OS_ReadLine bug in RO3.x / 4.x

OS_ReadLine seems to corrupt R4 on RO3.x and RO4.×.

I’m noting it here should someone Google it – unlikely, but you never know … I did! The PRM doesn’t actually say what happens to R4 on exit, but it was preserved on RO2 so the logical assumption would be that it’s also preserved on RO3+.

EDIT: This turned out to an issue with the LineEditor module, which corrupts R4 on 26bit non-SA OS’s.

I believe the PRMs state that any registers not listed in the “On exit” section will be preserved, unless explicitly stated otherwise (PRM1-29). The documentation for OS_ReadLine (PRM1-910) looks like R4 should be preserved on exit, so I think you’re correct that this was a bug.

They do indeed state exactly that.

Do you have any sample code which shows the bug? I find it somewhat surprising that a bug like that could go unnoticed for so long. I can’t see anything obviously wrong in the old kernel sources (R4 gets stacked on entry/exit), and some quick tests in BASIC in RISC OS 3.1 and 3.7 don’t show anything bad happening.

OS_ReadLine is a vectored SWI (ReadLineV), so it’s possible it’s something else sitting on the vector which is corrupting R4.

I tested it from BASIC using:

DIM Z% 256
SYS "OS_ReadLine", Z%, 128, 32, 128, &1234 TO ,,,,E%

On RO3.1 thru RO4.03 E% wasn’t &1234 on exit, off the top of my head it was &40 or something similar. I only picked up on it when I noticed a RO2 game doing odd things after text input.

It’s you.
Check what you have loaded that could be messing with ReadLine. Perhaps something like the LineEditor module?

Here is a vanilla install of RISC OS 3.10:

Here is a basic install of RISC OS 3.50 that loads PreDesk but drops out before entering the desktop (I was testing Harinezumi so I set the obey file to stop at that point):

And here is a reasonably customised install of RISC OS 3.70 (and, yes, I prefer coding BASIC in Edit with a yellow-on-blue colour scheme, I find it more readable than the black/white default):

In all cases, R4 is returned as &1234 as expected.

These are all under emulation, by the way.

That code works fine here, under emulated 3.1 & 3.7 and real 3.7. What kind of stuff do you have in your boot sequence? Aren’t there a couple of line editor utilities for RISC OS which add tab completion, history, etc. to the CLI? It wouldn’t surprise me if they worked by hooking into ReadLineV.

Just tested with Zap’s LineEditor v2.73 (20 Oct 1998) under RISC OS 3.70 – works fine.

I should add – since libraries and assembler programs will be written on the assumption that R4 is not trashed, and since R4 is not one of the APCS trashable registers, I would have thought something this big might have been seen before.

Do you have a utility like *Vectors on one of your machines that does not work as expected? What is sitting on ReadLineV?

Can you boot to a base shell? On older machines – start up with Shift and numpad-* held down – should drop you directly into the command line. Try your little program then. What’s the result?

Poked around StrongEd on the Pi image. Hmmm, it doesn’t appear to have a line editing facility like Zap… Have I missed something?

What in particular are you looking for?

You can certainly use the LineEditor module with StrongED, been doing that for years. You can use it with Edit too, it’s not specific to Zap – people seem to think it is because it’s distributed with Zap.

EDIT:

One thing just occured to me. Do you have the TaskWindow mode installed?

It has to be present (and used) or keystrokes won’t be passed on to the underlying task.

What in particular are you looking for?

A line editor that is a part of StrongEd.

I know that Zap’s LineEditor module is okay (already tested it) and that it works with other things (I use it a lot in ShellCLI). I was checking to see if you provided something similar.
Essentially, I’m looking around for other stuff that might mess with ReadLineV to see if I can figure out what is corrupting R4 on the OP’s setup. But it’s pretty much a shot in the dark as I don’t know what the OP is actually running.

One thing just occured to me. Do you have the TaskWindow mode installed?

No idea. I guess so as I started StrongEd, pressed ^F12 and a taskwindow (with buttons at the top and options/status at the bottom) appeared which worked and did stuff (including running the sample program above).

A line editor that is a part of StrongEd.

There isn’t, the LineEditor module works fine with StrongED if you want one. There’s certainly nothing in StrongED sitting on ReadLineV.

No idea.

You can tell by looking at the Infobar at the bottom. The name of the mode used will be displayed just right of centre. From your description I’d say TaskWindow mode is installed.

I could swear I SHIFT-booted at least the A440/1, in hindsight, RO3.11 has a nasty habit of retaining modules if you don’t do it from a power on.

Anyhow…I’ve updated the OP. The culprit was the LineEditor module (2.76d is the version I tested) and only affects 26bit non-SA OS’s.

I’ve left the patch in ADFFS anyhow, as it’s likely to be a commonly used module.

I’ve had a quick read of the source code, but I am still wondering how a program actually uses the LineEditor module?
My port of sash (a linux shell) does not inherit the modules facilities, though I have just noticed that in singletasking F12 that while it does not inherit the LineEditor, it does inherit the standard editing (as in moving the cursor with the arrow keys and using the copy key).
Or maybe it is just LineEditor can’t be passed on from one task to another.
The LineEditor functions do take over the singletasking F12 CLI editing but they don’t get passed onto another shell.