Re: [Plplot-devel] The libltdl method of dynamically loading our drivers is now working (was Quick help needed with changing a struct and a typedef in PLplot)

[Maurice L. <mj...@ga...>]

Alan W. Irwin writes:
 > (2) should be changed to
 > typedef lt_ptr (*PLDispatchInit)( PLDispatchTable *pdt );
 > 
 > Any C expert on the list know why *PLDispatchInit rather than
 > PLDispatchInit? I naively thought you would drop the asterisk because the
 > type was going from void * to lt_ptr, but as you all know, my C programming
 > skills are not the best especially when dealing with pointers and
 > complicated typedefs. The example of typedef on p. 147 of the little white
 > K&R book is analogous to this one, but I don't understand the point of
 > (*PFI) rather than (PFI) in that case as well.

It looks to me you're going from a void (not void*) to lt_ptr, since the
original typedef was:

    typedef void (*PLDispatchInit)( PLDispatchTable *pdt );

The new typedef:

    typedef lt_ptr (*func)( PLDispatchTable *pdt );

allows you to declare a function pointer in the following way:

    func dispatch_init;

which means that:

    dispatch_init is a pointer to a function that passes a pointer to a
PLDispatchTable and returns lt_ptr.  

It doesn't matter what "func" is called, in this case PLDispatchInit.

One confusing point is that ANSI C allows you to drop the explicit dereference
when making the actual function call, i.e.

    (*dispatch_init)(pdt);
and
    dispatch_init(pdt);

are the same.  But since you have to keep the distinction straight in the
declaration, the first form is probably better (IMO) as being more
transparent.

 > One other issue I ran into was how to initialize libltdl and release its
 > memory when done.  Same old problem (mentioned in the plcore.c comments) of
 > no clear PLplot library initialization . So I chose to do the initialize
 > whenever a stream was created and release whenever it was destroyed.  This
 > is overcomplicated, and it would be better to initialize just once before
 > any streams are created and release after they are all destroyed, but there
 > is no clear way to tell when plinit is first called so I had to do it stream
 > by stream.

Could it go in pllib_init()?  I added this earlier this year to help solve
this problem -- any routine the user might call first (if I didn't miss any)
calls this first to do library initialization.

 > The AT effort is going much faster than I thought (because of the great
 > start by Rafael and because autotools is pretty easy to learn from scratch)
 > so the merge to HEAD *could* happen soon (i.e., maybe even sometime during
 > this week unless I run into unforseen problems.) So I suggest you should get
 > involved immediately if you want to evaluate the AT approach before the
 > merge!

Will try, but am very very busy.  I appreciate all your effort on this tho.

 > Alan

-- 
Maurice LeBrun    mj...@ga...
Research Organization for Information Science and Technology of Japan (RIST)