chrisbazley wrote:
> A simple test of calling OS_Word from BASIC indicates that the value
> of R1 is preserved by this SWI, and therefore _kernel_osword has
[snip]
> block is set then _kernel_osword will return a negative value despite
> having succeeded!
If _kernel_osword() returns -1 because R1 has been preserved and
passed back it must be because R1 was passed as -1, meaning that
the control block was at &FFFFFFFF, which is not possible, as the
call would have been trying to put a block of data wrapping past
the top of memory.
You will either get a return result of whatever address the control
block was held at - which won't be within a few bytes of the top of
memory - or you will get exactly -1 (&FFFFFFFF) or exactly -2
(&FFFFFFFE).
There are many API calls that used to be described as returning <0
where they actually returned -1, and the documentation has been
updated to specify checking for exactly -1, not for <0. To me, this
is just the same.
(I have a file titled "Changes to APIs due to 32-bit addressing"
which includes the lines: A number of APIs, particulalrly the
Window manager, allow registers or values to be a pointer or <0.
These APIs are being updated so that only 0 or -1 are non-pointer
values.)
> modes. Is it likely that anyone (ab)uses the return value of
> _kernel_osword by casting it back to a pointer? If not then it would
The only place I've used it in in my Z80 emulator in C, which does:
chkerr(_kernel_osword(...));
...
chkerr(retval)
{ if (retval != -2) return(0);
...
> _kernel_osbput functions, which rely upon SWIs XOS_WriteC and XOS_BPut
> returning a positive value in R0 unless they set the overflow flag to
> indicate an error. Isn't this a foolish assumption? The PRM documents
It's a foolish assumtion to test for a negative value with the AI
documentation explicitly tells you to test for -1 or -2.
> All of this makes me think that relying on return values >= 0 to
> indicate success may be a lost cause, and it would be safer to check
> explicitly for -1 or -2. :-(
Exactly, just as the documentation tells you to do.
--
J.G.Harston - j...@arcade.demon.co.uk - mdfs.net/User/JGH
In 1939 $50 of groceries would fill three station wagons. Today I
can lift $50 of groceries with one hand. I must have got stronger.
|
