False compile error: Use of unassigned local variable even when code is correct

[CzBuCHi]

Describe the bug
Created simple project where stryker throws compile error for valid code -> threads code as 'noCoverage' -> invalid report

Logs
Zip file contains test solution and result of dotnet stryker -L > console.txt command

Expected behavior
No error

Desktop (please complete the following information):

• OS: Windows 11
• Framework Version net4.8 and net 8.0
• Stryker Version 4.8.1

Additional context
Test solution: Test.zip

[dupdob]

Hi
I suppose you are referring to this log line:
[15:11:11 WRN] Stryker.NET encountered a compile error in C:\projects\Test\test\Class.cs (at 11:15) with message: Use of unassigned local variable 'value' (Source code: value) [15:11:11 INF] Safe Mode! Stryker will flag mutations in Test as compile error.

This is normal behavior and there is no issue. As you can see, this is logged as a warning and not as an error.
I suppose we should improve on the log message. Here is the explanation
One of the mutation generated by Stryker resulted in a compilation error, probably when trying to elicit this line value = new object(); in the Test method.
In general, Stryker removes the offending mutation and tries compilation again. But sometimes, Stryker is not able to pinpoint the mutation causing said compilation error. When this happens, Stryker triggers safe mode and removes every mutation from the method. This ensures the code can be safely compiled and that Stryker can proceeds with testing mutations.

The offending mutation(s) will be in 'compilation error' status and will be counted as killed mutations.

I hope this clarify things. I will close this issue, but as it looks like this is a minimal project designed for reproduction purposes, feel free to reopen it with supplemental questions/information if you think your problem is not solved.

[CzBuCHi]

my issue is that in this example when i marked my code preprocessor directives based on mutants:

namespace Test;

public class Class
{
    public Type Test()
#if MUTANT_0
    {}
#else    
    {
        object value;
        try
#if MUTANT_1
        {}
#else 
        {
            value = new object();
        }
#endif
        catch (Exception exc) 
#if MUTANT_2
        {}
#else
        {

#if MUTANT_3
            ;
#else 
            Console.WriteLine(exc.Message);
#endif

#if MUTANT_4
            ;
#else 
            throw;
#endif
            
        }
#endif // MUTANT_2
        return value.GetType();
    }
#endif // MUTANT_0
}

and tried to compile each mutant individually i got those:

mutant 0: CS0161 'Class.Test()': not all code paths return a value
mutant 1: CS0165 Use of unassigned local variable 'value' on return value.GetType();
mutant 2: CS0165 Use of unassigned local variable 'value' on return value.GetType();
mutant 3: NO ERROR
mutant 4: CS0165 Use of unassigned local variable 'value' on return value.GetType();

so mutant 3 should be marked as survived (as test is not checking if Console.WriteLine was called or not) but is marked as compile error ...

from logs it seems that stryker 'first' compilation is with mutant 1,2 or 5 already applied -> compiler error CS0165
then it will try 'second' compile with mutant 0 -> compiler error CS0161 but reports CS0165 again - it seems like error message is not updated -> when stryker finally try to compile mutant 3 which return no error it will keep error from previous attempts ...

imho: 'first' compilation should be with no mutants applied - and if that throws compiler error then mark all mutants as 'CompileError'

PS: i dont have permission to reopen ...

[dupdob]

Hello
As i said:

Stryker triggers safe mode and removes every mutation from the method. This ensures the code can be safely compiled and that Stryker can proceeds with testing mutations.

So yes, it will flag mutant #3 as compilation error as well because it does not know better.

Stryker injects all mutants at once and compiles the project. If it fails, it looks at which lines fail and removes mutations on these lines. In your example, mutant #1 will cause an error on this line return value.GetType();.

This is a known limitations that is surprisingly difficult to overcome:

1. building the project once for each mutation will take an inordinate amount of time
2. identifying the mutation(s) responsible for this type of error requires tracing data flows with is obviously complex and would not offer a generalized solution
3. we also looked into having an incremental remove & restore strategy within the method, which is also hard to due to how Roslyn APIs are designed.

[CzBuCHi]

Just an idea: to remove user confusion ( why is stryker telling me i have compiler error when i have none??? )

stryker could print modified source code (with all those if (StrykerBlaBla.MutantControl.IsActive(1))... injected parts) to console / file
to make clear ehere those compiler errors are .... (you already doing that to log in --dev-mode so its should not be that hard)