kicad-developers team mailing list archive

[Lorenzo Marcantonio <l.marcantonio@xxxxxxxxxxxx>]

On Wed, May 01, 2013 at 06:32:46PM +0200, Miguel Angel Ajo wrote:
> this is not a problem when calling pcbnew objects from outside pcbnew,

Unless from the outside you call something that deletes an object you
have 'on hold' no, it's not a problem. Imaginary use case:

Let's say a script want to delete a track segment; it take hold of the
segment (searching around and stuff) and put it in a variable.

Now I don't actually know how track segment are deleted in pcbnew, but,
since it's imaginary, let's say that there is a board function that does
this, given a segment pointer (which is a plausible way to do stuff,
since usually object don't self-destruct). In pseudo-code:

board = GetBoard()
segment = board.FindSegment(stuff_im_interested_to)
... maybe other processing ...
board.DeleteSegment(segment)

At this point, since the segment object was deleted, the segment
reference becomes invalid (usually trying to use it gives a segfault but
anyway is undefined behaviour). It works, just be careful to not use it
after that (or put a null inside to be sure, if python has the concept).
The old SAFE_DELETE macro did something like that for C++ (delete x;
x=0;).

The problem exist in LUA too, BTW... actually since LUA is targeted
toward C, object are implemented as userdata (to anchor in the LUA
memory manager) but you have the same problem! The usual solution is
giving all the objects away to LUA (you gain free garbage collection in
the deal :D) but the object themselves must be coded specially for this
to happen; for example you have to register finalizers *and* every time
you take hold of another object you have to tell LUA that you have it,
otherwise it will be collected. If you ever had the (dis)pleasure of
coding COM object, the drill is the same. I don't know how python works but
I don't think that would be a lot different... it would be great if
there was a way to tell python "this object is no longer alive" and it
would null all the references (just put the magic call in the
destructors...); weak pointers in many languages works that
way (only automatically :D)

> from inside, it could be a problem if we multithreadWx does run as a single thread event loop, or is it multithread?

Never seen an ounce of thread synchronization in the whole kicad :D
>From what I have seen it all run in the main event thread (or whatever
thing in wx receives messages from the outside)

-- 
Lorenzo Marcantonio
Logos Srl