On Thu, 28 Dec 2006, Ivan Zhakov wrote:
> On 12/28/06, Malcolm Rowe <email@example.com> wrote:
> >On Thu, Dec 28, 2006 at 02:24:18PM +0530, Madan U S wrote:
> >> > Do you mean use name "baton" or use baton syntax with "void *"?
> >> Both.
> >> void, because, introducing svn_wc_entry_private_t still leaves space for
> >> ABI revving, when we have svn_wc_entry_private2_t.
> >> the word 'baton' because this would mean wc specific baton containing wc
> >specific information... and the term is pretty well-known in svn-dev-land.
> >Batons are usually provided by the caller, not the library. They're
> >handed back to a callback, hence the name. An existing example of
> >private data would be within the FS layer, within an svn_fs_t, where we
> >just call it 'fsap_data'.
> >From the caller's point of view, I don't believe there's any difference
> >between a 'void *' and a pointer-to-opaque type, since all pointers are
> >the same size (and you'll never be able to dereference either); we'd
> >therefore never need to create a svn_wc_entry_private2_t.
> Oh, sorry. Of course we should use pointer to svn_wc_entry_private_t.
> >That said, we don't need the type publically named for any other reason,
> >so just creating a 'void *private; /* private data for libsvn_wc */'
> >field would make sense to me.
> I prefer use pointer to published structure because we can use it with
> type cast. Just write: entry->private->keep_local = TRUE;
> If we'll use void *, we have to cast it like this:
> ((svn_wc_entry_private*)entry->private)->keep_local = TRUE;
> Which is ugly for me.
I too prefer a pointer to an opaque type, since the pointer will
always point to the same type. Declare the typedef in a public
header, note it's private data, and define it in a private header.
The couple lines the declaration takes doesn't cost us much in terms
of additional complexity for users of our header files, and ought to
leave us with a cleaner internal implementation.
Received on Fri Dec 29 08:53:18 2006
- application/pgp-signature attachment: stored