508MB RAM disc on 512MB machine

Reading R-Comp’s announcement of an OS update for the ARMX6 and increased RAM disc size, I thought to amuse myself by trying for the maximum. Absent mindedly forgot about the BB -xM only having 512MB.

Pulled the RAM disc column in the task manager until “System variable ‘Resource$Path’ not found” error box. MemNow shows 20 bytes of memory left.

Apart from my senility (or because of), should I be allowed to do that? Not sure if there’s a way to recover except by reset.

20-10-15 ROM

Apart from my senility (or because of), should I be allowed to do that? Not sure if there’s a way to recover except by reset.

Technically you’ve always been able to break the OS by starving it of free memory, it’s just a bit easier to do it now that the RAM disc size limit has been increased, and the OS is more likely to fall over than (say) RISC OS 3.1 due to many systems being reliant on being able to make temporary allocations from the RMA.

Part of me wants to leave this capability there, because I know there are bugs in how the OS handles out-of-memory situations – so as RISC OS 5.24 draws closer I’m hoping to spend some time fixing some of the more nasty ones. I’m OK with stuff stopping working if you starve the system of free memory, but I’d want to make sure that things will start working again once you shrink the RAM disc back down to a more sensible size.

But on the other hand, part of me realises that it’s a bit silly that the OS makes it so easy for the user to break the system. So maybe we should do something like have the task manager place a limit on how small it allows the user to shrink the free pool? E.g. it shouldn’t allow you to take away the last 256K of free memory, or, some percentage of total RAM size for machines with under a certain amount of RAM. I’d say that a 32MB machine is at the low end of the spectrum, so we’d set the clamp at min (256K, total_ram*256K/32M).

Windows just reserve a part of memory for its own use. So only applications can break, never the OS.

I can’t speak for the reliability of Windows 10; but I hope you realise that it has taken thirty miserable years to get to where somebody can actually say “only applications can break, never the OS” with a straight face when referring to Windows.

Windows 1 (1985) was icky and a bit crap. I have actually used this. It was less like Windows as we know it and more like a fancy version of DOSSHELL.
Windows 2 (1987-1989) was too much like Windows 1.
Windows 3 (1990) was the birth of what people knew worldwide. That it could actually manage to run more than one task was a huge benefit; but the main thing was the GDI system so stuff could be printed to the printer as it appeared on the screen without loads of complicated incantations for each printer, a generic driver offered the same resource for every application. Anybody who has ever had to configure WordPerfect 5.1 for an unusual printer will know what a refreshing change this was.
Problem was, GDI didn’t offer many handles, things liked to claim handles, and the machine falling over in a heap was quite common. You remember “General Protection Error”? That was the invalid memory access that Windows noticed; but there were plenty that it didn’t see – like anything that wanted to poke hardware directly. Or a crashing app spewing stuff into the first dozen kilobytes…
Windows for Workgroups 3.11 (1993) was slightly less buggy and even had software to share files just like I’d been doing for years with Econet. And thanks to the horrible interrupt controller in the 286 era, it was only marginally faster…
Windows 95 (1995) was a big change from the fugly 16 bit era, but it was about as stable as a ballerina on a rolling pin, leading to more than one BSOD while Big Bill was demonstrating it. For my part, I actually had the damn thing BSOD on me if I manually ejected the CD (didn’t need to be in use, just needed to be ejected). Oh, and the system was able to trash itself internally so much that the registry was hated and the standard recovery advice was “reformat and reinstall”. There were a bunch of OSR released, but a piece of poo is still a piece of poo no matter what colour you paint it.
Windows 98 (1998) was better but apparently had some strange incompatibilities with then-new machines. I don’t know, I used Windows 98 SE (1999) and found it to be actually quite reliable. Reliable considering that none of the Win32S versions did much in the way of memory access checking. I wrote my little driver to bit-bang IIC on the parallel port by simply poking data to the standard location of the parallel hardware. Did Windows complain? Heh, did it even notice?
Some people hated Windows ME (2000), some said it was okay. I never used it.
That was the last of the Win32S builds.
Windows XP (2001) was not just a revamping of the system, it was a shift to the somewhat more solid NT kernel that does a lot more validation on what an app can and cannot do. I am running XP on my netbook and it is reliable – I can count BSODs in units of years. A machine at work bombs out frequently. Why? The same basic weakness as RISC OS – what you let random third party rubbish (say hello to DLL hell) run with kernel level privilege, if the code is buggy it can bring the whole machine down. Additionally, XP is extremely vulnerable to viruses and the like due to the decision of making the user “administrator” by default. This means stuff can be installed that Windows doesn’t even attempt to protect against. Not that the protection is worth much. Do you know how I replaced Notepad with Metapad? I wrote a batch file to copy the executable into the Windows protected files folder and then delete the normal executable. Windows noticed this and obligingly replaced Notepad (fifteen seconds later!) with…you guessed it, Metapad. Nice job, Microsoft.
XP was 2001. We have had Vista (2006), 7 (2009), 8 (2012), 8.1 (2013) and now 10 (2015) since then. I’m hoping 10 does actually implement a decent separation of kernel and user so that your statement is correct – but unless it segregates all drivers from kernel space, it will still have weaknesses that could cause an application to crash the system. I may be wrong here, but I thought the only system to push pretty much everything non-kernel out of kernel space was Minix?
Actually, I am aware of what XP does if memory runs out as my netbook has an SSD so there is no swap file. I have 2GiB fitted and that’s it. What happens is that Windows will pop up an ! warning that virtual memory has run out (generic message, my virtual memory is zero). If I am lucky, a random application or two (hello Firefox) will die. If I am not so lucky, Explorer will die and I don’t mean the file browser, I mean the task bar and backdrop and Win-R and all that stuff, basically leaving a few tasks running and no way to do much of anything. Shutting down is fun – I close all applications, wait until disc activity stops, then long-press on the power button. A controlled shutdown is better, but with Explorer dead there is nothing to respond to the request…
I suspect running out of memory is why my old Android phone would reboot itself. At least iOS just kills errant tasks before they risk messing up the entire device. At least, so far as I have observed.

tl;dr: Windows might be reasonable today, just remember it was a laughing stock for a long time. Anything can suffer if memory completely runs out.

Jeffrey – I like the idea of Switcher refusing to release the last ~256K so there is some buffer space for various internal memory claims.

I have just done a test on my Pi. Older ROM, pre PMP.
Screen memory = 832K
SysHeap = 2816K
Module area = 161280K
Font cache = 32768K
System sprites = 16384K
RAM disc = 208K
Free = 0K

The RAM disc was never actually created, the machine entirely stiffed (not even the mouse pointer or keyboard LEDs work).

Okay, it was a highly contrived example, but none the less, an unpleasant thing to run into by accident…

So maybe we should do something like have the task manager place a limit on how small it allows the user to shrink the free pool?

That would be good. It’s not a problem that one can steal a stupid amount of memory but rather that it can’t be given back. If the RAM disc doesn’t die…