Direct Mapping for boolean arrays

[ncruces]

This is consistent with how these are handled in interface mapping.

Also boolean is ffi_type_uint32 everywhere in the source code, except here.

[matthiasblaesing]

In principal I agree with the direction that was started here, but it collides with the current size definition of boolean in JNA (it is defined to 4 bytes). See:

https://github.com/java-native-access/jna/blob/master/src/com/sun/jna/Native.java#L1296

This would not be in line with a 1 byte wide mapping.

There is a request for supporting C99 bool:

#721

I started some work on this, but it is still unready and I think in the wrong direction:

https://github.com/matthiasblaesing/jna/tree/boolean

I think a way could be:

• map the boolean in JNA compatible with JNIs jboolean, that is 8 bit unsigned (http://docs.oracle.com/javase/7/docs/technotes/guides/jni/spec/types.html)
• introduce a NativeBoolean (as the companion to NativeLong), that maps the platforms bool width

This should go into the 5.0.0 branch, as it invalidates assumptions about sizes and mappings.

[ncruces]

Yes, I understand it does, and I'm not necessarily very happy with the change. Still two changesets so...

8f2e773 IMHO is a simple fix.
Probably doesn't make much of a difference, but if it does, this is the only place where boolean is signed.

ba7881f simply makes direct mapping consistent with interface mapping.
Consistently wrong? Maybe, but consistent.

Also, note that the same reasoning applies to both boolean and char arrays:

• char maps to wchar_t, char[] already maps to jchar* (always 16-bit unsigned);
• boolean maps to uint32_t, boolean[] already maps to jboolean* (always 8-bit unsigned).

There's quite a lot of effort to handle the difference for String, but no such effort was made for char[].

So the only change here is making boolean[] work for direct mapping. I'm not even using this (likely no one is) so feel free to ignore it. But it is how it already works for interface mapping, and it seemed trivial enough.

---

Now, this is just my opinion, and I'm very, very new to JNA.

But in a sense, I feel JNA already tries to do a bit too much, and would avoid having it do even more. When things are passed by reference, I can do the marshaling by myself (e.g. convert between encodings). It's only when passing arguments and return values by value that JNA must help (with JNI and the ABI), so it doesn't at all bother me that arrays are not marshaled where simple values are.

I'll have a look at your boolean branch. But feel free to ignore this if I'm slowing you down.

[matthiasblaesing]

Please: No need to apologize, in fact I'm glad your are looking into this! Your points are well taken and your reasoning is sound.

Two things I'd like to ask you: Please add an entry to CHANGES.md. For the first PR I did that:

https://github.com/java-native-access/jna/blob/master/CHANGES.md

The second: Please rebase your changes on top of JNA master (I just merged your first PR)

[ncruces]

Regarding your branch, using stdbool.h (and custom mapping other boolean types) seems reasonable to me.

Any inconsistencies between jboolean/bool won't be much different from inconsistencies between jchar/wchar_t. It's unfortunate that C was late to standardize these things, but C99 was 18 years ago, and if there's ever going to be any standard boolean type, stdbool.h is it.

The real issue, I think, is breaking compatibility with previous JNA versions, and the mountain of APIs returning an int that would better map to a Java boolean. Then you can go the other way: keep things as they are, come up with a NativeBool mimicking NativeLong?

[matthiasblaesing]

@ncruces I merged this PR. Thanks for looking into this. I'll look into rebuilding the pre-build native parts in the next few days.

With regard to StdBool I think the only realistic option is the introduction of a NativeBoolean. At some point I'll look into this again, but not in the near future :-)