UTFStringRef.set() throws IllegalArgumentException

[Robmaister]

I'm trying to wrap a library that has some structs that expose bare char* strings. When I try to set one of these values using UTFStringRef.set, I get the following exception:

Exception in thread "main" java.lang.IllegalArgumentException: negative size: -4
        at jnr.ffi.provider.jffi.TransientNativeMemory.allocate(TransientNativeMemory.java:46)
        at jnr.ffi.provider.jffi.NativeMemoryManager.allocateDirect(NativeMemoryManager.java:41)
        at jnr.ffi.Struct$UTFStringRef.set(Struct.java:2364)
        at test.TestProject.setValue(TestProject.java:207)
        at test.SimpleExample.main(SimpleExample.java:8)

I can include some code, but unless I'm mistaken, there's a relatively straightforward bug here:

https://github.com/jnr/jnr-ffi/blob/master/src/main/java/jnr/ffi/Struct.java#L2364

valueHolder = getRuntime().getMemoryManager().allocateDirect(length() * 4);
valueHolder.putString(0, value, length() * 4, charset);

I believe that both instances of length() here were intended to be value.length() instead. By default, the constructor sets the value of length() to Integer.MAX_VALUE, and it would make sense to allocate a buffer of the same size as the incoming string.

Also along the same lines, does String.length() in Java include the size of a null terminator? Should it perhaps be (value.length() + 1) * 4?

If I'm completely off track here let me know and I'll include some code with what I'm trying to accomplish

[Robmaister]

Due to the incompatibility of JNR with Android, I'm going to have to stop using JNR and write my bindings directly with JNI. I should have probably checked that before starting to write my bindings.

Anyways, the immediate solution I came up with was to drop back to Pointer fields and manually marshal strings back and forth via getString(0) and putString(0) (with a null check on the result of get() from the struct field).

[headius]

Thanks for the workaround info, and I'm sorry we did not see this issue! Did you ever figure out what was causing the problem in JNR?

[Robmaister]

No worries! I'm not really fantastic about keeping up with my own repos haha

If I remember correctly changing length() to value.length() would have fixed it. I ended up having to focus on other stuff and now work elsewhere, so I can't really go back and try running it again with a patched version.

If you want to create a reproducible example, create a struct with a string field and try to set it on the Java side before passing that struct over to the native side.

At a higher level, we were better off re-implementing the protocol we were trying to port with Android's socket library for better integration into the API (handler/looper stuff) so I doubt whoever is continuing my work would be working with the same bindings I wrote.

[charleskorn]

I am still able to reproduce this issue - and I agree with @Robmaister's assessment:

If I remember correctly changing length() to value.length() would have fixed it.

This looks like it would resolve the issue.

[headius]

Thanks for waking this one up with a fix!