Re: "or" shortcut does not work - but why?

[Andreas Rozek <Rozek@...>]

Good evening!

>> I'll try to look at the TEST instruction, as suggested by Nick.
>
>A particularity of this instruction is that it assumes that "boolean
>expressions" in C (tests, && and ||) always result in 1 for true 
(instead
>of any non-zero value). Of course ANSI C ensures that, but maybe
>the compiler is doing something wrong there?

I guess, the relevant code (in lvm.c) looks like this:

      case OP_TEST: {
printf("TEST: l_isfalse = %d, GETARG_C = %d\n",l_isfalse(ra),GETARG_C
(i));
        if (l_isfalse(ra) != GETARG_C(i))
          dojump(L, pc, GETARG_sBx(*pc));
        pc++;
        continue;
      }

Well, I added a simple "printf" for testing purposes, of course
The output of that printf statement is as follows:

  TEST: l_isfalse = 1, GETARG_C = 1

Comparing this with the opcodes produced by lua:

0041  02000000           [01] loadbool   0   0   0    ; false
0045  07000000           [02] setglobal  0   0        ; Result
0049  05000000           [03] getglobal  0   0        ; Result
004D  1A400000           [04] test       0       1    ; to [6] if false
0051  16000080           [05] jmp        1            ; to [7]
0055  01400000           [06] loadk      0   1        ; "anything else"
0059  07000000           [07] setglobal  0   0        ; Result
...

I'm a bit unsure how to interpret these values:

 - l_isfalse = 1  - looks ok, doesn't it?
 - GETARG_C = 1   - is that correct? (shouldn't it be 0?)

lopcodes.h defines GETARG_C as follows:

#define GETARG_C(i) (cast(int, ((i)>>POS_C) & MASK1(SIZE_C,0)))

Perhaps I have to look deeper in that direction!?

Kind regards,

Andreas Rozek         Phone:  ++49 (7031) 436 5784
Brunnenstraße 30/2    Mobile: ++49 (151) 172 879 18
D-71032 Böblingen     EMail:  Andreas.Rozek@gmx.de
Germany               URL:    http://www.Andreas-Rozek.de