World: r3wp

http://www.rebol.net/cgi-bin/rambo.r?id=3314&


Memory usage climbs, but there is no longer any crash on View 1.3.1

i: 0 until [
	attempt [
		view layout [
		button feel [
		redraw: func [face action pos][
		;if action = 'draw [show face] ; memory use climbs very high
		if action = 'show [show face]  ; memory use looks quite stable
		]
		]
		]
	]
	1000 <= i: i + 1
]
unview

hmm, need to check all tickets with undetermined importance. otherwise 
they get overlooked.

view/new win: layout [] win/changes: reduce [:+] show win  ; crash

anton: confirmed

Thanks Henrik. You can also try putting different functions, natives, 
ops, in there for the same result. (I haven't tried them all, though.)

none does not cause a crash, though.

I have searched for "changes" in rambo and I did not find a similar 
ticket.

add it :)

OK done. :)

Hi, #3545 is marked as High importance. Actually, it is underestimated, 
because such problems are critical even for pre-alpha Rebol programs 
e.g. when trying to send keep-alive type messages to a port using 
Romano's ATCP scheme. I  wonder what is an optimal work-around for 
the current (Core 2.6 - based) scripts? (see #3465 too, marked - 
underestimatedly - as Medium importance) .

(I know, that we can use Timers for View scripts, but that does not 
solve the Core case)

timeouts are a problem for normal (non-async) core, but should work 
fine on async core.

the "solution" (a trick, with its own problems...) i use for the 
detective is http://www.colellachiara.com/soft/Libs/timers.r

anyway - this is something we *need* for /services... so i guess 
there will be a real solution very soon.

just a few weeks ago I wrote that with the new rebcmd for Win my 
problems with odbc and catch were gone.

Now I encountered a "Crash, should not happen" with a script doing 
not more than receiving data from tcp sockets, parsing the data, 
writing to files and sending a response back.

Some almost identical scripts were working with the old cmd for months 
without interruption.

Can you isolate the problem?

I will try, but I am not very optimistic

It's reproducible
Crash (should not happen) invalid string width 16: type 41

Now I get the crash every second day

Isolation is key to fix the problem. If you can't isolated it, please 
post the full script.

I can see where the crash happens.

It happens in or just before an awake function, when the peer closes 
the socket.


On_Data: func [port] [action-time: now con: port msg: copy: copy 
port 
	if any [none? msg empty? msg] [print ... 

and the print is not evaluated.

The script alone will probably not help you, as you will need the 
data too.

Maybe there is data also coming from a second port, as I have sometimes 
seen a crosstalk of events, if there is heavy traffic on the pc.

does it have a GUI too?

no, just a plain console application accepting tcp socket connections 
and data

is it doing anything woth noting during the awake? creating objects, 
or something like that?

When accepting a new connection, a new object is created.

After accepting new data there is mainly parsing, block and string 
creating, writing to the console and to files and sending a reply 
back

did you check any other versions of command? (this is on the latest 
version, right?)

I did not see this with rebcmd2..55.3.1.alpha, but maybe I did not 
test with enough traffic from more than one sender at the same time.

I just remember that some of the interim releases were not stable 
enough regarding read/lines, so I was waiting for a (near) final 
release for intensive testing.


Now I am checking, if different behaviour of the sender has some 
influence.

Normally at least one sender establishes a permanent connection, 
but most sender open a connection, do their communication  and close 
the socket at the end.

That's the reason, why I know, that the crash happens after closing 
the socket.  The last message was always from one of the senders 
with a one time connection. He got his answer, but the receiver did 
not print the message, that the remote socket was closed.

Maybe it has to do with timing and or GC problems, as the crash happens 
mostly at a time, when there is more traffic, but never when there 
is always  continuous communication. When I play back the communication 
without interruption, there is no crash.

my guess is that this will be very hard to figure out. please post 
all the info you can in a rambo ticket, then let's hope Carl will 
guess where the problem is. :)

sqlab it will be good if you can do some pkt traces. Then later we 
can do some pkt replays to repeat the problem, and track it. We will 
need basically the server code. The client can be performed by the 
replay utility.

Server code and pkt capture. (pcap format is preferred this is done 
with ethereal).

Jaime: That will not be possible, as the data are subject to privacy.

i will try to narrow down the server and see if i can produce a crash 
with anonimized data.
sorry.(

Right that does it. Feel/over is not always doing its documented 
job. "The over function is called whenever the mouse pointer passes 
over or off of a face."  (quote from http://www.rebol.com/docs/view-system.html
   )

view make face [offset: 100x100 pane: make face [feel: probe make 
face/feel [over: func [face action pos][probe action]]]]

The above example works but no longer works if you click and drag 
off the face.

sqlab "anonimized data could work". We need to isolate the problem, 
and we all want a very robust rebol.

So I will submit a bug report for that.

Also, I would like modifications to related ticket #3112 "How feel/over 
handles face vs subfaces":

- Type: should be changed from "Issue" -> "Bug"  (because I think 
it will be ignored otherwise)

- Importance: should be changed from "Undetermined" -> "Medium" at 
least.

Submitted the bug report.

click and drag off
 -> that's the away action for the engage function.

OK, so the above document needs to be changed to say something like: 
"... and not in the middle of a drag, whose events can be trapped 
with the engage function..."  ?

Thanks Gabriele, I am going to investigate some more. Perhaps you 
can add a comment to the report to change the doc instead (or at 
least).

In the works Anton. Thanks!

Thank you, Gregg.

well, the new rebcommand finally crashed on me too though took much 
longer than previously.  Usage jumped to 50% as before and stayed 
there.  I guess on a non hyperthreaded cpu, it would have jumped 
to 100%.

It's looking like I need to ditch odbc and move to using doc's mysql 
instead.

Thanks again Gabriele. Even though I've read that document several 
times over the years, and styled many widgets, somehow I kept getting 
confused by OVER. It was that particular phrase "The over function 
is called whenever the mouse pointer passes over or off of a face." 
Every time I re-read it I believed it :)

Anyway, now the doc is fixed I feel a lot better.

Gabriele, #3112 shows a different issue and should not be dismissed. 
(I admit my last bug report was a little bit crazy, and I'm sorry 
about that.)

Ach! too hasty again. That issue can be solved by using OVER in combination 
with DETECT.  That misunderstanding was caused by the same wording 
in the documentation of OVER. I will make an example for that too.