[foreign-jextract] RFR 8236434: Initial jextract API implementation

[Jorn Vernee]

Hi,

Some files seem to be missing from the webrev. I only see Declaration in 
the exported package, and when building it complains about missing types 
(e.g. Position, Type).

Jorn

On 20/12/2019 16:25, Maurizio Cimadamore wrote:
> Hi,
> here's the initial webrev for the jextract API:
>
> http://cr.openjdk.java.net/~mcimadamore/panama/8236434/
>
> And here's a link to the javadoc for said API:
>
> http://cr.openjdk.java.net/~mcimadamore/panama/jextract_javadoc
>
> Summary:
>
> * most of the configuration changes come from the "foreign" branch - 
> that is, when building the new "foreign-jextract" branch you will need 
> to specify an LLVM install (e.g. --with-libclang); I've removed few 
> things that seemed to be unused, but most of the stuff is really the 
> same - in principle I could have done w/o - but then it's really hard 
> to use and test the API.
>
> * In the implementation code, you will find a Java port of libclang - 
> this port has been fully written using the stuff in the foreign-abi 
> branch (hence, method handles and var handles). The lowest layer of 
> this API is contained in the 'libclang' folder and has been 
> autogenerated for the most part. In other words, there's not a single 
> line of JNI code in here.
>
> * It's interesting to note how - despite the many native calls in used 
> in this port, only _one_ required breaking into privileged mode, and 
> into ForeignUnsafe. I think that's pretty good!
>
> * I've paid extra attention when using layout constants - more 
> specifically, since libclang already tries to be portable and to 
> define things so that they have same sizes on all platforms (e.g. a 
> long in Linux becomes a 'long long' in Windows), I've defined a bunch 
> of size-specific ABI layout constants in ClangLayouts, and then used 
> these everywhere in the library port. In the library implementation 
> there are few places where we need to use real ABI-dependent layouts 
> (when creating a Type object from a clang type), so you'll see some 
> uses of another set of constants defined inside LayoutUtils. 
> Hopefully, all this gymnastic should result in something that works 
> across all ABIs.
>
> * I've added a very simple smoke test - I know we need to do more 
> here, but I think it's better to start somewhere. Besides, this test 
> should provide a useful basis to, eventually, build a more general 
> framework to validate the result of the jextract API parsing.
>
> * There's no minimal jextract as of yet - since this changeset is 
> already very big, I think it's better to first push the API, and then 
> review the extractor code separately.
>
> Cheers
> Maurizio
>