Compiler error CS0843 when built against mono-2.8

[Andrew Jorgensen]

Compiling Evolution.dll...
warning CS2002: Source file `./src/AddressContactDetailItem.cs' specified multiple times
./src/ContactItemSource.cs(39,17): error CS0843: An automatically implemented property `Evolution.ContactAttribute.Detail' must be fully assigned before control leaves the constructor. Consider calling default contructor
./src/ContactItemSource.cs(39,17): error CS0843: An automatically implemented property `Evolution.ContactAttribute.Key' must be fully assigned before control leaves the constructor. Consider calling default contructor
Compilation failed: 2 error(s), 1 warnings

[Robert Dyer (psybers)]

Call me crazy, but this looks like an error in the 2.8 compiler! 'this.Key' and 'Key' are scoped to the same reference and thus their analysis seems to fail incorrectly!

[Andrew Jorgensen (ajorg)]

Nope, they are not the same. When you do "Key = key" you are calling the accessor, which is a method. When you do "this.Key = key" you are making an assignment.

[Robert Dyer (psybers)]

And the inter-procedural analysis should be able to determine that the Key accessor (when setting) does a 'this.Key = key' and thus the analysis (as I said!) is flawed.

Please don't change bug status.

[Robert Dyer (psybers)]

Is the problem here that there is no default constructor defined to assign those 2 properties? Thus if you call a default constructor on this type you wind up with a case where the use-def analysis would indeed be correct in stating those 2 properties were never assigned before use?

[Robert Dyer (psybers)]

And FYI I am digging through the official language specification for C# to look up the definite assignment rules. Just because I think it should be this way does not necessarily make it so. I'll get back to you.

[Andrew Jorgensen (ajorg)]

I did try to compile a simpler example of this on .NET and it does emit the same error. Unfortunately it still emits the error when I change it to this.Key. Mono, at least, does allow that.

[Robert Dyer (psybers)]

Ok that makes more sense to me, because as I said 'this.Key' and 'Key' are semantically equiv.

The question still is why the compilers can't figure out that property is assigned, when it clearly is. If you figure out a solution for .Net let me know - I would prefer that over a solution that works on Mono but is technically still a bug!

[Andrew Jorgensen (ajorg)]

I am told by my team here that my patch was not correct (and should cause another error) and that the correct fix is more like the attached.

16:13 <alan_> ajorg, The correct fix is to append " : this ()" to the constructor to call the default parameterless one

Also, there is a discussion of this problem on stackoverflow

16:15 <lewing1> ajorg: http://stackoverflow.com/questions/420433/automatic-properties-and-structures-dont-mix

[Robert Dyer (psybers)]

Have you verified the new patch with the 2.8 compiler?

[Andrew Jorgensen (ajorg)]

Yes, I just built with these changes (the actual attached patch change line endings when I tried it! no idea why) and it does compile on mono 2.8.

[Andrew Jorgensen (ajorg)]

Some fuzz on line-endings may be required.

[Robert Dyer (psybers)]

Well now I feel stupid, this whole time I thought we were dealing with a class (not a struct). I am really not sure why they used a struct there... but your patch seems valid in this case.

Fixed in rev 695.