a little change in _send functions of libnet

[Stefan (metze) Metzmacher]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Rafal,

> I'm thinking about changing _send functions so that they don't
> take TALLOC_CTX argument. This means the composite context would
> be based on libnet_context from the very beginning (right in
> composite_create call). TALLOC_CTX would in such case be used
> in _recv calls, for transferring the libnet function results.

yes, it should be used in the _recv call for transferring the results,
but also in the _send() call passed to composite_create()

> This could simplify the code a bit, and make all memory allocation
> depend on single libnet_context.

I don't see how this will simply the code, if we pass libnet_ctx to
composite_create()

I think the rule should be to pass the callers mem_ctx to
composite_create() then allocate the calls private state under the
composite_context and then all subrequests on the private_state.

then the recv() call transferrs results to the given mem_ctx
and calls talloc_free(composite_context)

then only the results are left on the callers mem_ctx and the rest is
free'ed.

> Thoughts ?

I don't think that's a good idea, because if the composite_context
is a child of a given memory then the request will be removed with a
talloc_free() on the parent context too, with out an explicit
talloc_free() on the request composite_context.

I know we do current what you propose in some other client libs,
but I'm not sure if that's good there too...

metze
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Using GnuPG with SUSE - http://enigmail.mozdev.org

iD8DBQFFhcWBm70gjA5TCD8RApY/AKDSXIe3UNKSbBbkS/RTEob4zo2/DgCgiZmS
UypV9srFXR4g6XOckR/v674=
=5J+P
-----END PGP SIGNATURE-----