Bug 3879 – compile-time assertion error in mtype.c: foreach through associative array w/ templated-class as value

I'm working on a file with these classes so far: abstract class State(T:GameEntity) { string _name; @property { const string name() const { return _name; } this(string name) { _name = name; } (and some empty virtual function decls that get defined by derived classes) } // and then the actual implementation for the various State nodes will be done through derived classes // so if I had a class DogRobot that derives from GameEntity, then it would // internally use a StateMachine!(DogRobot), which itself would be using // State!(DogRobot) for its states. Somewhere in DogRobot.d it would also // define those new State!(DogRobot)-derived classes that would then be // added to the State Machine via addState() class StateMachine(T:GameEntity) { ... State!(T)[string] _states; // THIS IS LINE 100 IN THE ERROR MESSAGES MENTIONED LATER -- assoc array for quickly retrieving states by their names // public, actual state objects added to the machine here void addState(State!(T) state) { ... (compiles fine) ... } void linkStates(string fromName, string toName, string eventName) { ... (compiles fine) ... } // so far so good, further down I have a function called printGraph() for debug output, and this is where I've run into assertion problems with DMD 2.040 compiler void printStates() { std.stdio.writefln("States:"); // each and every one of these below causes a compile-time error, // though different ones sometimes: // this leads to error: "__overloadset is not a template" on line 100 foreach (string key, State!(T); _states) std.stdio.writefln(" %1", key); // the rest of these each, individually, result in an assertion error // at compile-time which pops up twice. // In the actual code I commented-out the previous problematic lines // before trying the next approach/flavor. DMD 2.040 compiler's error // for each of these reads as follows: // // Assertion failure: 'impl' on line 3886 in file 'mtype.c' // // abnormal program termination // Assertion failure: 'impl' on line 3886 in file 'mtype.c' // // abnormal program termination // Various problematic code for THIS error is as follows: // first attempt foreach (string k; _states.keys) std.stdio.writefln(" %1", k); // approach #2 foreach (State!(T) s; _states.values) std.stdio.writefln(" %1", s.name); // approach #3, but char[] k for a string-type key is probably // incorrect in a different way I'm guessing... just a desperation try foreach (char[] k; _states.keys.sort) std.stdio.writefln(" %1", k); // approach #4 typedef State!(T) tstate; foreach (tstate s; _states.values) std.stdio.writefln(" %1", s.name); } // end printStates() } // end StateMachine class // end code snippet In the meantime I can probably get around this problem by getting rid of templates entirely and very carefully using delegates to act as "plugged-in behavior"... I was following the approach in an AI book I have, but there are other ways I can go about doing the same thing, I'm sure.

This sounds like a duplicate of bug 3692.

(In reply to comment #1) > This sounds like a duplicate of bug 3692. I took a look at that bug and did a little further tinkering. It seems you might be correct. I just removed the std.variant import directive that I hadn't mentioned (not thinking it had anything to do with the problem) and commented-out an unrelated class definition in the same file that used Variant, and the results are now: 1) The first line, that gave "__overloadset is not a template on line 100", now compiles fine. 2) The "foreach (string k; _states.keys)" approach now compiles fine. 3) The "foreach (State!(T) s; _states.values)" approach now compiles fine. 4) The "foreach (char[] k; _states.keys.sort)" approach now gives an appropriate error message (cannot implicily convert expression etc) instead of crashing with an assertion. 5) The final approach where I do declare State!(T) as a new typedef called 'tstate' and then do "foreach (tstate s; _states.values)" gives an error message instead of crashing -- "cannot implicitly convert expression"... which is surprising, though unrelated to what this bug report is about, and I'm not sure if I should submit another report on this particular thing because maybe doing typedefs on templated-classes is intentionally not-allowed? Anyway, I think you're right... removing any and all mention of std.variant from the source file changed the situation entirely.

Since bug 3692 is fixed in DMD2.041, this one should be working now. Reopen if it that is not correct.