A very strange bug. DMD 2.058 64-bit Linux

Caligo iteronvexor at gmail.com
Sun Feb 26 12:06:06 PST 2012


bug.d
---------------->8---------------->8----------------
@trusted:

import std.datetime : benchmark;
import std.stdio    : writefln, writeln;

alias double Real;

void ben(alias fun)(string msg, uint n = 1_000_000) {
  auto b = benchmark!fun(n);
  writefln(" %s %s ms", msg, b[0].to!("msecs", int));
}

struct Matrix(int row, int col) {

private:
  alias row Row;
  alias col Col;
  alias Real[Row * Col] Data;

public:
  Data _data = void;
  alias _data this;
  this(const Real[Row*Col] data) pure nothrow { _data = data; }
}

M inverse(M)(const auto ref M m)  {
  writeln(m[]);
  M minv = m;
  return minv;
}

unittest {
  alias Matrix!(4, 4) Matrix4x4;
  auto m9 = Matrix4x4([4, 0, 0, 0, 0, 0, 2, 0, 0, 1, 2, 0, 1, 0, 0, 1]);
  ben!( {auto r = inverse(m9);} )("4x4 inverse:");
}
----------------8<----------------8<----------------

t1.d
---------------->8---------------->8----------------
import std.stdio;

void main(){ }
----------------8<----------------8<----------------

It took me a long time to pinpoint this because it's tricky to trigger the bug.

Once you have those two files, compile with this:

dmd -unittest t1.d bug.d

and then run t1:

./t1

The output you get should look like this:

...
[0, 4, 0, 0, 0, 0, 0, 2, 0, 0, 1, 2, 0, 1, 0, 0]


Obviously the output is wrong.  'm9' for some reason is getting
overwritten.  In my project this caused big problems because there are
other m# with different values, and their values would literally get
copied to m9.  Calling inverse() on m9 then would fail because the
other matrices are not invertible.  Placing a writeln() in inverse()
helped me realize that what was being passed to inverse() was being
modified somewhere.  I'm still now sure how m9 is being modified.

Another point, compiling with this:

dmd -unittest bug.d t1.d

and then running bug:

./bug

doesn't trigger the bug.


Could someone else please confirm this behavior?


More information about the Digitalmars-d-learn mailing list