Hollywood app crashes

[amyren]

I have developed this painting program, PolarPaint with Hollywood.
It is the Undo feature that causes the program to crashm with a "Table Field 2 not initialized" error
It is easy to replicate, just draw a few dots on and then use the undo button a couple of times and the program will crash.

I tested the same executable on WinUAE, and my A1200 (pistorm) and even under Amiberry 7 on linux mint, and it runs without issues.
The executables for windows and MorphOS does work fine as well.
I suspect that the issue must be a bug in the amiberry for arm version used in the A600GS. On the A600GS the error will happen on OS3.9 and 3.2 as well as on Amibench.

Here is a small minumum code to replicate the issue from Hollywood:

UndoStack1 = CreateList()
SetFontColor(#WHITE)
SetFillStyle(#FILLCOLOR)
counter = 100
For i = 1 To 10
   InsertItem(UndoStack1 , counter)
   Local undoBrushID1 = UndoStack1[ListItems(UndoStack1)-1] ; Get the last undo state
   Wait(40)
   Box(300, 100, 50, 20, #BLACK)
   TextOut(300, 100, undoBrushID1)
   counter = counter+1
Next
Wait(50)
For i = 1 To 10
   Box(300, 100, 50, 20, #BLACK)
   Local undoBrushID1 = UndoStack1[ListItems(UndoStack1)-1]
   TextOut(300, 100, undoBrushID1)
   Wait(40)
   RemoveItem(UndoStack1, ListItems(UndoStack1) - 1)
Next
WaitLeftMouse

[F0LLETT]

Is jit enabled?

[broadblues]

I have developed this painting program, PolarPaint with Hollywood.
It is the Undo feature that causes the program to crashm with a "Table Field 2 not initialized" error
It is easy to replicate, just draw a few dots on and then use the undo button a couple of times and the program will crash.

I tested the same executable on WinUAE, and my A1200 (pistorm) and even under Amiberry 7 on linux mint, and it runs without issues.
The executables for windows and MorphOS does work fine as well.
I suspect that the issue must be a bug in the amiberry for arm version used in the A600GS. On the A600GS the error will happen on OS3.9 and 3.2 as well as on Amibench.

Here is a small minumum code to replicate the issue from Hollywood:

UndoStack1 = CreateList()
SetFontColor(#WHITE)
SetFillStyle(#FILLCOLOR)
counter = 100
For i = 1 To 10
   InsertItem(UndoStack1 , counter)
   Local undoBrushID1 = UndoStack1[ListItems(UndoStack1)-1] ; Get the last undo state
   Wait(40)
   Box(300, 100, 50, 20, #BLACK)
   TextOut(300, 100, undoBrushID1)
   counter = counter+1
Next
Wait(50)
For i = 1 To 10
   Box(300, 100, 50, 20, #BLACK)
   Local undoBrushID1 = UndoStack1[ListItems(UndoStack1)-1]
   TextOut(300, 100, undoBrushID1)
   Wait(40)
   RemoveItem(UndoStack1, ListItems(UndoStack1) - 1)
Next
WaitLeftMouse

I don't have hollywood to compile with, could you compile that test cde and send me it at andy@broad.ology.org.uk then I can examine the logs when it crashes and see if anything is obvious.

[amyren]

Is jit enabled?

I tried both, same result.

Off topic, but since you mentioned JIT. Does the setting work at all for you? I have this OS3.9 setup and I cant spot any difference in usage if I enable JIT or not. Also SysInfo speed tests are the same, about 12 times faster compared to A4000 040/25

[F0LLETT]

I tried both, same result.

Off topic, but since you mentioned JIT. Does the setting work at all for you? I have this OS3.9 setup and I cant spot any difference in usage if I enable JIT or not. Also SysInfo speed tests are the same, about 12 times faster compared to A4000 040/25

It does, also depends of CPU speed setting.

I'd listen to what Broadblues said and get back to him here.

[amyren]

It does, also depends of CPU speed setting.

I'd listen to what Broadblues said and get back to him here.

Unless caught in spam filter, Broadblues got the attachment yesterday

About the JIT I guess I could create a separate thread for it to get in depth of it

[broadblues]

It was in my spam draw in YAM, will test ....

[broadblues]

After opening a black window it loses again and displays a errr requester saying:

Table field 1 was not initialised!

File: Unnamed1 (current line 7)

No crashes occurred.

[broadblues]

WRT to JIT

If CPU Speed is set to maximum then JIT will mean faster 68k CPU speed.

If CPU Speed is set to specific speed, then JIT will mean less Arm CPU used to obtain that speed

That's how I understand it anyway.

[amyren]

It does not crash the system, it is the app itself that crashes (exits on error)

The black window should show a counter that display the numbers from 100 to 109, then count down to 100 again. After this the window should stay open until you press left mouse button.
If it doesnt, it means the app crashed.
Try the executable on winuae or a real amiga to test.

[AmigaNG]

I think it’s a know slight issue on Arm JIT

https://github.com/BlitterStudio/amiberry-lite/issues/4

https://github.com/BlitterStudio/amiberry/issues/1125

I found JIT to be a little less stable than it is with winuae/ x86, Turning it off or tweaking setting in emulation config can help, also amiberry updates fixed a few issue I was running into in the past, so it good it getting more stable.

[amyren]

I think it’s a know slight issue on Arm JIT

https://github.com/BlitterStudio/amiberry-lite/issues/4

https://github.com/BlitterStudio/amiberry/issues/1125

I found JIT to be a little less stable than it is with winuae/ x86, Turning it off or tweaking setting in emulation config can help, also amiberry updates fixed a few issue I was running into in the past, so it good it getting more stable.

I think you are right.
There must be an issue with my A600GS setup, since it apears not to change anything if I turn JIT off or On. And the sysinfo readings I get suggests that JIT is enabled always.

To test this theory I did replace the SD card with one with a regular OrangePI linux install on. There I have Amiberry-Lite v5.7.5 installed. Turning off JIT and it runs without the error, with JIT on it acts just as on the A600GS setup.

Also I edited the script a bit to avoid the "crash with error" situation. This script will cause the program to run without premature exit, but will reveal that JIT messes up the sequence of reading tables in Hollywood.
With JIT it will count up from 100, but skip the number 102. Then when it counts down it will skip 102 again, and jump to 101, 100, and then 102.

Hollywood script:
UndoStack1 = CreateList()
SetFillStyle(#FILLCOLOR)
counter = 100
For i = 1 To 10
   InsertItem(UndoStack1, counter)
   Local undoBrushID1 = GetItem(UndoStack1, ListItems(UndoStack1)-1)
   Wait(40)
   Box(300, 100, 50, 20, #BLACK)
   TextOut(300, 100, undoBrushID1)
   counter = counter+1
Next
Wait(50)
For i = 1 To 10
   Box(300, 100, 50, 20, #BLACK)
   Local undoBrushID1 = GetItem(UndoStack1, ListItems(UndoStack1)-1)   
   TextOut(300, 100, undoBrushID1)
   Wait(40)
   RemoveItem(UndoStack1, ListItems(UndoStack1) - 1)
Next
WaitLeftMouse

[amyren]

After opening a black window it loses again and displays a errr requester saying:

No crashes occurred.

Just emailed you a updated executable to test with and without JIT

[broadblues]

Thanks I can confirm the issue, I notice the first github issue shared above is one of yours?

Hollywood is a bit of a monolithic program to track something like this own with, and I'm no expert on the JIT in the first place, ideally a small fragment off C or assembly code that can reproduce the bug is needed.

[amyren]

Yea, that first github issue was me:)
More than 5 years ago, so I guess the bug may remain a while