First stab at generating JNI callbacks #1
Conversation
This doesn't seem like a change to commit into master. It's more of a starting point for a conversation about the pieces that are still missing.
| . # TODO: verify that these don't duplicate | ||
| $(type)_cb_wrapper(\ | ||
| . # This method signature needs to be built on the model's knowledge of | ||
| . # the function callback type |
There was a problem hiding this comment.
This is really a recurring theme here.
| . elsif type = "zloop_timer_fn" | ||
| zloop_t *loop, int timer_id, void *arg\ | ||
| . else | ||
| garbage *unknown, void *arg\ |
There was a problem hiding this comment.
Need a better way to shortcut on "unknown callback type."
But the root cause is that I don't know how to find out what we know about this callback
| . else | ||
| Int\ | ||
| . endif | ||
| Method(env, wrapper->obj, wrapper->mid, wrapper->args); |
There was a problem hiding this comment.
There's a different, but related, issue here.
We need to invoke the callback with the relevant callback args that we got here in addition to wrapper->args.
It's easy enough to do with the same sort of if/elsif construct I used above, but that approach already feels ridiculous.
There was a problem hiding this comment.
In order to make this happen with the current approach, we really need multiple ICallback variations. Each with an appropriate method signature.
| // Shortcut to throw a NoSuchMethodError | ||
| return; | ||
| } | ||
| $(type) $(c_name)_$(type) = $(type)_cb_wrapper; |
There was a problem hiding this comment.
This
If nothing else, it's duplicated in too many places for comfort and should be part of the container's data instead.
| @@ -899,6 +948,14 @@ $(project.GENERATED_WARNING_HEADER:) | |||
| #include "$(project.header:)" | |||
| #include "$(my.cname:).h" | |||
|
|
|||
| // TODO: This doesn't really belong in here | |||
There was a problem hiding this comment.
I'm just fuzzy about a) where would be better and b) the best way to make that happen
| @@ -169,6 +195,7 @@ function resolve_method (method) | |||
|
|
|||
| if variadic = 1 | |||
| if va_start <> "format" | |||
| # I think this is where the Zloop constructor breaks | |||
There was a problem hiding this comment.
Different issue. Doesn't belong here.
| @@ -197,6 +224,8 @@ function resolve_method (method) | |||
| jni_native_invocation_c += "$(comma)($(argument.c_type)) (intptr_t) &$(argument.c_name)" | |||
| elsif jni_is_class = 1 | type = "anything" | type = "sockish" | |||
| jni_native_invocation_c += "$(comma)($(argument.c_type)) (intptr_t) $(argument.c_name)" | |||
| elsif argument.callback ?= 1 | |||
| jni_native_invocation_c += "$(comma) $(argument.c_name)_$(argument.type)" | |||
There was a problem hiding this comment.
Big gaping hole: need to replace the "args" parameter with a pointer to a jni_cb_struct.
Which means we need to malloc that. And make sure it gets freed.
This doesn't seem like a change to commit into master. It's more of a
starting point for a conversation about the pieces that are still
missing.