Problem with passing ref parameters to properties

[Iain Buclaw]

On 10 May 2012 15:18, SebastianA <sebastian.ahlman at remedygames.com> wrote:
> We are seeing a problem with passing ref parameters to properties. This bug
> only occurs in certain situations. Consider the following code:
>
> ====
> void runTest()
> {
>        Thing t;
>        t.vPosition = (Clock.currStdTime % 2 == 0) ? Vec(2, 2) : Vec(3, 3);
>        Vec v = t.vPosition;
>
>        outputDebug("%d %d\n", v.x, v.y);
> }
>
> struct Vec
> {
>        int x;
>        int y;
> }
>
> struct Thing
> {
>        @property Vec vPosition() { return mPosition; }
>        @property Vec vPosition( const ref Vec value ) { return mPosition =
> value; }
>
> private:
>        Vec mPosition;
> }
> ===
>
> Now, this code should output either "2 2" or "3 3" depending on the time.
> However, on release builds this code will output "0 0" every time it is run.
> This seems to happen regardless of the optimization level (we have tried
> with -O1 and -O3). Some things to note is that the bug will NOT occur if any
> of the following is true:
>
> - We compile the code in debug mode. - We specify a constant known at
> compile time (eg. "true") instead of the non-deterministic time value. - We
> save the vector into a temporary local variable before passing it to the
> property. - We remove the "const ref" from the property setter and pass the
> vector by value.
>
> Debugging the code reveals that the value passed to the getter is indeed
> 0,0, which rules out the getter as the error source.
>
> We are building this on x64 Windows using gdc version 4.6.1 20110627 (gdc hg
> r826:396ce79e6402(default), using dmd ) (tdm64-1).
>
> Any idea what's going on?


Hello Sebastian,


Thanks for your report.  I will check this out and get back to you
this afternoon.

You should be able to produce a debug tree of the code generated in
GCC AST using the command line switch -fdump-tree-original.  This will
output the code sent to the backend in a C-like syntax, and should
help you decode where it may be going wrong.

Have you also checked this test case against the DMD compiler to see
if you get the same or different behaviour that GDC produces?


Regards
-- 
Iain Buclaw

*(p < e ? p++ : p) = (c & 0x0f) + '0';