Bug 9051 – Passing an immutable global with post-blit to a CTFE function

I'm wondering if this is supposed to work. By all means I don't see why it has to try this(this) in this case (and moreover fail). struct Set{ uint[] arr; this(this){ } //comment out to make it compile } pure auto getMeASet(uint[] arr) { return Set(arr); } immutable set = getMeASet([1,2,3,4]); pure auto getSecond(in Set set) { return set.arr[1]; } pragma(msg, getSecond(set)); And the diagnostic spits all kinds of nonsense: D:\D\ctfe_glob.d(20): Error: variable ctfe_glob.__cpcttmp6 of type struct const(Set) uses this(this), which is not allowed in static initialization D:\D\ctfe_glob.d(20): while evaluating pragma(msg, getSecond((const const(Set) __cpcttmp6 = Set([1u, 2u, 3u, 4u]); , __cpcttmp6)))

The error message is not coming from CTFE. Here's a simple case from a comment in the compiler source code (declaration.c): /* The problem is the following code: * struct CopyTest { * double x; * this(double a) { x = a * 10.0;} * this(this) { x += 2.0; } * } * const CopyTest z = CopyTest(5.3); // ok * const CopyTest w = z; // not ok, postblit not run * static assert(w.x == 55.0); * because the postblit doesn't get run on the initialization of w. */ I think that if we didn't call this(this), wrong code would result. There is at least a diagnostic bug here - if it must behave this way, the code that creates the comma expression should be issuing the error. Interestingly, if you move it inside a function, the result is different: void foo() { pragma(msg, getSecond(set)); } qqq.d(19): Error: static variable set cannot be referenced at compile time qqq.d(19): called from here: __cpcttmp6.__cpctor(set) qqq.d(19): called from here: getSecond((const const(Set) __cpcttmp6 = __cpcttmp6.__cpctor(set); , __cpcttmp6))

(In reply to comment #1) > The error message is not coming from CTFE. Here's a simple case from a comment > in the compiler source code (declaration.c): > > /* The problem is the following code: > * struct CopyTest { > * double x; > * this(double a) { x = a * 10.0;} > * this(this) { x += 2.0; } > * } > * const CopyTest z = CopyTest(5.3); // ok > * const CopyTest w = z; // not ok, postblit not run > * static assert(w.x == 55.0); > * because the postblit doesn't get run on the initialization of w. > */ > > I think that if we didn't call this(this), wrong code would result. I agree in general. And But given that getMeASet(uint[] arr) is an r-value. Surely the result of it should be just moved? The other thought is: can't this(this) be run at compile-time then? What are limitations? > There is at least a diagnostic bug here - if it must behave this way, the code > that creates the comma expression should be issuing the error. > > Interestingly, if you move it inside a function, the result is different: > The same thing happens if you try passing set by reference be it const or otherwise. > void foo() > { > pragma(msg, getSecond(set)); > } > > qqq.d(19): Error: static variable set cannot be referenced at compile time > qqq.d(19): called from here: __cpcttmp6.__cpctor(set) > qqq.d(19): called from here: getSecond((const const(Set) __cpcttmp6 = > __cpcttmp6.__cpctor(set); > , __cpcttmp6))

(In reply to comment #2) > (In reply to comment #1) > > The error message is not coming from CTFE. Here's a simple case from a comment > > in the compiler source code (declaration.c): > > > > /* The problem is the following code: > > * struct CopyTest { > > * double x; > > * this(double a) { x = a * 10.0;} > > * this(this) { x += 2.0; } > > * } > > * const CopyTest z = CopyTest(5.3); // ok > > * const CopyTest w = z; // not ok, postblit not run > > * static assert(w.x == 55.0); > > * because the postblit doesn't get run on the initialization of w. > > */ > > > > I think that if we didn't call this(this), wrong code would result. > > I agree in general. And But given that getMeASet(uint[] arr) is an r-value. > Surely the result of it should be just moved? > > The other thought is: can't this(this) be run at compile-time then? What are > limitations? Yeah, that's kind of like bug 7988. The calling of this(this) should be part of the CTFE'd bit. CTFE could cope with it, no problem, it just doesn't get a chance. BTW if we moved the this(this) into CTFE, the example I posted would also work!

The postblit should not be used as it has impossible to fix flaws. Development for it has halted. The code works if the copy constructor is used: struct Set{ uint[] arr; this(const ref Set) {} //comment out to make it compile } pure auto getMeASet(uint[] arr) { return Set(arr); } immutable set = getMeASet([1,2,3,4]); pure auto getSecond(in Set set) { return set.arr[1]; } pragma(msg, getSecond(set)); I'm gonna close this as WONTFIX.