1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
|
Important for 1.0
===
- How we will handle DCOP needs sorting out. Among other things, we
need to check that service and service-ownership semantics map to DCOP
reasonably well.
- Activation needs some careful additional thinking-through.
- Audit @todo and FIXME for security issues
- The convenience functions in dbus-bus.h should perhaps have
the signatures that they would have if they were autogenerated
stubs. e.g. the acquire service function. We should also evaluate
which of these functions to include, in light of the fact that
GLib/Qt native stubs will probably also exist.
- the "break loader" and valid/invalid message tests are all disabled;
they need to be fixed and re-enabled with the new message args stuff.
I think I want to drop the .message files thing and just have code
that generates messages, more like the tests for
dbus-marshal-recursive.c
- modify the auth protocol to also support other initial-handshake
type of information:
Perhaps the auth protocol should be able to negotiate a protocol
version to the least-common-denominator between client and server?
Though in practice ever using this feature would be pretty tough,
since protocol probably modifies the API. But we could have it there
as a safety net.
- need to define bus behavior if you send a message to
yourself; is it an error, or allowed? If allowed,
we need to have a test for it in the test suite.
- array lengths should probably be returned as size_t rather than int
(though they are kind of a pita to pass in as size_t with the
varargs, so maybe not - what does glib do with g_object_get()?)
- rename the service thing. unique service names (":1") and well-known
("org.foo.bar") should have different names probably; something like
"address" for the unique and "alias" for the well-known, or
"application id" for the unique and "common name" or "published
name" for the well-known; not sure yet.
- things are broken now when mixing endianness, because DBusMessage
doesn't autoswap itself when you access a message of alternate
endian.
- add string array support back to append_args()
Important for 1.0 GLib Bindings
===
- finish dbus-glib-tool support for adding introspection
data to GObject and autoexporting GObject using same
- the GLib bindings varargs take DBUS_TYPE_WHATEVER and
return stuff allocated with dbus_malloc(); should this
be made more "G" at some expense in code duplication?
You also still have to use some D-BUS functions such as
dbus_message_get_args() which takes a DBusError.
Probably we need to either fully encapsulate and hide
dbus/dbus.h, or encapsulate it slightly less e.g. no
GError. Or maybe it's as simple as "never return dbus_malloc()
memory" and just fully encapsulate the get_args() type of
stuff.
- dbus_gproxy_connect_signal() has to take a signature for the signal
so it can figure out how to invoke the callback, or we have to rely
on having introspection data.
Might as Well for 1.0
===
- Probably no point in a version number in the daemon name
(s/dbus-daemon-1/dbus-daemon/)
- add dbus_message_has_path(), maybe has_member/interface
- dbus_message_iter_init_array_iterator has "iter" and "iterator"
in the same function name
- connection_open/connection_disconnect lacks symmetry, open/close
or connect/disconnect
- protocol version in each message is pretty silly
Can Be Post 1.0
===
- Property list feature on message bus (list of properties associated
with a connection). May also include message matching rules
that involve the properties of the source or destination
connection.
- Disconnecting the remote end on invalid UTF-8 is probably not a good
idea. The definition of "valid" is slightly fuzzy. I think it might
be better to just silently "fix" the UTF-8, or perhaps return an error.
Owen says we should only validate the UTF-8 on dbus_message_get_string()
(changing get_string to have an error return, and allowing a type error
as a possible return)
- assorted _-prefixed symbols in libdbus aren't actually used by
libdbus, only by the message bus. These bloat up the library
size. Not sure how to fix, really.
- build and install the Doxygen manual in Makefile when --enable-docs
- if you send the same message to multiple connections, the serial number
will only be right for one of them. Probably need to just write() the serial
number, rather than putting it in the DBusMessage, or something.
- perhaps the bus driver should have properties that reflect attributes
of the session, such as hostname, architecture, operating system,
etc. Could be useful for code that wants to special-case behavior
for a particular host or class of hosts, for example.
- currently the security policy stuff for messages to/from
the bus driver is kind of strange; basically it's hardcoded that
you can always talk to the driver, but the default config file
has rules for it anyway, or something. it's conceptually
screwy at the moment.
- when making a method call, if the call serial were globally unique,
we could forward the call serial along with any method calls made
as a result of the first method call, and allow reentrancy that was
strictly part of the call stack of said method call. But I don't
really see how to do this without making the user pass around the
call serial to all method calls all the time, or disallowing
async calls.
If done post 1.0 will probably be an optional/ugly-API type
of thing.
- I don't want to introduce DBusObject, but refcounting and object
data could still be factored out into an internal "base class"
perhaps.
- document the auth protocol as a set of states and transitions, and
then reimplement it in those terms
- recursive dispatch, see dbus_connection_dispatch()
- do we need per-display activation; if so I'd like to do this by setting a
"display ID" property on screen 0, with a GUID, and keying activation by
said GUID. Otherwise you get all kinds of unrobust
string/hostname-based mess. per-screen is then done by appending screen number
to the display. If displays have a deterministic ID like this, you can
do per-display by simply including GUID in the service name.
- optimization and profiling!
|
