I have this amazing ability to totally misinterpret the problem but I'll just throw out something that is probably totally irrelevant  

My programs are:



The program is invoked and returns to the library.

A flag can be set to determine how to shut down.

The reason for this is that I want end-users to be able to write custom code in main but without tampering with certain code in the library.

  matherp said  

G@bor

Please try this version which has the cache functionality disabled and should be pretty much identical to B1

PicoMite.zip

The issue is almost certainly the build running out of C heap but I don't know why.

Hello Peter,

I am back. I tried this version what you have sent, working good again.
Hope you can solve this problem in next betas.

Thanks a lot!

G@bor

ville56
Please try this.

PicoMite.zip

Peter,

B3 works for me without any problems with "option tracecache on 128", performance is about the same as with B2.
Edited 2026-05-01 15:47 by ville56

V6.03.00B3

PicoMiteRP2040V6.03.00B3.zip

PicoMiteRP2350V6.03.00B3.zip
BIG internal change
Fixes bug in SELECT CASE when OPTION TRACECACHE is ON
Various performance tweaks - all versions

Please report anything odd but PLEASE state which version you are running VGA/HDMI etc.
Edited 2026-05-01 21:27 by matherp

On an original Pico 2040.

RTC working so I2C must be ok.

Will fix. LIST PINS has the same bug.

Note to all: Any programs in the library will need re-loading
Edited 2026-05-01 23:31 by matherp

@Peter
Quick initial tests.
A nice speed boost! hut.bas was 8.1 (b1) before; now it’s 6.7 with tracecache and 10.2 without.
However, I noticed that when mods are played back from the FM, they start up with stuttering.
And in B3 there are graphics mode issues. For example, starting demo.bas from FM and aborting it with Ctrl-C leads to this strange behavior when returning to FM. If you then exit FM, the old graphics return. Image 2.
If you restart it, it often recovers.
With other programs as well, after loading and starting them, you sometimes see the screens from the previous programs. If needed, I can record an HDMI video.





Matthias

> option list
PicoMiteHDMI MMBasic USB RP2350A Edition V6.03.00B3
OPTION SERIAL CONSOLE COM2,GP8,GP9
OPTION SYSTEM I2C GP20,GP21
OPTION FLASH SIZE 4194304
OPTION COLOURCODE ON
OPTION KEYBOARD GR
OPTION RESOLUTION 640x480 @ 315000KHz
OPTION SDCARD GP22, GP26, GP27, GP28
OPTION AUDIO GP10,GP11', ON PWM CHANNEL 5
OPTION RTC AUTO ENABLE
OPTION MODBUFF ENABLE  192
OPTION PLATFORM HDMIUSB
>
Edited 2026-05-01 23:34 by homa

I don't see the problem with the graphics progs I'm running. Please post the code for your examples that give the issue

PicoMite MMBasic RP2350A V6.03.00B3
Copyright 2011-2026 Geoff Graham
Copyright 2016-2026 Peter Mather

> option list
PicoMite MMBasic RP2350A V6.03.00B3
OPTION SYSTEM I2C GP20,GP21
OPTION FLASH SIZE 4194304
OPTION COLOURCODE ON
OPTION HEARTBEAT OFF
OPTION PICO OFF
OPTION CPUSPEED (KHz) 200000
OPTION DISPLAY 50, 100
OPTION SDCARD GP22, GP6, GP7, GP4
OPTION RTC AUTO ENABLE
OPTION F1 help
OPTION F5 list commands
OPTION F6 list functions
OPTION F7 list pins
OPTION F8 option list
OPTION F9 fm

B:/backup2350w
14:50 01-05-2026      33332  aadkas3-test2-V6.03.00B2-O-RP2350A-W.bas
14:50 01-05-2026      16297  aadkas3-test3-RP2350A-no-comment.bas

> load "aadkas3-test2-V6.03.00B2-O-RP2350A-W.bas

> files
A:/
  <DIR>  .
  <DIR>  ..
00:00 00-00-1980         58  .fm
15:52 01-05-2026          0  aadkas3-test2-V6.03.00B2-O-RP2350A-W.bas
00:03 01-01-2024      16297  aadkas3-test3-RP2350A-no-comment.bas
00:03 01-01-2024      34844  aadkas3-test3-V6.02.02B5-RP2350A.bas
00:03 01-01-2024      34895  aadkas3-test3a-V6.02.02B5-RP2350A.bas

> save "aadkas3-test2-V6.03.00B2-O-RP2350A-W.bas
Error : Error during device operation

Jan.

The solution was:

> drive "a:/format"
> files
A:/
<DIR> .
<DIR> ..
16:23 01-05-2026 4 bootcount
2 directories, 1 file, 1806336 bytes free

Have a nice day.

Jan.

@Peter: Check out this video.
https://youtu.be/ooXQd0MIFFQ
I restarted the HDMIUSB device, so it was a clean boot. Despite the graphics and HEAP errors, a reset wasn’t necessary. At the point where only the graphics were visible, I could have entered “mode 1” in the console, which is what I did.

The ZIP file contains the files from Drive A shown in the video demonstration.

The "option tracechace on" setting in the last programme (at 1:05) leads to incorrect calculations and display; without this setting, everything works fine again,

Matthias

driveA.zip

  Quote  

The "option tracechace on" setting in the last programme (at 1:05) leads to incorrect calculations and display; without this setting, everything works fine again,

Just to be clear, without caching enabled in that specific program there are no other issues?

Try this

PicoMite.zip

WOW are you quick. The basic raycaster calculation in ccast1.bas now runs without errors when "OPTION tracecache on" is enabled. FPS has gone from 15 to 25!! That’s a massive improvement.
The other issues – mod stuttering, the graphics mode with FM and "Error : Not enough System Heap memory" – are still there.
Edited 2026-05-02 02:20 by homa

  Quote  

Error : Not enough System Heap memory

What is the sequence to make this happen?

On the graphics mode do you just mean the flash of garbage before FM redraws?

The mod stuttering is just as playback starts or does it continue? If just the start it may be just one of those things until the Picos cache catches up.

Have you watched the video?

FM does not return correctly; a graphics error occurs. By the way: is there an option to load/run a programme in CMM2 mode from within FM?

The mod stutters only at the start and when you continue using FM, such as when changing directories, etc.
But if you start it from the CLI, it doesn’t stutter!


After editing ccast.bas in FM (F2), then pressing RUN (ENTER) and exit, and then EDIT again, a HEAP error will occur at the latest.

I think I've fixed everything except perhaps the graphics error. Please try this version and if the graphics problem is still there report exactly the sequence to create it

UPDATED 17:19 UTC

PicoMite.zip
Edited 2026-05-02 03:21 by matherp

V6.03.00B4
PicoMiteRP2040V6.03.00B4.zip
PicoMiteRP2350V6.03.00B4.zip

Fixes bugs introduced in LIST PINS and LIST SYSTEM I2C
Fixes bug in cache optimisation of NOT command
Fixes stall of audio on startup and directory change in FM
Fixes memory leak when exiting a program with TRACECACHE enabled back to FM
Fixes memory leak and display corruption when exiting a program writing to a framebuffer back to FM
Fixes temporary screen flash when exiting a program running a different mode back to FM

Here is a document describing the various optimisations in B3 onwards. Note, the change to the Tokenised Line Format section. It is this change that means all library functions must be re-loaded from source.


option-profiling-cache.pdf
Edited 2026-05-02 04:10 by matherp

From the Option Profiling PDF: "any event that restructures program memory or the variable table (NEW, DIM, ERASE, OPTION BASE, OPTION EXPLICIT) invalidates all entries".

Does this include REDIM? If so, could Struct Sort array().membername [, flags] have an additional optional parameter to indicate the number of records to be sorted?

I "REDIM PRESERVE" arrays of structures which may have a varying number of elements to the exact size, sort and process, and then REDIM back to MAXSIZE to accept the set of structure elements.

(Or perhaps there is a better way to do this.)

Peter,
Sorry for the delay, I was still out and about.
I’ve now gone straight for B4!
Your work pace is impressive :-) and the new caching skills too!
All the bugs found so far have been fixed; even when playing MP3s, I can now switch between directories in FM without it stuttering!  

Matthias

P.S. Did you see my question about the CMM2 LOAD/RUN command from within the FM? I now find the speed in the editor to be very good; I get the impression it has improved significantly. The same goes for the “long” line feature.
The only thing is that I still find it a bit slow to scroll through long directories in the FM (via TerraTerm)—maybe it’s faster natively. Can’t TerraTerm or VT100 also move around sections?