Re: [Caml-list] Library recompilation with OCamljava

[Kenneth Adam Miller <kennethadammiller@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 6225 bytes --]

Ok, to replicate the bug, just go to
https://github.com/KennethAdamMiller/ocaml-makefile where I'm updating the
fork of ocamlmakefile in order to allow a target to compile for ocamljava.
Once I update (I've committed, but haven't pushed yet, will within a matter
of hours hours), you can clone or pull, and cd to the threads subdirectory,
which is an example project that consumes the ocamlmakefile. Then just do

make jbc  #or make java-byte-code

ex1 and ex2 crashe with uncaught under/overflow exception and some other
overflow errors (not reliable); I think it's because the synchronization
that is expressed in the ocaml code isn't very well followed at java's
level; there seems to be some very great imbalance in thread scheduling
(possibly slow signal handling). In java, there's a long pause where the
output doesn't make it to the screen until the uncaught exception hits,
when it does, the numbers being passed have a huge difference. In contrast,
ocamlopt or ocamlc compiled code executes rather evenly, with each event
receiving scheduler attention pretty much equally. That's my jest.

Agreed! I'd much much rather construct some method of automatic (or even
just expedient manual intervention!) that offers ocaml -> C sublibrary :
java -> C sublibrary support! I've already started down the path of
ocamljava, and right now, even ocaml-RPC with protocol buffers looks like a
long haul as well! ocamljava is just a much better solution.

Precisely! ocaml-ctypes is exactly what's being used by the library that
I'm porting to call into C sub libraries. It would be really sweet if the
ocamljava compiler could detect the ocaml-ctypes and generate these
mappings automatically. This would eliminate a lot of error prone code,
since C code tends to interpret data raw at some point... How might I got
about writing this to boot? I'm moderately new to ocaml, lacking deep
expertise in it, but I'm an aggressive learner. Please explain the best
path forward, I want to create a robust and reusable solution.

On Sat, Oct 18, 2014 at 5:09 AM, forum@x9c.fr <forum@x9c.fr> wrote:

> Hi Kenneth,
>
>
> Le 17 oct. 2014 à 23:33, Kenneth Adam Miller <kennethadammiller@gmail.com>
> a écrit :
>
> So, after doing some more work, I think the answer to 2) is no/no. After
> looking more into piqi, I now just need to find a way to do protocol
> buffers based RPC to an ocaml service...
>
> Anybody know how to do that?
>
> On Thu, Oct 16, 2014 at 4:08 AM, Kenneth Adam Miller <
> kennethadammiller@gmail.com> wrote:
>
>> So, I'm attempting a large library recompile with ocamljava. It's pretty
>> audacious, because some people in my workplace are rather unwilling to
>> learn ocaml, yet a very well respected and needed library is authored in
>> it. Everyone knows java, so we hope very much to recompile the library just
>> with ocamljava.
>>
>> Two important things that make me nervous in pursuing this effort are:
>>
>> 1) once fully recompiled, will the library work in java as it did with
>> native/ocaml byte code?
>>
>> I've already started modifying the popular OCamlMakefile project to add a
>> java-byte-code target, and I've found that a threaded example is admitted
>> by the compiler produces an uncaught exception when run on the JVM... That
>> example may fall out of what the ocamljava docs has been safely implemented
>> under ocamljava, I'm not sure...
>>
>>
> This sounds  like a bug... It would be nice to report it at
> https://github.com/xclerc/ocamljava,
> particularly if you have a (small) reproduction case. The "thread" library
> is indeed only lightly
> tested, mainly because the "concurrent" library (which is specific to
> OCaml-Java) includes
> another implementation of threads, that is much closer to Java threads.
>
>
> 2) I've noticed that the rather large library that I need to compile, as
>> my final target, and aside from the toy targets I'm testing ocamljava with,
>> actually consumes some other C libraries and functions in it's dependency
>> path...
>>
>> This is difficult; the traditional OCamlMakefile builds traditional c
>> stubs to .o files, correctly compiled with ocamlc. But when put on the
>> ocamljava command line, but ocamljava doesn't know what to do about it.
>> Would there be any way that I can have support for this ocaml feature as
>> well, or facilitate some way to link in or enable ocaml code that calls
>> into C being compiled down to java?
>>
>>
> OCaml-Java does not know how to handle C files, as additional primitives
> should be implemented
> through Java files. As a consequence, when porting a project from
> ocamlc/ocamlopt to ocamljava,
> you have to port C files to Java files.
>
>
>
>> 3) What if the answer to 2) is no/no?
>>
>> If I can't use ocamljava, which is the most desired and elegant way,
>> allowing beautiful language inter-operation, how can I *best* facilitate
>> calls to the ocaml library? Is there a fast way to generate callbacks to
>> ocaml in java or any other language? It's very highly preferable not to
>> have to delegate back through the JNI due to type safety and fragility.
>> Alternatively, I looked at the OCaml library Restful, and wondered to
>> myself if there could be any kind of fast definition between ocaml types
>> and an exchange language, like json or something. Ideally, I'd like to be
>> able to generate a url per function in a very very simple declarative
>> manner, so that I can take ocaml libraries, and make them operate as a
>> service, where ocaml library functions correspond to URLs.
>>
>>
> I think a neat way to use a C library from an ocamljava-compiled program
> would be to have a
> Java "backend" for Jeremy Yallop's ctypes (
> https://github.com/ocamllabs/ocaml-ctypes).
> I never had the time to implement that, but toyed with this idea and think
> the best way to
> implement it would be to go through JNA (https://github.com/twall/jna)
> rather than JNI.
> JNA includes a "dlopen"-like mechanism, and automatically maps simple
> types from Java
> to C. My knowledge of ctypes is quite limited, but I see no showstopper.
>
>
> Regards,
>
> Xavier
>

[-- Attachment #2: Type: text/html, Size: 8783 bytes --]