Bug 5500 – Appender.put causes an invariant violation in struct with overloaded opAssign

The execution of this code throws an assertion error. import std.array; struct Foo { int a, b; Foo opAssign(Foo rhs) { a = rhs.a; b = rhs.b; return this; } invariant () { assert (a <= b); } } unittest { auto app = appender!(Foo[]); auto foo = Foo(1, 2); app.put(foo); } Stack trace: core.exception.AssertError@test(13): Assertion failure ---------------- 5 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x00009a61 onAssertError + 65 6 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x00012e82 _d_assertm + 30 7 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x00002593 void test.__assert(int) + 27 8 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x000025ef void test.Foo.__invariant() + 23 9 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x000025ae test.Foo test.Foo.opAssign(test.Foo) + 14 10 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x00002bfd void std.array.Appender!(test.Foo[]).Appender.put!(test.Foo).put(test.Foo) + 113 11 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x0000262b void test.__unittest1() + 55 12 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x00002573 void test.__modtest() + 11 13 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x0000a121 extern (C) bool core.runtime.runModuleUnitTests().int __foreachbody247(ref object.ModuleInfo*) + 45 14 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x0000560f int object.ModuleInfo.opApply(scope int delegate(ref object.ModuleInfo*)) + 79 15 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x0000a012 runModuleUnitTests + 134 16 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x00013202 extern (C) int rt.dmain2.main(int, char**).void runAll() + 38 17 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x00013156 extern (C) int rt.dmain2.main(int, char**).void tryExec(scope void delegate()) + 38 18 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x000130e7 main + 179 19 test.d.E6E9779DFA87AC0EBDE59516F9288677 0x0000255d start + 53

(In reply to comment #0) > The execution of this code throws an assertion error. > I can't seem to reproduce this. dmd 2.051, linux

(In reply to comment #1) > (In reply to comment #0) > > The execution of this code throws an assertion error. > > > > I can't seem to reproduce this. > > dmd 2.051, linux Weird. I have this error every time I run the code in - dmd 2.050 & 2.051 Mac OS X (differents systems) - dmd 2.051 Linux (Ubuntu 10.10) / Parallels Desktop I will try on a PC with MSWindows this evening.

I'm not getting this to happen with dmd 2.051 on Arch Linux either (Phobos and druntime are from git). However, if I use the example that you originally gave on D.learn: import std.array, std.datetime; unittest { auto app = appender!(Interval!Date[]); auto interval = Interval!Date(Date(2000, 1, 1), Date(2011, 2, 3)); app.put(interval); // Error: datetime.d(20208): Invariant Failure: begin is not before or equal to end. } it happens some of the time but not all of the time. So, whatever the bug is exactly, it's non-deterministic, and it may easily get affected by the OS that you're on. It may be that some slight tweaks would make it pop up on boxes which currently can't reproduce it. I'm convinced that there is a bug here, but it's non-deterministic enough, that depending on the code and the machine, it could either be reproduced easily or only with difficulty.

Actually, I think I know what's happening (roughly speaking). This problem is caused by bug# 5058. Appender has uninitialized data which it then tries to assign to using the assignment operator. Because the invariant is called before opAssign (which I argue in bug# 5058, that it shouldn't be), the invariant is called on garbage data. Sometimes, that garbage data violates the invariant, and sometimes it doesn't. So, unless there's something that Appender can do to work around bug# 5058, it's likely not a bug in Appender so much as it's a bug in how invariants work with regards to opAssign.

> Actually, I think I know what's happening (roughly speaking). This problem is > caused by bug# 5058. I think you are right, these bugs look very similar. It definitely is related to opAssign, since it never occurs when opAssign is not redefined. This would also mean that the compiler-generated opAssign is either not calling invariant() or calling it at the right time, after data is properly initialized. But I don't know what happens behind the scene in this matter.

I can't reproduce this: DMD 2.062 LDC2 (based on 2.061) Arch Linux x86_64 Can anyone else reproduce this? This bug is quite old...

Cannot repro in 2.065. Closing as WORKSFORME.