Patch generally looks good (just glanced, don't have time for a full review now) - though.. the gio/gioenumtypes.c hunk is unnecessary because that file is generated. I'm curious - does this patch make GDBus work on FreeBSD? (related: does the test suite pass?) Thanks, David

Yes, this patch is absolutely required to make GDBus work on FreeBSD.  With it, GNOME 2.31 starts and seems to work.  I haven't run any test suites yet, but I think they should work.  The only thing I couldn't implement is the generic interface to get socket credentials.  That requires a two-way street on FreeBSD.  One end writes the credential message, and the other end reads it.

(In reply to comment #2)
> Yes, this patch is absolutely required to make GDBus work on FreeBSD.  With it,
> GNOME 2.31 starts and seems to work.

Cool.

>  I haven't run any test suites yet, but I
> think they should work.  The only thing I couldn't implement is the generic
> interface to get socket credentials. 

g_socket_get_credentials(), right? That's fine, we don't guarantee that it actually works; it's fine to return G_IO_ERROR_NOT_SUPPORTED.

> That requires a two-way street on
> FreeBSD.  One end writes the credential message, and the other end reads it.

Right, that's g_unix_connection_{send,receive}_credentials() which must always work on POSIX (it's implemented one way or another; e.g. either by passing creds or by using things like g_socket_get_credentials()).


I'll commit the patch tomorrow. Thanks for making sure it works on FreeBSD!

Committed with a minor change (a printed warning references 'struct ucred' when you meant 'struct cmsgcred' instead):

http://git.gnome.org/browse/glib/commit/?id=964eb62343b53cf9172d409adacbb58d78896092

Thanks.

Hello, I have a small problem about this part:
@@ -140,6 +153,33 @@ g_unix_credentials_message_deserialize (
  out:
     ;
   }
+#elif defined(__FreeBSD__)
+  {
+    GCredentials *credentials;
+    struct cmsgcred *cred;
+
+    if (level != SOL_SOCKET || type != SCM_CREDS)
+      {
+        goto out;
+      }
+    if (size < CMSG_LEN (sizeof *cred))
+      {
+        g_warning ("Expected a struct ucred (%" G_GSIZE_FORMAT " bytes) but "
+                   "got %" G_GSIZE_FORMAT " bytes of data",
+                   CMSG_LEN (sizeof *cred),
+                   size);
+        goto out;
+      }
+
+    cred = data;
+
+    credentials = g_credentials_new ();
+    g_credentials_set_native (credentials, G_CREDENTIALS_TYPE_FREEBSD_CMSGCRED, cred);
+    message = g_unix_credentials_message_new_with_credentials (credentials);
+    g_object_unref (credentials);
+ out:
+    ;
+  }
 #endif

Why you need CMSG_LEN()? This adds extra sizeof(struct cmsghdr) bytes to the size to check. However, the corresponding Linux code does not have this one.

Also considering g_unix_credentials_message_serialize(), there is no CMSG_LEN() there. Why do you need it in g_unix_credentials_message_deserialize()?

The problem is that I am testing new ibus (1.3.99.20110419.tar.gz) under FreeBSD 8.2-STABLE. I see warning messages:

(ibus-daemon:97531): GLib-GIO-WARNING **: Expected a struct cmsgcred (96 bytes) but got 84 bytes of data

(ibus-daemon:97531): GLib-GIO-WARNING **: unknown control message type 65535:3

They repeated many times. sizeof(struct cmsgcred) is 84, and CMSG_LEN(sizeof(struct cmsgcred)) is 96. And ibus does not start correctly.
After I removed the CMSG_LEN()s in g_unix_credentials_message_deserialize(), it works.

So I suspect that this is a problem in this commit?

Please reopen a bug when you add a comment; otherwise it is unlikely to be noticed.

Yeah, the CMSG_LEN() definitely seems wrong.

(In reply to comment #2)
> I haven't run any test suites yet, but I
> think they should work.

Famous last words... Henry (or Joe), can you try "make check" in gio/tests and see if any more changes are needed?

**
ERROR:gsettings.c:646:test_l10n: assertion failed (str == "Unbenannt"): ("Unnamed" == "Unbenannt")
  /gsettings/no-path:                                                  OK
  /gsettings/basic-types:                                              OK
  /gsettings/complex-types:                                            OK
  /gsettings/changes:                                                  OK
  /gsettings/l10n:                                                     FAIL
GTester: last random seed: R02S3e750f50f305c01eb71f4ab7d1f1b72c


It fails in gsettings...

(In reply to comment #8)
> **
> ERROR:gsettings.c:646:test_l10n: assertion failed (str == "Unbenannt"):
> ("Unnamed" == "Unbenannt")

hm... that passes for me. Anyway, I'm committing fixes for this and a few other problems