default tcp host

[Daniel P. Berrange]

On Wed, Jun 20, 2007 at 12:46:00PM -0400, Havoc Pennington wrote:
> Awesome, some mostly style-type comments inline, I think the basic 
> approach should work fine
> 
> Daniel P. Berrange wrote:
> >+#define MAX_SOCKET 10
> 
> Just sucking it up and dynamically sizing this array probably makes 
> sense - it never resizes, the nfd is passed in at construct-time, so 
> it's just as easy.

Ok, will do.

[snip other coding style issues]

> >+  else if (!strcmp(bind, "*"))
> 
> The convention in libdbus is == 0 rather than !

Will fix all these style issues.

> 
> >+ * @param retfd location to store returned file descriptors
> >+ * @param nretfd maximum number of returnable file descriptors
> 
> Rather than create this limitation / burden on the caller, just do it like:
>  int **fds_p,
>  int  *n_fds_p
> and return an allocated array.

Ok.

> Similarly for the returned port, though in that case you can have the 
> caller pass in an initialized DBusString to be modified, instead of 
> manually returning a char*
> 
> >+	      strncpy(retport, port, retportlen-1);
> >+	      retport[retportlen-1] = '\0';
> 
> DBusString is preferred to any manual string manipulation, basically 
> char* is only used for "immutable-ish" strings (precision guidelines r 
> us!). The codebase avoids most of string.h to minimize that category of 
> security (and crash) bugs. (For read-only use, you can quickly init a 
> DBusString with an existing char* with _dbus_string_init_const())

The reason I used a pre-allocated char * is because that's what the 
getnameinfo() contract requires when resolving the port from a 
sockaddr_storage  struct. So while I could pass in a DBusString to
the method, I'll eventually need to use a pre-allocated char * for
the impl.

> Since listen_tcp_socket / connect_tcp_socket are fairly complex now, it 
> may be worth thinking about how some of the logic can be moved 
> cross-platform so it doesn't have to be synced to Windows, though feel 
> free to punt on that and we'll figure it out later. I don't want to do 
> it if it results in a big mess, and I can't really think of a good 
> non-big-mess way to deal with this. Once there is a Windows 
> implementation it might be easier to see how much of it is common to the 
> platforms.

I was looking at the existing Windows impl of the listen/connect methods
and they in fact seem to have two completely separate implementations of
each method toggled on #ifdef ENABLE_DBUSSOCKET  which AFAICT is never
enabled in the codebase.

Windows has getaddrinfo/getnameinfo, so it ought to be possible to have
an almost 100% unified impl, with just a few inline #ifdef's to deal
with way Windows handles errno with its sockets code. That all said I
have zero Windows machines, so while I can provide some code as a
staring point, I'm unable to compile/debug Windows.

Regards,
Dan.
-- 
|=-            GPG key: http://www.berrange.com/~dan/gpgkey.txt       -=|
|=-       Perl modules: http://search.cpan.org/~danberr/              -=|
|=-           Projects: http://freshmeat.net/~danielpb/               -=|
|=-   berrange at redhat.com  -  Daniel Berrange  -  dan at berrange.com    -=|
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.freedesktop.org/archives/dbus/attachments/20070620/5ee8f084/attachment.pgp