#2641 Incompatible pointer types, even though one is void *

[Philipp Klaus Krause]

In SDCC #9994, I see the following error messages, when initalizing a void * from a different pointer type (there awas an error before, but in #9994 it was made more verbose to print the types involved).

static  struct { unsigned char STK[128]; } s;

void *v = &s + 1;

For the stm8 and z80 backends I see:

test.c:3: error 129: pointer types incompatible 
from type 'struct __00000000 near* fixed'
  to type 'void generic* fixed'
test.c:3: error 129: pointer types incompatible 
from type 'struct __00000000 near* fixed'
  to type 'void generic* const fixed'

Which looks a bit weird, since there is no near in the stm8 and stm8 backends. All pointers should be generic.

For the hc08 backend I see:

test.c:3: error 129: pointer types incompatible 
from type 'struct __00000000 xdata* xdata'
  to type 'void generic* xdata'
test.c:3: error 129: pointer types incompatible 
from type 'struct __00000000 xdata* xdata'
  to type 'void generic* const xdata'

And what is that 00000000?

Philipp

[Philipp Klaus Krause]

Even weirder, even the following gives silimar errors:

static  struct { unsigned char STK[128]; } s;

void *v = (void *)(s.STK + 1);

PHilipp

[Rajmund Szymański]

In this case:

struct _s { unsigned char STK[128]; };

static struct _s s;

void *v = (struct _s *)&s + 1;

there are no errors.

[Philipp Klaus Krause]

A closer look at the ISO C11 standard, section 6.6 indicates that SDCC does not have to accept '+' in address pointer constants. In that case the bug is not SDCC rejecting the code, but rather SDCC giving a wrong error message.

Philipp

[Philipp Klaus Krause]

In revision [r10036], I changed the default error message in initPointer() to complain about the expression not being constant instead of incompatible types.

Philipp

[alvin]

Does that mean your example is still an error?

static  struct { unsigned char STK[128]; } s;
void *v = (void *)(s.STK + 1);

I think section 6.6 explicitly mentions this case:

7 More latitude is permitted for constant expressions in initializers. Such a constant
expression shall be, or evaluate to, one of the following:
...
— an address constant for a complete object type plus or minus an integer constant
expression.

[Philipp Klaus Krause]

I was only looking at verse 9 of 6.6, but yes, verse 7 allows more. I'm not sure about the cast though.

But it seems SDCC should at least accept

static  struct { unsigned char STK[128]; } s;
void *v = (s.STK + 1);

So this bug is open again.

Philipp

[Rajmund Szymański]

Interesting example:

struct _s { char t[10]; };
struct _s s1;
void *v1 = &s1 + 1;                // ERROR
void *v2 = (struct _s *)&s1 + 1;   // OK

void main()
{
    struct _s s2;
    void *v3 = &s2 + 1;              // OK
    void *v4 = (struct _s *)&s2 + 1; // OK
    void *v5 = &s1 + 1;              // OK!
    void *v6 = (struct _s *)&s1 + 1; // OK! 
}

[Philipp Klaus Krause]

Not much of a surprise. The file-scope objects need constant expressions as initializers, the function-scope auto ones do not.
Even when this bug is fixed, still a lot more will be allowed for the function-scope auto ones, than for the file-scope ones.