ROOL Forum: Recent posts

Sun, 25 Oct 2026 08:02:07 +0000

as you don’t like AI being used

That’s just the opposite of what I said. My message begins with “I’m not against”.

Oh as an aside as you don’t like AI being used you better stop using all those new pieces of software you have been using lately from Andrew Youll, as they were helped in development with some AI.

As I said, to port a GPL game with AI will raise very few issues. But here, licenses issues are probably not solved, and code attribution can’t be made.

Andrew was very clear in his show talk that the RISC OS 64 developers have been very careful to avoid anything which is related to GPL.

They do, But do the AI has the same ethics?

If it’s StarCoder2, helped with some carefully chosen source, no problem of license (but still some problems of code attribution).
If it’s Claude… hell no. Ask Claude what it thinks of licenses and code attribution issues.

But RISC OS is a broad community with room for all views to co-exist and software options for both.

Once again, you’re right.
Anyway, I never said this project can’t exist.
Just, that I’m not very excited by it :)

My position today is:

• Help to find API, examples = OK
• Ports = OK (code out = 99.999% code in, so license is easy to choose and code attributions are easy to make)
• New tools with clear code attributions = could be OK (it seems to be the case with the GFX accelerator for Pi4)
• New tools without clear identification of the code used = beware

As you said, The GPL license has caught out RISC OS Companies in the past including Castle.

Of course, new rules will emerge, and my position will change.

Sun, 25 Oct 2026 06:04:58 +0000

I think it’s worth pointing out that I did not suggest that the tokeniser was plagiarised.

I apologise if that’s the impression I’ve given, that wasn’t my intention.

But yes where possible and permissible and attributable by license it would be better to reuse code rather than reinvent the wheel.

Sun, 25 Oct 2026 00:14:38 +0000

Sorry Gavin if you were working on porting it too; contributions to this codebase would be very welcome though if there’s things you see you’d like to work on! I suspect there’s more performance to squeeze out of TD & RA especially. I need to find time time to put my ARM610 in the RiscPC and see how they perform.

It was quite a slog play-testing through the many bugs, especially on OpenTS which was only first released in late August, but it seems pretty stable now so hopefully it’s something people can enjoy. It’s been a long time since I was smashing my way through Dune II with all three houses on my A5000 BITD. I don’t mind admitting it was quite emotional the first time netplay worked between the Pi and RiscPC. Well worthwhile trying out just for that, and the Full Motion Video!

I’m glad you liked the Tarmac report, that was a project and a half! I was actually dusting that code off again a few weeks ago for the first time in a quarter of a century and I have a few ideas for things I want to do with it, building RedSquirrel on Windows 11 was the first battle! :)

Sat, 24 Oct 2026 21:50:54 +0000

I undertook this exercise when I was asked by Steve Fryatt about the BASIC tokeniser that Claude had created for the NetStats app, to see if it was plagiarised or not.

I think it's worth pointing out that I did not suggest that the tokeniser was plagiarised.

I simply asked because I was curious where it had come from. Before I wrote Tokenize, I asked around the various RISC OS places to see if what I needed¹ existed, and was advised that it didn't - but that it "shouldn't be too hard to write". It was the first project of mine that went into public source control, because I was asked if it could so that others could use it in their toolchains.

Given that it's now in GitHub and can be pulled in and used as part of a build action, I was just curious to know why the AI had apparently decided that creating a whole new tokeniser within another project was a useful way to spend its tokens, or if there was another tokeniser project out there that I hadn't come across...

1. An ARM BASIC tokeniser that would run on Linux as part of the cross-building tools. ↩

Sat, 24 Oct 2026 21:40:57 +0000

I’ve been told it’s led to Druck’s departure from here but I’ve not seen that post – so I’ll simply say that if people are being made to feel unwelcome and leaving – then that isn’t good for any community. In the end we’ll just be left with the people who enjoy conflict and can put up with the most bullshit.

Interesting. I hadn't spotted the DiscKnight or NetSurf shenanigans...

I don't know whether my comments on here had marked me out such that people were seeking me out to talk to, but it was an "interesting" show today. There was a lot of "Woooo! 64-bit! Stop worrying; just jump on board!" from conversations with non-developers. From developers in similar positions to me, however, things were often a lot more downbeat: I got the impression that there are more than a few folk somewhere between wondering when that project that they've been tinkering with for a while will be overtaken and rendered pointless by someone weilding AI, through to actively looking over their shoulders to see when it will be their turn to have their work forked or cloned.

Given that for most people this is just a fun hobby that may (or may not) generate some pocket money, it didn't feel that we were in a good place with respect to developer retention. I can only hope that I had a biased sample that was too small to be statistically valid...

I must also admit that I'm unsure which side of the bullshit debate I'm on, however, given that I've had posts on this subject described as "nonsense" by Andrew H.

I’ll hold my tongue about what was said about the potential for elder abuse given the demographic of the scene at the moment

Er, what?

Sat, 24 Oct 2026 21:31:51 +0000

I will not be hurrying to trust my data with a tool that seems not to have had very much testing …

Which is one reason that Andrew said in his excellent talk that Risc OS 10 will not be using Filecore as the main disc ‘layer’ (my words not his). Risc OS 10 uses an open source disc format instead, with a suitable license (BSD based??). EXT4 uses the GPL license so cannot be used and NTFS has cost and licensing implications so also cannot be used.

Using tried and tested code should increase the reliability of the filing system(s).

There were a lot of technical details in a short space of time during Andrew’s excellent talk, so apologies if some of my recollections are not correct. I’m sure we’ll be getting lots more technical details soon.

Ralph

Sat, 24 Oct 2026 20:54:54 +0000

Guessing then I’m going to have a look for an old PC with a 3.5 floppy drive to archive all the discs that I have. thought that would be the case trying to connect a DVD drive to the Pi 4 and also running RPCEmu, and sadly the answer is no.

Sat, 24 Oct 2026 20:46:31 +0000

though if charity stands are genuinely being pillaged for “all the good bits to go on ebay” before the show even opens (as has been alleged), it makes the more vile allegations much easier to believe.

I am sorry people where upset with a sale from the charity stand just before doors officially opened. I except your crititisism and will make sure it doesn’t happen again. All items are sold on a first come first served basis.
May I thank everyone who purchased or donated to the stall.

Sat, 24 Oct 2026 20:35:16 +0000

Andrew Youll

Who?

Is it really a clone or a new version of software which does a similar thing but on more Filesystems? When was DiscKnight last time updated/developed? DiscKnight is good at what is does.

When it was updated is hardly relevant, it was a man-made product with a lot of work put into it, and someone having an argument (on this very forum) then throwing out a Clauded version not even a week later is at best grossly disrespectful.

I still use Sergio Monesi’s FSCK. It’s even older, but on old-Filecore disks? Unbeaten. (Discknight doesn’t support the older formats)

I’ve been told it’s led to Druck’s departure from here but I’ve not seen that post – so I’ll simply say that if people are being made to feel unwelcome and leaving – then that isn’t good for any community. In the end we’ll just be left with the people who enjoy conflict and can put up with the most bullshit.

I’ve spent the last 3 months digging deeply into Filecore, finding the inaccuracies in the PRMs and supplementary documents (and where the bear-traps have been left). It’s taken a lot of time and effort to do that – the write-up was AI-assisted, but the reading and effort reviewing the LLM’s work might have been better spent by my writing the whole document myself. It’s left me with a significant amount of respect for Druck’s hard work (as has using Armalyser!)

I will not be hurrying to trust my data with a tool that seems not to have had very much testing, certainly not to the level that ten or more years of people sending corrupted disks to Druck has achieved.

Data recovery is hard, and if you do it wrong you can bugger things up even worse than they started.

AI is not perfect but can be useful if used properly and that’s the important bit ‘used properly’

The problem is, as we’ve established, it’s being used by some as a tool of plagiarism and spite.

Sat, 24 Oct 2026 20:15:33 +0000

What if the person instructing the AI agent has specifically told it to avoid GPL code, for example?

You can advise most LLM models to avoid GPL but they will be honest on that they can’t confirm that no GPL code was utilised in their training material.

Which comes back to the can a programmer whose ever viewed GPL code produce anything that isn’t GPL?

I know this a very basic view, and question as the arguement is LLM has in theory infallible recall, but just try working with one to realise this isn’t the case.

Addendum: The other thing you can do is ask the LLM to do comparisons of code you believe mat have been utilised to build the code it’s produced but this is going to take extra resources.

I undertook this exercise when I was asked by Steve Fryatt about the BASIC tokeniser that Claude had created for the NetStats app, to see if it was plagiarised or not.

Sat, 24 Oct 2026 20:12:31 +0000

Andrew was very clear in his show talk that the RISC OS 64 developers have been very careful to avoid anything which is related to GPL and this partly influenced their choice of which filing systems to support. My understanding is that they also had Aaron’s permission to look at the RISC OS 6 source code which has allowed them to bring in features from that.

The v3d driver that is being developed and being used by RISC OS technologies is based on the v3d Linux driver which is GPL.

Even if its that they’ve just used it as a reference guide to how the GPU works it would be classed as a derivative that’s why my hardware acceleration for the H264 and HEVC for the Pi are GPL as the Linux vchiq and HEVC block src was used as a reference, so it is derived from that work.