Simplify translation of `call 0`/`jmp 0`/`call 5`/`jmp 5`

[tkchia]

Proposed changes:

• I removed the push ax and pop ax surrounding each system call translated from call 5/jmp 5. I believe this is more correct: some CP/M (and MS-DOS 1.x) syscalls return status codes in the a (or al) register, and push ax ... pop ax will destroy these.
• There should be no need to set cl = dl = 0 when exiting the program via int 0x21, ah = 0x4c. The only parameter the syscall accepts is an exit status, which will come from al.
  • (An alternative might be to use int 0x20 here, which will work on MS-DOS < 2. Yet another alternative might be to simply leave call 0 and jmp 0 as-is.)

Thank you!

[ibara]

Hello --

I think the push ax/pop ax came from DRI's tool and probably doesn't make much sense for MS-DOS.

What do you think about leaving jmp and call to 0000h as-is? That should work for all MS-DOS and allow us to de-special case 0000h entirely, which would be nice.

[tkchia]

Hello @ibara,

I think the push ax/pop ax came from DRI's tool and probably doesn't make much sense for MS-DOS.

Thanks for the information.

What do you think about leaving jmp and call to 0000h as-is? That should work for all MS-DOS and allow us to de-special case 0000h entirely, which would be nice.

Leaving jmp 0 and call 0 as-is should Just Work on all versions of MS-DOS (including 1.x) — as long as the cs register points to the Program Segment Prefix.

PSP:0 will always point to an int 0x20 instruction. int 0x20 wil terminate the program as long as cs = PSP.

For .com files, cs = PSP at startup always. For .exe files though — needed for small memory model support — this will need to be specially ensured.

Thank you!

[ibara]

OK. Let's stick with what you have for now. We can revisit the MS-DOS < 2 question later.

[tkchia]

@ibara : OK, thank you!