Fw: [sdcc-devel] pointers

[Johan K. <joh...@id...>]

> > Now you really can do something like:
> >
> > char code * data * * xdata cpdpgp_x;
> >
> > data char i;
> >
> > void test () {
> >     i = ***cpdpgp_x;
> > }
>
> I believe the generated code is not correct.  As I understand,
> cpdpgp_x is 3 bytes in xdata, pointing to a 1 byte pointer in
> data (compiled w/ --model-small), pointing to a 2 byte pointer
> in data, which ultimately points to 1 byte in code.

No. You should read it from left to right. So, a code pointer (two bytes) to
a data
pointer (one byte) to a generic pointer (three bytes) in xseg.

> Adding _generic to make things clearer, this is what we
> should have: char code * data * _generic * xdata cpdpgp_x;

That's exactly what it is, but you can't do that anymore, _generic will now
be seen as an identifier. Maybe we should warn about this. In fact that was
the whole point. E.g. xdata valids from left to right, whereas _xdata used
to read from right to left (but was never implemented correctly, that's why
I wanted to remove it).

> Also, I tried compiling in large model, and I get the exact
> same assembly output.  As I understand it, the middle pointer
> should be allocated in xdata with large model, right?

No, only the last type specifier (in fact it's a storage specifier) between
the last '*' and the declarator determines the storage. That one is set
(xdata), so it won't make any difference in large or small model.

Johan