None of the JNI 1.1 function pointers were optional, and neither are any of the function pointers in later versions of the spec.
Having the Option implies that null pointer checks are required but, more notably, they also suggest that null pointer checks could be used for JNI >= 1.2 pointers, which could be a dangerous mistake since these are effectively extending beyond the table of pointers that was defined for JNI 1.1 and should be assumed to be invalid pointers that mustn't be touched.
It's also notable that we sometimes have to call GetVersion to determine the full set of pointers that are valid.
Recently the use of Option also raised some questions about our ability to, infallibly, handle Rust panics when we want to map a panic to a Java exception via JNI:
For now I've just created this as a draft PR since I also think it could be worth considering changing JNINativeInterface_ into a union that would look something like:
So then it becomes clearer that you need to access the function pointers according to the known version and in doing so you gain some safety because Rust can stop you from trying to call a JNI 1.2 method while dereferencing the .1_1 functions.
The .1_1 table would also include the padding for the reserved null functions in the middle of the table but can rename them so it would be harder to accidentally dereference them.
None of the JNI 1.1 function pointers were optional, and neither are any of the function pointers in later versions of the spec.
Having the
Optionimplies that null pointer checks are required but, more notably, they also suggest that null pointer checks could be used for JNI >= 1.2 pointers, which could be a dangerous mistake since these are effectively extending beyond the table of pointers that was defined for JNI 1.1 and should be assumed to be invalid pointers that mustn't be touched.It's also notable that we sometimes have to call
GetVersionto determine the full set of pointers that are valid.Recently the use of
Optionalso raised some questions about our ability to, infallibly, handle Rust panics when we want to map a panic to a Java exception via JNI:https://github.com/jni-rs/jni-rs/issues/432#issuecomment-1470934365
--
For now I've just created this as a draft PR since I also think it could be worth considering changing
JNINativeInterface_into a union that would look something like:So then it becomes clearer that you need to access the function pointers according to the known version and in doing so you gain some safety because Rust can stop you from trying to call a JNI 1.2 method while dereferencing the
.1_1functions.The
.1_1table would also include the padding for the reservednullfunctions in the middle of the table but can rename them so it would be harder to accidentally dereference them.