#2239 incorrect address generated for global with offset

[ClinicallyObeast]

unsigned char data[100];
unsigned char* datapos;

#define ENT_DAT(_p) (*((unsigned int*)((unsigned char*)(_p) + 3)))

void setFoo()
{
    // this line goes wrong. the address generated should be
    // datapos -5 + 3 but you get datapos + 0x00FE (not 0xFFFE)
    ENT_DAT(datapos - 5) = 0xffff;
}

int main()
{
    datapos = data + 5;
    setFoo();
    return 0;
}

code i get:

_setFoo:
;bug1.c:9: ENT_DAT(datapos - 5) = 0xffff;
    ld  a,(#_datapos + 0)
    add a, #0xFE
    ld  d,a
    ld  a,(#_datapos + 1)
    adc a, #0x00                     <<<<< is wrong, should be 0xff ?
    ld  e,a
    ld  l, d
    ld  h, e
    ld  (hl),#0xFF
    inc hl
    ld  (hl),#0xFF
    ret
_setFoo_end::

sdcc -v
SDCC : mcs51/gbz80/z80/z180/r2k/r3ka/ds390/pic16/pic14/TININative/ds400/hc08/s08
3.3.0 #8604 (May 11 2013) (MINGW32)

more info contact,

hughserious@gmail.com

thanks,
-- hugh.

[ClinicallyObeast]

markdown fail.

try this:
#define ENT_DAT(_p) (((unsigned int)((unsigned char*)(_p) + 3)))

[ClinicallyObeast]

markdown fail again. too bad.

[Philipp Klaus Krause]

Using sdcc 3.3.2 #8930 I get:

error 10: 'lvalue' required for 'assignment' operation.

Which looks good to me.

Philipp

[ClinicallyObeast]

it's because the "*" were deleted by markdown, try this:

PS. i have 3.3.0, is there 3.3.2 anywhere to download.

thanks,
-- hugh.

---

unsigned char data[100];
unsigned char* datapos;

#define ENT_DAT(_p) (((unsigned int)((unsigned char*)(_p) + 3)))

void setFoo()
{
// this line goes wrong. the address generated should be
// datapos -5 + 3 but you get datapos + 0x00FE (not 0xFFFE)
ENT_DAT(datapos - 5) = 0xffff;
}

int main()
{
datapos = data + 5;
setFoo();
return 0;
}

On 29/12/2013 21:15, Philipp Klaus Krause wrote:

Using sdcc 3.3.2 #8930 I get:

error 10: 'lvalue' required for 'assignment' operation.

Which looks good to me.

Philipp

---

[Philipp Klaus Krause]

Just attach a file with compileable source. No use trying to get anything meaningful other than natural language across in the bug report itself.

Philipp

[ClinicallyObeast]

thanks. file attached.

On 29/12/2013 22:39, Philipp Klaus Krause wrote:

Just attach a file with compileable source. No use trying to get
anything meaningful other than natural language across in the bug
report itself.

Philipp

---

[Philipp Klaus Krause]

Thanks. I now see the issue in all ports I tried (z80, mcs51, stm8).

Philipp

P.S.: 3.3.2 is not a release. It is just the version reported by the current development verion.

[Maarten Brock]

• Description has changed:

Diff:

--- old
+++ new
@@ -1,7 +1,7 @@
 unsigned char data[100];
 unsigned char* datapos;

-#define ENT_DAT(_p) (*((unsigned int*)((unsigned char*)(_p) + 3)))
+\#define ENT_DAT(_p) (*((unsigned int*)((unsigned char*)(_p) + 3)))

 void setFoo()
 {

[Philipp Klaus Krause]

Increasing priority, since bad code is generated silently.

Philipp

[Ben Shi]

I simplified test case to

char * p;

void f1(void)
{
* (p + 3 - 5) = 0xaa;
}

void f2(void)
{
* ((char * ) (p + 3 - 5)) = 0xaa;
}

f1() was right, and f2() went wrong even in the AST stage.

[Ben Shi]

Fixed in reversion #9160.