4 /* FIXME: better place? */
5 int vasprintf (char **ptr, const char *fmt, va_list ap);
6 char *bprintf(const char*fmt, ...) _XBT_GNUC_PRINTF(1,2);
8 Replace fifo with dynars
14 Rename cursor to iterator
25 * Check the gcc version on powerpc. We disabled -floop-optimize on powerpc,
26 but versions above 3.4.0 should be ok.
27 * check whether we have better than jmp_buf to implement exceptions, and
28 use it (may need to generate a public .h, as glib does)
36 (errors, logs ; dynars, dicts, hooks, pools; config, rrdb)
39 * Mallocators and/or memory pool so that we can cleanly kill an actor
42 * Better split casual errors from programing errors.
43 The first ones should be repported to the user, the second should kill
44 the program (or, yet better, only the msg handler)
45 * Allows the use of an error handler depending on the current module (ie,
46 the same philosophy as log4c using GSL's error functions)
49 * Hijack message from a given category to another for a while (to mask
50 initializations, and more)
51 * Allow each actor to have its own setting
52 * a init/exit mecanism for logging appender
53 * Several appenders; fix the setting stuff to change the appender
54 * more logging appenders (take those from Ralf in l2)
57 * speed up the cursors, for example using the contexts when available
60 * better formalisation of what modules are (amok deeply needs it)
61 configuration + init() + exit() + dependencies
62 * allow to load them at runtime
63 check in erlang how they upgrade them without downtime
66 * we may need a round-robin database module, and a statistical one
67 * a hook module *may* help cleaning up some parts. Not sure yet.
68 * Some of the datacontainer modules seem to overlap. Kill some of them?
75 * implement the P2P protocols that macedon does. They constitute great
79 * Spawn threads handling the communication
80 - Data sending cannot be delegated if we want to be kept informed
81 (*easily*) of errors here.
82 - Actor execution flow shouldn't be interrupted
83 - It should be allowed to access (both in read and write access)
84 any data available (ie, referenced) from the actor without
85 requesting to check for a condition before.
86 (in other word, no mutex or assimilated)
87 - I know that enforcing those rules prevent the implementation of
88 really cleaver stuff. Keeping the stuff simple for the users is more
89 important to me than allowing them to do cleaver tricks. Black magic
90 should be done *within* gras to reach a good performance level.
92 - Data receiving can be delegated (and should)
93 The first step here is a "simple" mailbox mecanism, with a fifo of
94 messages protected by semaphore.
95 The rest is rather straightforward too.
97 * use poll(2) instead of select(2) when available. (first need to check
98 the advantage of doing so ;)
100 Another idea we spoke about was to simulate this feature with a bunch of
101 threads blocked in a read(1) on each incomming socket. The latency is
102 reduced by the cost of a syscall, but the more I think about it, the
103 less I find the idea adapted to our context.
105 * timeout the send/recv too (hard to do in RL)
107 * multiplex on incoming SOAP over HTTP (once datadesc can deal with it)
109 * The module syntax/API is too complex.
110 - Everybody opens a server socket (or almost), and nobody open two of
111 them. This should be done automatically without user intervention.
112 - I'd like to offer the possibility to speak to someone, not to speak on
113 a socket. Users shouldn't care about such technical details.
114 - the idea of host_cookie in NWS seem to match my needs, but we still
115 need a proper name ;)
116 - this would allow to exchange a "socket" between peer :)
117 - the creation needs to identify the peer actor within the process
119 * when a send failed because the socket was closed on the other side,
120 try to reopen it seamlessly. Needs exceptions or another way to
121 differentiate between the several system_error.
122 * cache accepted sockets and close the old ones after a while.
123 Depends on the previous item; difficult to achieve with firewalls
126 * Implement gras_datadesc_cpy to speedup things in the simulator
127 (and allow to have several "actors" within the same unix process).
128 For now, we mimick closely the RL even in SG. It was easier to do
129 since the datadesc layer is unchanged, but it is not needed and
131 gras_datadesc_cpy needs to provide the size of the corresponding messages, so
132 that we can report it into the simulator.
133 * Add a XML wire protocol alongside to the binary one (for SOAP/HTTP)
137 * Inter-arch conversions
139 - Convert in the same buffer when size increase
140 - Exchange (on net) structures in one shoot when possible.
141 - Port to really exotic platforms (Cray is not IEEE ;)
142 * datadesc_set_cste: give the value by default when receiving.
143 - It's not transfered anymore, which is good for functions pointer.
145 - Cleanup the code (bison?)
146 - Factorize code in union/struct field adding
147 - Handle typedefs (needs love from DataDesc/)
148 - Handle unions with annotate
150 - Handle long long and long double
151 - Forbid "char", allow "signed char" and "unsigned char", or user code won't be
152 portable to ARM, at least.
153 - Handle struct/union/enum embeeded within another container
154 (needs modifications in DataDesc, too)
158 - Check struct { struct { int a } b; }
160 * gras_datadesc_import_nws?
163 * A proper RPC mecanism
164 - gras_rpctype_declare_v (name,ver, payload_request, payload_answer)
165 (or gras_msgtype_declare_rpc_v).
166 - Attaching a cb works the same way.
167 - gras_msg_rpc(peer, &request, &answer)
168 - On the wire, a byte indicate the message type:
169 - 0: one-way message (what we have for now)
170 - 1: method call (answer expected; sessionID attached)
171 - 2: successful return (usual datatype attached, with sessionID)
172 - 3: error return (payload = exception)
173 - other message types are possible (forwarding request, group
177 * Group communication
178 * Message declarations in a tree manner (such as log channels)?
180 [GRASPE] (platform expender)
181 * Tool to visualize/deploy and manage in RL
182 * pull method of source diffusion in graspe-slave
184 [Actors] (parallelism in GRAS)
185 * An actor is a user process.
186 It has a highly sequential control flow from its birth until its death.
187 The timers won't stop the current execution to branch elsewhere, they
188 will be delayed until the actor is ready to listen. Likewise, no signal
189 delivery. The goal is to KISS for users.
190 * You can fork a new actor, even on remote hosts.
191 * They are implemented as threads in RL, but this is still a distributed
192 memory *model*. If you want to share data with another actor, send it
193 using the message interface to explicit who's responsible of this data.
194 * data exchange between actors placed within the same UNIX process is
195 *implemented* by memcopy, but that's an implementation detail.
197 [Other, more general issues]
198 * watchdog in RL (ie, while (1) { fork; exec the child, wait in father })
199 * Allow [homogeneous] dico to be sent
200 * Make GRAS thread safe by mutexing what needs to be
207 * finish this module (still missing the saturate part)
208 * add a version guessing the appropriate datasizes automatically
210 * provide a way to retrieve the host load as in NWS
211 * log control, management, dynamic token ring
212 * a way using SSH to ask a remote host to open a socket back on me