forward references...again

[kede]

Hi all,

I know this has been flagged as a bug before, but it doesn't seem to be getting fixed.
So I'm gonna see if i can fix it myself.

In its simplest form the following three files won't compile:

----
[ vec.d ]
module vec;
import mat;

struct Vec { 
  float x, y, z; 
  Mat mat() { Mat m; return m; }
}

[ mat.d ]
module mat;
import vec;

struct Mat { Vec cols[3]; }

[ main.d ]
import vec, matrix;

int main() { Vec v; Mat m; return 0; }
---- 

dmd -c matrix.d; will compile
dmd -c vec.d;  fails

However, its the parsing of matrix.d that fails when u try to compile vec.d.

Basically, on the first semantic pass of vec.d the module mat will get parsed causing
a semantic pass on struct Mat, which flags errors because it cannot determine the size of the struct.  Then it returns back down to the vec module and successfully determines the size of the Vec struct as 12, and returns back to the main driver
[d-lang.c:d_parse_file()] which exits because of the flagged errors.


Anyways, I'm looking for some feedback concerning how to fix this in an acceptable fashion.  From what I can see, this problem seems to have already made the code in many modules somewhat less than readable, to me at least.  So rather than placing another hack into the code my initial idea is to add an addition semantic pass prior to the first, to scan through each module, tail recursing through the imports only if they are needed to determine the size of aggregate types... or maybe this will just end up being a hack too and the first semantic pass should really be able to handle this... :D

Thoughts or feedback before I start on this, or reasons why I shouldnt waste my time ;)

kede