[v4 3/6] d3dx9: implement fxlc constants (expressions) in effect.

[Paul Gofman]

On 03/19/2016 12:56 AM, Matteo Bruni wrote:
>
>> +    while (param->member_count && param->element_count)
>> +    {
>> +        if (param->element_count > 1)
>> +        {
>> +            FIXME("Unexpected param having both elements and members.\n");
>> +            return D3DERR_INVALIDCALL;
>> +        }
>> +        param = &param->members[0];
>> +    }
> I don't quite understand what you're trying to do here but it doesn't
> look right.
>
That is related to the strange case I met in the real world effect data
when the parameter has structure class, element_count 1, member_count >
1. As I understood from the pre-existing effect parsing code,
element_count 1 together with member_count > 0 (and > 1) is legal. When
element_count is 1, and member_count > 1, members contain an "array" of
1 element, which is actual struct parameter with element_count = 0 and
member_count is the same as in parent weird parameter. It is what
existing d3dx9_parse_effect_typedef understands (element_count is taking
precedence over member_count). I could not reproduce this struct
representation compiling effect from fxc code yet, but seen it in real
world effect. CTAB structure in the shader which correspond to used that
strange struct parameter does not have this weirdness and has element
count 0, so I cannot match CTAB parameter to effect parameter without
going a level deeper in effect parameter.

The existing effect parsing code clearly assumes that if both
element_count > 0 and member_count > 0, element_count is taking
precedence (that is, members array has element_count elements). I am not
sure if it considers illegal element_count > 1 and member_count > 0, but
I do not know currently how to treat this (if it is ever possible) as
did not see such an example, so currently this is a FIXME in the code
(which will hopefully never happen).