Tested using:
//////////////////////////////////////////////////////////
default
{
touch_start(integer total_number)

}

state other
{
state_entry()

state_exit()

}
///////////////////////////////////////////////////////

Sample results, omitting "."s (when compiled in LSL):
[23:01] Object: state_entry: 5.414494
[23:01] Object: state_exit: 7.305904
[23:02] Object: state_entry: 5.415421
[23:02] Object: state_exit: 7.320724
[23:02] Object: state_entry: 5.415062
[23:02] Object: state_exit: 7.317482
[23:03] Object: state_entry: 5.420517
[23:03] Object: state_exit: 7.304108
[23:03] Object: state_entry: 5.409958
[23:03] Object: state_exit: 7.302544

Sample results, omitting "."s (when compiled in mono):
[23:04] Object: state_entry: 5.452568
[23:05] Object: state_exit: 5.278564
[23:05] Object: state_entry: 5.430317
[23:05] Object: state_exit: 5.251865
[23:05] Object: state_entry: 5.409817
[23:05] Object: state_exit: 5.252609
[23:05] Object: state_entry: 5.408588
[23:05] Object: state_exit: 5.252937
[23:05] Object: state_entry: 5.410522
[23:05] Object: state_exit: 5.255127

say(string prefix, integer number) {
while (--number >= 0) { llOwnerSay(prefix + " " + (string)llGetTime()); }
}

default
{
state_entry()

touch_start(integer number) { say("Touched", 1); state next; }

state_exit()
{ say("Exited", 5); }
}
state next {
state_entry()
{ llOwnerSay("state: next"); say("Entered", 5); }

touch_start(integer number) { say("Touched", 1); state default; }

state_exit()
{ say("Exited", 5); } }
}

...produces the following results....
[13:46] Object: state: default
[13:46] Object: Entered 0.001223
[13:46] Object: Entered 0.024586
[13:46] Object: Entered 0.024586
[13:46] Object: Entered 0.044923
[13:46] Object: Entered 0.067178
[13:46] Object: Touched 4.541522
[13:46] Object: Exited 5.187782
[13:46] Object: Exited 5.853748
[13:46] Object: Exited 6.522691
[13:46] Object: Exited 7.190498
[13:47] Object: Exited 7.856604
[13:47] Object: state: next
[13:47] Object: Entered 8.392305
[13:47] Object: Entered 8.412863
[13:47] Object: Entered 8.412863
[13:47] Object: Entered 8.436759
[13:47] Object: Entered 8.457339
[13:47] Object: Touched 10.840280
[13:47] Object: Exited 11.484820
[13:47] Object: Exited 12.151639
[13:47] Object: Exited 12.819146
[13:47] Object: Exited 13.509143
[13:47] Object: Exited 14.178793
[13:47] Object: state: default
[13:47] Object: Entered 14.710886
[13:47] Object: Entered 14.733949
[13:47] Object: Entered 14.733949
[13:47] Object: Entered 14.755330
[13:47] Object: Entered 14.778624

It's good to know that Linden Lab now knows the cause, but the workaround of removing critical code from state_exit() is prohibitively expensive in terms of testing and deployment to over nineteen thousand customers.

Please, please, please, please look into this issue as soon as possible.

integer ix;
integer count = 100;

for (ix = 0; ix < count; ++ix) {
}

Code runs about one hundred times slower in state_exit handler. This did NOT happen prior to 1.24 servers.

This problem needs immediate attention. The problem happens for LSL or Mono-compiled code.

Possibly related? In any case, both issues deal with state changes and performance.

Thank you very much.

"I think what's happening here is that state changes are being clamped to 1 per frame."

Clearly this is not the issue shown here. Anything done while in state_exit takes longer. Below is a test script and results.

If you do not want to read that, the short story is:

A state transition takes 0.02 seconds (?!)
Every operation while in state_exit() takes a very long time to run. I am not bothering to count the operations, but each iteration takes about 0.41 seconds in the sim I was in when called in state_exit() and less than 0.000001 seconds in state_entry().

I wonder if a moth catcher has fallen into state_exit()?

float entryEntry;
float entryExit;
float exitEntry;
float exitExit;
float defaultStart;
float insaneStart;
integer first = 1;
integer N = 2;

integer
test()
{
integer i;
integer k = 0;

for (i = 0; i < N; ++i)
k += i;
return k;
}

default
{
state_entry()
{
defaultStart = llGetTime();
if (first) { first = 0; insaneStart = llGetTime(); state insane; }

llOwnerSay("\nN is " + (string)N +
"\nLast Entry: " + (string)(entryExit - entryEntry) +
"\nLast Exit: " + (string)(exitExit - exitEntry) +
"\nGoing Insane: " + (string)(entryEntry - insaneStart) +
"\nReturning Sane: " + (string)(defaultStart - entryExit));
++N;
}

touch_start(integer total_number)

}

state insane {
state_entry()

state_exit()

}

The results were:

N is 2
Last Entry: 0.000000
Last Exit: 1.891130
Going Insane: 0.088676
Returning Sane: 2.558797

N is 3
Last Entry: 0.000000
Last Exit: 2.327432
Going Insane: 0.044499
Returning Sane: 2.972273

N is 4
Last Entry: 0.000000
Last Exit: 2.673080
Going Insane: 0.066175
Returning Sane: 3.362556

N is 5
Last Entry: 0.000000
Last Exit: 3.176846
Going Insane: 0.021635
Returning Sane: 3.760851

N is 6
Last Entry: 0.000000
Last Exit: 3.627911
Going Insane: 0.022079
Returning Sane: 4.228088

N is 7
Last Entry: 0.000000
Last Exit: 4.009933
Going Insane: 0.022327
Returning Sane: 4.677006

You're welcome to go and test there. Assuming that 1.24.4 checks out on aditi, it will be fully deployed to the main grid during this week.

1.24.4 'does not' fix this issue as confirmed as writing.

please revisit this issue

Vektor Linden is coordinating the testing of all the 1.24.4 fixes.

If you can create a simple repro and paste it here then Vek can quickly confirm and Babbage can go back to the lab again...

====================================
integer Max = 1000;

check(string where) {
integer ix;
float t = llGetTime();
for (ix = 0; ix < Max; ++ix) {
}
llSay(0, where + ": " + (string)((llGetTime() - t) / (float)ix));
}

default {
state_entry() { check("state_entry"); }
touch_start(integer total_number) { check("touch_start"); state two; }
state_exit() { check("state_exit"); }
}

state two {
touch_end(integer count) { state default; }
}
============================
Mono-compiled:
[18:15] Object: state_entry: 0.000022
[18:15] Object: touch_start: 0.000022
[18:15] Object: state_exit: 0.000022
LSL-compiled:
[18:17] Object: state_entry: 0.000289
[18:17] Object: touch_start: 0.000267
[18:17] Object: state_exit: 0.000289

Without the fix, state_exit handler takes about 100 times longer, for either type of object code.

I rezzed 100 times even and it not happen ! gdi

I swear ! first first time I try it was on the 3rd rez !

would trying to recreate the first time with my alt have a diff?

SL just baffles me at times....I not a coder I just a builder trying to get along.

Renee

EDIT to add: If I were a scripter I wouldn't be building houses, lol.

Kick myself in head - note to self: Self- Kick your head

Renee, don't feel bad... I saw that some sims were still running 1.24.3. I bet you just got a 1.24.3 sim during your first test.

This ugly bug has reared it's head