What happens if ‘out’ parameter is not assigned by the calee?
C# specification is fairly vague about scenarios that can only possibly happen outside of a closed C#-only system. The situations where some parts do not follow the rules lead to “unspecified” behavior. How bad can it get though?
In C# out
calling convention is built on mutual trust between caller and callee. out
parameter is essentially a ref
parameter that callee promises to assign before returning. It makes that promise by marking the parameter with out
attribute. Broken promise here primarily affects definite analysis and features based on that analysis.
Since C# cannot break the protocol, lets use another language that can. VB to the “rescue”.
Imports System.Runtime.InteropServices
Public Class Class1
Public Shared Sub Foo(<Out> ByRef x As Integer)
' Bah, could not be bothered to assign x
End Sub
End Class
And we will call that code from:
static void M()
{
while (true)
{
int x;
ClassLibrary1.Class1.Foo(out x);
System.Console.Write($"{x++} ");
}
}
The output is
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 . . .
Per C# spec a new set of local variables is created when execution enters their containing lexical scope. We are supposed to get a new x
every time we enter the { }
block. In reality compiler reuses the same slot for x
and relies on definite analysis to ensure that you will never know the difference by observing a value left by a previous iteration. The analysis assumes that the value is overwritten by Foo
before we read from x
, but Foo
does not actually do that so the program compiles with undefined behavior.
So far it does not look too bad. Yes, we see the values from the previous iteration, but the behavior seems deterministic.
Well, that is just because we are lucky and x
is always stored in the same location.
Lets try something more complex.
static async Task M()
{
while (true)
{
int x;
await AwaitsSometimes();
ClassLibrary1.Class1.Foo(out x);
System.Console.Write($"{x++} ");
}
}
private static Random random = new Random(42);
static async Task AwaitsSometimes()
{
if (random.Next() % 2 == 1)
{
await Task.Yield();
}
}
Now behavior is:
Release build:
0 0 1 2 3 4 0 0 0 1 2 3 4 0 0 1 0 0 0 0 0 0 0 1 2 0 0 1 0 1 2 3 0 1 2 0 . . .
Debug build:
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 . . .
What is going on?
When AwaitsSometimes
randomly awaits, it forces the caller to suspend and schedule the rest of itself as an asyncronous continuation of Task.Yield
. While doing that, caller saves the values of the local variables in a display storage, so that when resumed, it could continue with the same state. It does not, however, save the value of x
considering that as pointless. For all it knows from the definite assignment analysis, at the point of await AwaitsSometimes()
, x
contains undefined junk which will be overwritten before the variable is used anyways.
Since we go into suspension sometimes, we observe the loss of the value sometimes. Not the behavior you want from an otherwise straightforward program…
Why then Debug behavior is different?
In Debug builds compiler makes all locals in the same scope as await
to be saved into the display storage without consulting with definite assignment analysis. That is done to make debugging experience better. When resuming after await, all locals will keep their values and can still be examined in the debugger. When values of locals are not actually used by the program after await, such practice could be very wasteful.
That makes us “lucky” again in Debug, but Release shows that behavior is truly undefined.
Garbage in, garbage out. Do not break promises.