On 10/16/20 11:11 AM, Ali Çehreli wrote:
On 10/16/20 6:12 AM, tchaloupka wrote:> struct Foo { > Bar bar; > bool err; > > ~this() { > // scope(failure) destroy(bar); // < this fixes the Bar > destructor call > enforce(!err, "Test err");Well, that check means "cannot continue", which means the compiler stops executing the destruction code because it can't. (It would a serious bug if it continued execution in a state that the program knows to be invalid.)
The destruction of members is outside the destructor's purview. It can't turn the destruction off, so it should logically be considered part of an enclosing function.
Note that enforce is throwing an *Exception*, not an *Error*, it does not mean the program is in an invalid state. Throwing an Error, the compiler is allowed to not clean up after itself.
Imagine if throwing an Exception disabled RAII in any enclosing function scopes because it considered that an invalid state.
Conceptually, bar's destructor is called on that closing brace but we decided to abort mission earlier. Your calling destroy(bar) may or may not be wrong in case of 'err'. Only you know at that point.
If that is the case, a hand written destructor should turn off automatic destruction of members. Either the compiler handles member destruction or it doesn't.
-Steve
