Hi! I've been helping with some code to interface to a C library,
and I think I've discovered a bit of a hangup in the 64-bit linux
version of DMD.
If a struct is eight bytes or less (the size of two uints), it
can be passed just fine between D and C code, but if it's any
larger than that, the data gets messed up somewhere along the
way. This doesn't happen when the code is compiled in 32-bit mode
(with -m32)
Here's some example code to trigger the bug:
structyc.c:
-----
#include <stdio.h>
struct passingstruct {
unsigned int a;
unsigned int b;
int c;
};
void printatest (struct passingstruct value)
{
printf ("hello!\n");
printf ("%u %u %d\n", value.a, value.b, value.c);
}
-----
structyd.d:
------
import std.stdio;
struct passingstruct {
uint a;
uint b;
int c;
void printme () {
writefln ("%d %d", a, b);
}
}
int main (char[][] args)
{
auto thingy = passingstruct( 300, 200, 0);
writefln ("%u %u %d", test.a, test.b, test.c);
printatest (test);
return 0;
}
extern (C)
{
void printatest (passingstruct value);
}
-------
When compiled and linked in 64-bit mode, it produces this output:
300 200 0
hello!
394682976 32695 402542592
However, 32-bit mode produces the expected output:
300 200 0
hello!
300 200 0
Additionally, if the third element (.c) is taken out of the
"testman" struct, or if b and c are made ushort or ubyte, the
program works fine in 64-bit mode. GDC(an admittedly out-of-date
version--ubuntu only packages 4.6.3) also successfully passes the
whole struct in 64bit mode.
