Bug 14229 – RAII ordering is wrong

Status
RESOLVED
Resolution
FIXED
Severity
major
Priority
P1
Component
dmd
Product
D
Version
D2
Platform
x86_64
OS
Windows
Creation time
2015-02-27T20:57:00Z
Last change time
2015-02-28T02:53:08Z
Assigned to
nobody
Creator
etcimon

Comments

Comment #0 by etcimon — 2015-02-27T20:57:59Z
I'm having issues with the ordering of copy constructors / destructors, and this causes my reference counting utilities to break down. Here's the relevant code: ``` import std.stdio; struct A { this(B!"A->" i) { b = i; writeln("A.__ctor"); } ~this() { writeln("A.__dtor"); } this(this) { writeln("A.__copy"); } B!"A->" b; } struct B(string ident) { this(C!"B1->" i, C!"B2->" j, C!"B3->" k) { c = i; c2 = j; c3 = k; writeln(ident ~ "B.__ctor"); } ~this() { writeln(ident ~ "B.__dtor"); } this(this) { writeln(ident ~ "B.__copy"); } C!"B1->" c; C!"B2->" c2; C!"B3->" c3; } struct C(string ident) { this(int i, int j, int k) { a = i; b = j; c = k; writeln(ident ~ "C.__ctor"); } ~this() { writeln(ident ~ "C.__dtor"); } this(this) { writeln(ident ~ "C.__copy"); } int a; int b; int c; } C!"B1->" getC(A a) { C!"B1->" c = a.b.c; writeln("Returning B1->C"); return c; } void main() { A a; C!"B1->" c1 = C!"B1->"(1,2,3); C!"B2->" c2 = C!"B2->"(4,5,6); C!"B3->" c3 = C!"B3->"(7,8,9); B!"A->" b = B!"A->"(c1, c2, c3); a.b = b; writeln("Getting C"); C!"B1->" c; c = getC(a); writeln("Got C"); } ``` The relevant output is: ---- Getting C B1->C.__copy B2->C.__copy B3->C.__copy A->B.__copy A.__copy B1->C.__copy Returning B1->C A.__dtor A->B.__dtor B3->C.__dtor B2->C.__dtor B1->C.__dtor B1->C.__dtor Got C ---- There's a mistake here. The copy pattern under "Getting C" is fine. From the top down. To be equally fine, destruction pattern should reverse it! It should propagate the destruction tree from the bottom up. ie. you should be seeing: B3->C.__dtor B2->C.__dtor B1->C.__dtor A->B.__dtor A.__dtor Unfortunately, this might end up in the early destruction of an object (segfault), and an absolute nightmare for the poor programmer debugging this while expecting a bottom up destruction pattern.
Comment #1 by etcimon — 2015-02-28T02:53:08Z
I realized this is not an issue, because the destructors are called before the destruction. So, this doesn't map the actual free'ing of the structs.