import std.stdio;
struct Bar {
uint num;
Bar opAssign(uint otherNum) {
num = otherNum;
return this;
}
}
void main() {
Bar bar = 1;
writeln(bar.num); // Prints 0.
}
Comment #1 by clugdbug — 2009-08-04T02:22:05Z
I think this is invalid.
Bar bar=1; is a construction, not an assignment.
Comment #2 by clugdbug — 2009-08-07T01:16:33Z
Actually there is a bug:
Bar bar = 1;
should not be accepted.
Here's a really frightening case:
struct A { char [] a; }
struct B { long x; }
void main() {
B s;
A q = s;
}
--------------
The cause is in VarDeclaration::semantic in declaration.c, line 1083 (DMD 2.031), in the section dealing with initialisation of a struct. The code is:
if (!ei->exp->implicitConvTo(type))
{ Type *ti = ei->exp->type->toBasetype();
// Don't cast away invariant or mutability in initializer
if (!(ti->ty == Tstruct && t->toDsymbol(sc) == ti->toDsymbol(sc))) {
ei->exp = new CastExp(loc, ei->exp, type);
}
}
and in D1, it's just (line 1074):
if (!ei->exp->implicitConvTo(type))
ei->exp = new CastExp(loc, ei->exp, type);
If it can't be implicitly converted to a struct, why the heck should an explicit cast be inserted???
PATCH(D1): Delete those 2 lines.
On D2, we need the same logic as for assignment.
Comment #3 by clugdbug — 2009-08-08T22:19:09Z
I have confirmed that after completely removing that section of code, the DMD test suite still passes all tests. I tried to construct a valid case which required that code, but was unable to find one -- it looks as though that code is ONLY creating bugs.
Comment #4 by clugdbug — 2009-09-07T01:16:45Z
*** Issue 3259 has been marked as a duplicate of this issue. ***
Comment #5 by clugdbug — 2009-09-24T06:40:24Z
There must have been something wrong with the way I tested this. It fails one of the test suite tests. Not yet patched.
Comment #6 by clugdbug — 2009-09-24T07:02:06Z
Hmm. It seems that code I commented out is used to invoke static opCall(). That seems very wrong to me, as it's allowing all sorts of other garbage to compile.
Comment #7 by clugdbug — 2009-09-25T01:06:47Z
Here's a revised patch, same place (VarDeclaration::semantic in declaration.c, around line 1083). Replace the existing code with this version, which explicitly allows opCall and NOTHING ELSE:
---
// Dynamic initialization uses static opCall.
if (!ei->exp->implicitConvTo(type))
{ // Look for opCall
if (search_function(sd, Id::call))
{ // Rewrite as e1.call(arguments)
Expression * eCall = new DotIdExp(loc, e1, Id::call);
ei->exp = new CallExp(loc, eCall, ei->exp);
}
}
---
This passes everything in the DMD test suite, except for one line in one test (xtest46.d, test30) which I believe to be broken. I don't think that all of the opDot() and 'alias this' possibilities are tested in the test suite, however.
Here's a tiny test which concisely tests the main features:
struct A { char [] a; }
struct B { long x; }
struct C { int a;
static C opCall(int b) { C q; q.a=b; return q; }
}
static assert(!is(typeof( (){ A s; B q = s; })));
static assert(!is(typeof( (){ B x =1L; })));
static assert(is(typeof( (){ C g = 7; })));
static assert(is(typeof( (){ C g = 7; C h = g;})));