WIP: add safer unwrap

[DaAitch]

Hello,

this PR prevents that ObjectWrap::Unwrap can be used with any type. It has to be a type having ObjectWrap<T> as sub class.

I spend some time finding a bug in my appliction, which did not segfault, because I did something like this test/objectwrap-saferunwrap.cc and it compiles and the reason I found out something is really going wrong is, that the stacktrace shows calling a method ->A() actually called ->B(). Then I realized, that it might be is a vtable pointer fixup problem and understand that of course it's possible, but not allowed to reinterpret cast void * to any type but the the original one.

So to prevent that s.o. tries to seductively unwrap an ObjectWrap<Car> or ObjectWrap<Truck> to any sub class like IVehicle (which obviously is not a ObjectWrap) it should not compile.

What do you think?

If this has a chance to land, we my need "should not compile" tests.

[mhdawson]

I like the concept, do you have ideas of how we'd add the "should not compile" tests?

[gabrielschulhof]

We should call the function SafeUnwrap not SaferUnwrap and the C preprocessor variable should be NAPI_SAFE_UNWRAP. If present, it should do the safe napi_unwrap, otherwise it should do the regular napi_unwrap. The way it is now is kind of confusing to me.

@DaAitch is this related to #732 (multiple inheritance)?

@mhdawson maybe we should have a second target in the test/binding.gyp file which only gets run in the CI and is considered failed if it successfully compile. Might be kind of a crude check though, because it might be quite an effort to start analyzing the compiler's stderr output to see if it generated the right error message. Our many supported platforms each have a compiler that produces its own individual message, and the message may change with the version of the compiler, so if we analyzed compiler stderr output we might be signing up for quite a maintenance burden.

The PR needs a rebase.

[mhdawson]

@gabrielschulhof having that second target which fails makes sense to me :)

[mhdawson]

@DaAitch any chance you can rebase so we can move forward.

[DaAitch]

@mhdawson sorry for the delay, I didn't read notifications for a long time :(. If you still think this may help, please feel free to copy and adapt my code and close this PR. At present I think I'm too far away from this topic. Sorry for that

[mhdawson]

I think at this point based on the last comment we'll close, and somebody can reopen in the future if they'd like to get it landed.