JNI using tools or by "hand"?

Question

I am going to be honest and upfront here, I am posing this question because I am scheduled to give a presentation to a lot of people including my Boss and his Boss and even her Boss in the begining of October. I have tried to gather some information on the web about JNI tools and have found that thare are more than a few ways to do JNI but no one really says how much they liked/disliked each method. Since JNI is just a part of my presentation (I figure about 45 minutes worth) I don't have time to try each one of the many options out. In return for any information, if i get enough information to make it worthwhile, I will post the compiled list including credit for information provided to the java community websites.
What I need to know is what you have tried, how it worked for you, and pro's you found, any con's, and if you would recomend it to another.
For example I have tried swig a little.
Tool: Swig
How it worked for me: It got the job done after a lot of manipulation of the C code.
Pros: It was fast and relativly easy on the java side.
Cons: I had to write tons of wrappers to the C code because there was tons of crap swig could not take, arrays were a big hastle with swig. The java generated code was also really hard to read and seems horribly inefficient.
Would I recomend it: NO!
More background info if anyone is interested: I was tasked with this because I suggested we stopped using so many questionable tools to autogenerate code for us and JNI was one of the areas I pointed out we could benefit by doing it by hand. Though I have never done it by hand myself. Since a huge portion of our work is maintaining the code, if we learn how to write JNI by hand and do so not only will the JNI be more readable and efficient, but we will actually be able to maintain it without adding wrappers to C further slowing things down and increasing overhead. Am I wrong?

Answer 1

JustSomeGuy wrote:
Have you tried swig before? We could not modify it to get it to play nice with some types, especially any array that is not a standard single dimension array. We ended up having to use C wrappers.That by itself would not be a reason to dismiss all code generation.
jschell wrote:
Nothing you said here would lead to that conclusion for me.
In terms of "messy", because it seemed awkward or because it required a lot of manual steps?
If the second then why not simply automate that process.
If the first that isn't a reason. Generated code doesn't need to look nice. Basically, we write a GUI that runs about 50 different functions in C and a few fortran ones and displays their results as a model. Those functions each have a team of programmers that is constantly changing them, often these changes include changes that affect JNI code, but usually only one small portion of it. Presumably each function actually consists of a large number of methods, rather than for example, just a couple where some new data types are added.
Presumably a meta data solution would not work either - where a C layer provides a generic structure (name/value pairs are the easiest example) and then changes are just added generically.
For changes that occur frequently one should probably consider a dynamic interface anyways unless one can demonstrate that such a design is detrimental.
The generated code makes it impossible to read and edit by hand because it is really obfuscated, so we have been regenerating the code every time, which is not always a quick/easy process. By obfuscated you mean hard to read correct? Rather than as in the java term where class files are reworked to make reverse engineering harder.
Why is is not easy?
Being able to simply change easy to read JNI code by hand would be much faster for us in almost every situation. That is of course unless you know of a better tool than swig, which would be why I created this thread.
The explanation is not clear. A code generation tool is a run and forget type of operation. It might take a while to actually run but that is a different problem. If that is the problem is it because the make process is monolithic in that everything is regenerated?
So, since you seem to be knowledgeable, what tools have you tried and how did they work out for you?
I have been creating my own code generators for more than 10 years. Mostly build time but sometimes run time as well.
Noting of course that creating your own tool could be a possibility.