fix Issue 6659 - Destructor in range foreach called after initialization

[9rnsr]

http://d.puremagic.com/issues/show_bug.cgi?id=6659

If for statement has two or more declarations in its Intiliaze part, we should handle their destructors correctly.
This pull transform such for statement like follows.

for (auto x=X(a), y = Y(b); ...; ...) {}

to:

try {
  try {
    for (auto x=((x=X.init).this(a)), auto y=((y=Y.init)).this(b)); ...; ...) {}
  } finally {
    y.~this();
  }
} finally {
  x.~this();
}

The way is basically same as in CompoundStatement::semantic.

[Trass3r]

The old problem: this nice description isn't present as source code comments and will be lost in here.

[9rnsr]

Added comment. Thanks for you point out!

[Trass3r]

np :)

[Orvid]

Shouldn't this require a comma before the true in order for this to be properly valid? This currently triggers a parsing error in Mono-D and I'm trying to figure out if the error is in DMD or if Mono-D should be parsing this as valid.

[yebblies]

Yes and no. It would probably be better if we didn't accept this syntax, but it does appear to be valid in the grammar.

for ( Initialize Test(opt) ; Increment(opt) )

Where Initialize is {scope(success) { assert(value == 1); value = 2;} }.