Late At Work last night, we were trying to figure out why something was failing. A validation check was failing when it shouldn't have been.
We ended up adding a print statement to this code (disassembled from Reflector in order to check that the code was actually what we had written):
public static string Redacted(string name, DateTime lastModified)
{
long ticks = lastModified.Ticks;
if ((ticks != (ticks - (ticks % 10000L))) &&
(lastModified != DateTime.MaxValue))
{
Log.Debug(string.Format("Last Modified Date = '{0}'. Ticks = '{1}'. TicksCalc = '{2}'",
lastModified.ToString("dd/MM/yyyy hh:mm:ss.fff"),
ticks, ticks - (ticks % 10000L)));
It printed (reformatted):
Last Modified Date = '22/03/2011 12:16:22.000'.
Ticks = '634363497820000000'.
TicksCalc = '634363497820000000'
But the condition is that "ticks
" (which is equal to Ticks printed abo开发者_高级运维ve) is not equal to "(ticks - (ticks % 10000))
" (which is equal to TicksCalc)! 634363497820000000 != 634363497820000000?!
In order to determine what is going on here, we added another two statements:
long ticks = lastModified.Ticks;
/* Added following two lines: */
long num2 = ticks - (ticks % 10000L);
Log.Debug((ticks == num2).ToString());
/* */
if ((ticks != (ticks - (ticks % 10000L))) &&
(lastModified != DateTime.MaxValue))
{
Log.Debug(string.Format("Last Modified Date = '{0}'. Ticks = '{1}'. TicksCalc = '{2}'",
lastModified.ToString("dd/MM/yyyy hh:mm:ss.fff"),
ticks, ticks - (ticks % 10000L)));
As it should have, this one printed true
(when testing with the same value), and didn't write the second line.
Feeling a bit lost, we then removed the two lines again, recompiled, and reran. The original behaviour repeated itself.
This morning, I recorded a video.
The video first of all shows hitting a breakpoint in the method using the 'broken' code, then rebuilding and rerunning using the 'working' code. Note that even though the debugger displays that the if
condition evaluates as to false
, the body is still entered.
I've seen things like this happen before when observed by the debugger, due to the debugger forcing some things to be evaluated, but this happens whether or not the debugger is employed.
Furthermore, this only happens in Release mode (i.e. with JIT optimizations enabled).
Here are the disassembled methods for both versions: working, not working. I can't really read assembly, so am posting them here in the hopes of elucidation.
I hope that the answer isn't something obvious that I've overlooked completely...!
Edit: Here is the IL. I don't think there's anything wrong with it because it decompiles to the correct C#:
- Not working
- Working
Update:
Confirmed as a bug by Microsoft, to be fixed in the next release.
I experimented a bit with simplified code: http://nopaste.info/2c99a0e028_nl.html
The most interesting variation is:
static readonly long variableZero=0;
const long constZero=0;
public static void Broken2( long ticks2)
{
long ticks = ticks2+variableZero;
if (ticks != (ticks - (ticks % 10000L)))
{
string.Format("Last Modified Date = '{0}'. Ticks = '{1}'. TicksCalc = '{2}'",
"n/A",
ticks, ticks - (ticks % 10000L)).Dump();
}
}
If I replace variableZero
with constantZero
it works.
So I'm pretty sure it is either a jitter or a compiler bug.
I've filed a bugreport on MS Connect: https://connect.microsoft.com/VisualStudio/feedback/details/671105/jitter-or-c-compiler-bug#details
Update: The strange behavior only occurs if no debugger is attached. i.e. when Jit optimization is enabled. So I'm pretty sure it's a jitter bug.
And for people without linq-pad there is now a plain C# console project: http://nopaste.info/00a0e37328_nl.html
Check out this thread.
If statement weirdness in Visual Studio 2008
It comes down to this, you can't trust the debugger all the time.
To "fix" that if statement, add an empty else {} statement to it. The debugger will work as expected.
That does indeed look like an - ahem - jitterbug. Can you set a break-point on the "if" statement and show us a screenshot of the disassembly view after it hits?
I had something similar a while ago. In my case it had to do with the fact that i was comparing 2 integer values where one value was actually a reference to a boxed integer and the other was a real primitive value.
The thing is that if you print out the value of the boxed integer and the primitive, they look the same, but comparing them is another thing. You'll get a reference comparison in stead of a value comparison.
The answer is easy:
long ticks = lastModified.Ticks;
long num2 = ticks - (ticks % 10000L);
if ((ticks != num2) && (lastModified != DateTime.MaxValue))
{ do your thing here! }
That's crazy. Have you tried, for no good reason, to reorder the if statement?
if (lastModified != DateTime.MaxValue && ticks != (ticks - (ticks % 10000L))
Also, if this doesn't work (as it shouldn't, considering it shouldn't be a problem in the first place), can you show the actual IL for the code in in the problematic form?
One other thing, couldn't the ticks
check be simplified to:
(ticks % 10000L) != 0
精彩评论