it really correct? I am far from being guru here, but 
 it sounds strange - that would mean, that as far as there are events 
 coming, you are not allowed to quit wait, no? I think that 'wait 
 waits for either the event, or an timeout to occur. If there is any 
 kind of event on port in wait-list, 'wait return. It can either return 
 with first port with event, or, when using /all refinement, with 
 block of all ports, which have event available on them at the time 
 of return ...

but - you surely know what you are talking about, so I have to be 
wrong :-)

WAIT has many different behaviours depending on the passed argument.

When called with [ ] or [ integer! ], WAIT will process all ports 
stored in system/ports/wait-list. In that case, WAIT will block until 
:

- a timeout is specified and no more events are received for that 
timeout period
- an event returns a TRUE value.

Not sure about the WAIT [port1 port2 ...] case. I guess that it will 
process all events from system/ports/wait-list but will exit on first 
event received by port1, port2,...

just tried to add this before cheyenne starts:
do-events: does [
	forever [wait [0:0:1] print now/time/precise]
]
And from second console run:
loop 1000 [read http://localhost]


the result is, that it prints the time in every second while it serves 
the requests.

In the same time it was serving MP3 from modified micro-http service 
using something like:
    on-received: func [data][
		write-client rejoin[
			{ICY 200 OK} crlf
			{icy-name: LocalRadio} crlf
			{icy-genre: whatever} crlf
			{icy-url: http://127.0.0.1:811} crlf
			{content-type: audio/mpeg} crlf
			{icy-pub: 1} crlf
			{icy-br: 128} crlf 
			crlf
		]
		write-client path-to-mp3-file
	        close-client
    ]

And running my RebolBB with responses <16ms

So Doc was wrong, it's unusual...

Oldes: thanks for taking the time to test it. The correct behaviour 
of WAIT is the one you've described. I'm glad you've corrected me 
on that one. I'm pretty sure that I've run such tests with different 
results when working on UniServe five years ago. Maybe something 
changed in REBOL kernel since then or maybe my test script was just 
flawed. Nice to see that there's still something to learn from R2.

wow, audio streaming in rebol and uniserve .. cool

I need to learn more about thay!

thay -> that

on-write is dissabled in the recent versions. Why?

What makes you think that?

; --- temporary deactivated
			;on-write port

0.9.17	21/12/2004 {
			o Added 'on-write-done event.
			o 'on-write call in 'write-peer temporaly deactivated.
			o bugfix in 'open-port, custom port-id now correctly used.
		}

Right, that was a workaround for a packet sending issue, but never 
had to reactivate it.

It was useful on client side to kickstart first packet sent to a 
server (kind of fast shortcut to avoid a round in event loop before 
sending the first packet), but I had more issues than benefits from 
it, so I left the code deactivated in case I would need it in future, 
but it looks like I could remove those lines. Do you have a need 
for it?

Btw, 'on-write is not disabled in UniServe, it's just no more directly 
called in that specific case.

probably no..  just found it as you told me, that I should control 
the data from it.

Right, controling bandwidth requires changing UniServe's internals, 
mostly in 'on-write handler.

The balloon goes up! I want to move Rugby's transport layer to Uniserve, 
or even better, Cheyenne. (http tunneling). For this to really work 
I only need to know if you have primitives for timers in the event 
loop (inside the 'wait): do-every time! [ code ]  do-after time! 
[code]

The rest will be porting transport and state machines on the server, 
but as Rugby already had a CGI interface it hould be simle to use 
the server with Cheyenne.

That feature is waiting for weeks to be implemented (need it for 
adding a mail relay agent to Cheyenne and for cron-like jobs scheduling), 
I'll give it some time this weekend.

Is this convergence ?  :)

No, this will be world domination ;-)

Rugby to the rescue! I always wanted general multiplexing architecture 
being part of REBOL natively, but then I also wanted simplicity of 
Rugby. Now I will get both? :-)

Maarten - what about Chord concept, is it still valid? Or is there 
anything better out there worth implementing?

btw - what is missing in R3 to start porting of Uniserve to R3? Higher 
level protocols? Or tasks?

Stability maybe? ;-)

First ugby on Uniserve. Then Chord on top of that.... Chord on Rugby 
is already working.

Nenad, this may help you: http://www.colellachiara.com/soft/Libs/timers.r