Re: [sdcc-devel] pic16 bug

[Borut R. <bor...@si...>]

Raphael,

is this something that should be fixed also for the release?

Borut


Raphael Neider wrote:
> Hi David,
>
> Sorry for the rather late reply...
>
>> i think i found a bug in the sdcc pic16 port:
>
> Yep, you have.
>
>> struct foo *g=&f;
>>  
>> $ sdcc -c -mpic16 -p18f452 bug.c
>> Internal error: validateLink failed in SPEC_SCLS(cexpr->etype) @ 
>> SDCCast.c:1513: expected SPECIFIER, got null-link
>>  
>> I would like to fix this bug, does anyone know what might be the 
>> cause of this bug?
>
> The initialized struct foo* g is causing trouble. The workaround is to 
> have
>    g = &f;
> somewhere early in your program, e.g. in your main().
>
> To avoid the assertion, you may apply the attached patch using
>   patch -p0 < init-ptr-struct.patch
> in sdcc/src/pic16. It actually fixes this bug and makes it into a 
> missing feature ;-). The patch may also guide you as to where to look 
> for implementing this feature. Patches would be highly appreciated, 
> though if I recall correctly, this bug is related to spurious symbol 
> allocations in the backend, which leaves out requried type information 
> and thus causes this (and other kinds of) trouble. So, this is not 
> likely to require a simple, local fix somewhere in the backend but 
> rather a more or less global approach...
>
> I will rework the initialization code over the next months (?) and 
> also try to get rid of these spurious symbols, referencing the `real' 
> ones instead. So do not waste too much time on this one.
>
> Regards,
> Raphael