#2850 [stm8] direct address pointer access fails for += operant

[Michael Mayer]

I want to access the EEPROM area as an array. I defined a address pointer to 0x4000 (the EEPROM memory area). Simple access works fine, but the "+=" operator compiles as "=". Sample code:

unsigned char array[32];
#define adr ((unsigned char*)0x4000)

void main()
{
    array[0] = 1;
    array[0] += 2;  // works as expected

    adr[0] = 1;
    adr[0] += 2;    // compiles as adr[0] = 2;
}

Compiling using the lastest snapshot build #10694:

$ sdcc -v -mstm8 -c test.c
SDCC : mcs51/z80/z180/r2k/r3ka/gbz80/tlcs90/ds390/pic16/pic14/TININative/ds400/hc08/s08/stm8 3.8.3 #10694 (Linux)
published under GNU General Public License (GPL)

The resulting code for array[0]+=2 is as expected, but adr[0]+=2 is compiled as adr[0]=2:

_main:
;   test.c: 7: array[0] = 1;
    ldw x, #_array+0
    ld  a, #0x01
;   test.c: 8: array[0] += 2;   // works as expected
    ld  (x), a
    add a, #0x02
    ld  (x), a
;   test.c: 10: adr[0] = 1;
    mov 0x4000+0, #0x01
;   test.c: 11: adr[0] += 2;    // compiles as adr[0] = 2;
    mov 0x4000+0, #0x02
;   test.c: 12: }
    ret

Is there an issue with my address definition or is this a compiler bug?

[Philipp Klaus Krause]

I can reproduce this bug in current SDCC [r10754]. It seems to happen in the front-end (output from --dump-ast looks already wrong) for all backends (tried z80, stm8).

Philipp

[Philipp Klaus Krause]

Looks like IS_LITERAL is true for the type of adr[0] in decorateType() in SDCCast.c, despite it being an ARRAY_OP, so SDCC tries to optimize away the addition.

Philipp

P.S.: I.e. the type of adr[0] is "unsigned char literal", while the type of array[0] is "unsigned char fixed".

[Philipp Klaus Krause]

Fixed in [r10782].

Philipp

[Maarten Brock]

Isn't this fixed in the wrong place? How did adr[0] become a literal in the first place? Or is it maybe marked isaddr?

And I think this one certainly deserves a regression test case.

[Philipp Klaus Krause]

The the propertyof being a literal is copied from adr to adr[0] (in the old code in COPYTYPE, in the new code in copyLinkChain()). So I changed it to make array indexing behave more like pointer dereference.

I didn't add a regression test case, since the bug is only observable when adr is defined to a fixed memory, making the regression test case highly target-specific.

Philipp