Re: [Ocamlfuse-devel] Lowlevel interface, anyone have time to write stubs?

[Goswin v. B. <gos...@we...>]

Vincenzo Ciancia <ci...@di...> writes:

> Dear Goswin,
>
> I can't believe I took so much time to reply to you. As usual I am
> very sorry and I hope you did not loose interest in the thing.
>
> Ocamlfuse has a new developer, who was my co-developer in relfs and in
> general is more constant than me on things. The high level api has
> been adjusted for compatibility with FUSE 2.7, even though many
> operations are still missing. The low level api as usual lacks
> manpower, but I promised to try to help you for that. You already know
> that you can join the sf.net project as a developer whenever you want
> (just tell me your sf.net user name when you feel like). Now let's get
> to your old questions. I need more explanations on my side to get back
> in the topic.
>
>> Instead of constructing a bunch of closures on the fly how about a
>> slightly different concept?
>
> Which closures you mean?
>
>> In ocaml you define a user visible
>>
>> type ops = {
>>    init    : unit ->  unit;
>>    destroy : unit ->  unit;
>>    lookup  : int64 ->  string ->  entry;
>>    getattr : int64 ->  (stats * float);
>> }
>>
>> and have one fixed set of private callbacks:
>>
>> fuse_init    : ops ->  unit ->  unit;
>> fuse_destroy : ops ->  unit ->  unit;
>> fuse_lookup  : ops ->  fuse_req ->  int64 ->  string ->  unit;
>> fuse_getattr : ops ->  fuse_req ->  int64 ->  unit;
>> ...
>>
>> let fuse_init ops () = ops.init ()
>> let fuse_getattr ops req inode =
>>      try
>>        let (stats, timeout) = ops.getattr inode
>>        in
>>          assert((fuse_reply_attr req stats timeout) = 0)
>>      with Error err ->  assert((fuse_reply_err req err) = 0)
>>
>>
>> The C code gets the "ops" passed down, tells the GC that it will keep
>> it and passes it to fuse_lowlevel_new(). Then when a callback comes it
>> it retrieves it with
>>
>> /**
>>   * Get the userdata from the request
>>   *
>>   * @param req request handle
>>   * @return the user data passed to fuse_lowlevel_new()
>>   */
>> void *fuse_req_userdata(fuse_req_t req);
>>
>> and passes it back up to the ml calbacks.
>
> I myself am a bit outside of the topic after so much time, but perhaps
> fuse_req_userdata was simply not exposed at the time I wrote the
> bindings (most of these low-level functions were exposed in response
> to the needs of ocamlfuse and of the C# bindings, which suffered from
> the same problems). Could you mock-up the call sequence you
> expect/imagine, from when an event is received from the kernel by
> libfuse, to when it is processed by the user-defined ocaml callback
> and then back to the kernel? I just need a line per call and if it's C
> or CAML land. That will be of great help.
>
> As soon as I understand what you mean we can try to change the current
> high-level api that way, if it works, low-level should be similar.
>
> Vincenzo

I started on my own lowlevel bindings and commited them to the ocaml
forge:

http://forge.ocamlcore.org/projects/libfuse-ocaml/
http://git.ocamlcore.org/libfuse-ocaml/libfuse-ocaml.git/

Maybe that best explains what I ment.

MfG
        Goswin