From: Charles Mills <cmills@...>
Date: 2005-01-14T04:06:17+09:00
Subject: Re: extending/embedding ruby (callbacks) [LONG]


Peter Schrammel wrote:
> Charles Mills schrieb:
> > Peter Schrammel wrote:
> > It may help if you explain what your trying to do a bit more.
> OK:
> The ruby part just does one thing:
> extend the RFuse Class with some methods:
> getdir
> getattr
>
> Then it creates a RFuse object a calls main on it.
>
> main is in C: It registers the  wrapper functions "rf_getdir",
> "rf_getattr" to fuse and calls the main function of fuse, so the ruby

> programm is sleeping and waiting.
>
> if somebody does a "ls /tmp/fuse" the fuse system calls the
registered
> function rf_getdir with 3 args:
> path, handler, filler
> where
> path is the requested path
> handler is a hanlde used by filler
> filler is a function, that takes 3 args: handler,string,mode
>
> now rf_getdir should do the following:
> create an object of RFiller(set the filler fuction and the handler)
and
> pass the path, and the Rfiller object to self.getdir (on the ruby
side).
>
> self.getdir calls filler.push("filename",mode) and returns
>
> the RFiller object should be destroyed now and rf_getdir is done.
>
> The problem is:
> even if I don't call ruby's getdir method I have endless recursions.
>
> >
> >
> > the function below should be registered using
rb_define_alloc_func()
> > and is typically called rfill_alloc()
>
> done that...thanks
>
>   >
> > you don't need this, if your worried about rfiller_instance being
GC'ed
> > make it volatile.
>
> how to do this? sorry my ruby extension programming is at newbie
level :-)

no worries.  you should read this:
http://www.rubygarden.com/ruby/ruby?action=browse&id=GCAndExtensions

volatile is a C keyword/qualifier.

>
> >
> >
> > not sure how the flow of control goes in your program but seems
pretty
> > obvious you have infinite recursion going on right here.
> >
> ok: But if I do just this:
> static int rf_getdir(const char *path, fuse_dirh_t h, fuse_dirfil_t
f)
> {
>    VALUE rfiller_class;
>    VALUE rfiller_instance;
>    rfiller_class=rb_const_get(rb_cObject,rb_intern("RFiller"));
>    rfiller_instance=rb_funcall(rfiller_class,rb_intern("new"),0);
> 	// no call to anything!
>    return 0;
> }
>
> I still get the same error... :-(
> Commenting out the "rfiller_instance="... line everything is all
right.
>
> To me it seems the GC wakes up one in a while does something ...
> strange. Or I run out of memory because the GC never wakes up.
>

Maybe you should start with something smaller.  Probably something that
doesn't use function pointers.  Perhaps there is a way to test the
classes individually... or you could add some sanity tests.

-Charlie