5 Document the fact that gras processes display the backtrace on sigusr and sigint
8 /* FIXME: better place? */
9 int vasprintf (char **ptr, const char *fmt, va_list ap);
10 char *bprintf(const char*fmt, ...) _XBT_GNUC_PRINTF(1,2);
13 - rename SWAG to RING?
14 - Rename cursor to iterator
16 gras_socket_close should be blocking until all the data sent have been
17 received by the other side (implemented with an ACK mechanism).
28 * Check the gcc version on powerpc. We disabled -floop-optimize on powerpc,
29 but versions above 3.4.0 should be ok.
30 * check whether we have better than jmp_buf to implement exceptions, and
31 use it (may need to generate a public .h, as glib does)
39 (errors, logs ; dynars, dicts, hooks, pools; config, rrdb)
42 * maybe a memory pool so that we can cleanly kill an actor
45 * Better split casual errors from programming errors.
46 The first ones should be reported to the user, the second should kill
47 the program (or, yet better, only the msg handler)
48 * Allows the use of an error handler depending on the current module (ie,
49 the same philosophy as log4c using GSL's error functions)
52 * Hijack message from a given category to another for a while (to mask
53 initializations, and more)
54 * Allow each actor to have its own setting
55 * a init/exit mecanism for logging appender
56 * more logging appenders (take those from Ralf in l2)
59 * Add configuration and dependencies to our module definition
60 * allow to load them at runtime
61 check in erlang how they upgrade them without downtime
64 * we may need a round-robin database module, and a statistical one
65 * a hook module *may* help cleaning up some parts. Not sure yet.
66 * Some of the datacontainer modules seem to overlap. Kill some of them?
67 - replace fifo with dynars
68 - replace set with SWAG
75 * implement the P2P protocols that macedon does. They constitute great
79 * use poll(2) instead of select(2) when available. (first need to check
80 the advantage of doing so ;)
82 Another idea we spoke about was to simulate this feature with a bunch of
83 threads blocked in a read(1) on each incoming socket. The latency is
84 reduced by the cost of a syscall, but the more I think about it, the
85 less I find the idea adapted to our context.
87 * timeout the send/recv too (hard to do in RL)
89 * multiplex on incoming SOAP over HTTP (once datadesc can deal with it)
91 * The module syntax/API is too complex.
92 - Everybody opens a server socket (or almost), and nobody open two of
93 them. This should be done automatically without user intervention.
94 - I'd like to offer the possibility to speak to someone, not to speak on
95 a socket. Users shouldn't care about such technical details.
96 - the idea of host_cookie in NWS seem to match my needs, but we still
98 - this would allow to exchange a "socket" between peer :)
99 - the creation needs to identify the peer actor within the process
101 * when a send failed because the socket was closed on the other side,
102 try to reopen it seamlessly. Needs exceptions or another way to
103 differentiate between the several system_error.
104 * cache accepted sockets and close the old ones after a while.
105 Depends on the previous item; difficult to achieve with firewalls
108 * Add a XML wire protocol alongside to the binary one (for SOAP/HTTP)
112 * Inter-arch conversions
114 - Convert in the same buffer when size increase
115 - Exchange (on net) structures in one shoot when possible.
116 - Port to really exotic platforms (Cray is not IEEE ;)
117 * datadesc_set_cste: give the value by default when receiving.
118 - It's not transfered anymore, which is good for functions pointer.
120 - Cleanup the code (bison?)
121 - Factorize code in union/struct field adding
122 - Handle typedefs (gras_datatype_copy can be usefull, but only if
123 main type is already defined)
124 - Handle unions with annotate
126 - Handle long long and long double
127 - Forbid "char", allow "signed char" and "unsigned char", or user code won't be
128 portable to ARM, at least.
129 - Handle struct/union/enum embeeded within another container
130 (needs modifications in DataDesc, too)
134 - Check struct { struct { int a } b; }
136 * gras_datadesc_import_nws?
139 * Other message types than oneway & RPC are possible:
140 - forwarding request, group communication
143 * Group communication
144 * Message declarations in a tree manner (such as log channels)?
146 [GRASPE] (platform expender)
147 * Tool to visualize/deploy and manage in RL
148 * pull method of source diffusion in graspe-slave
150 [Actors] (parallelism in GRAS)
151 * An actor is a user process.
152 It has a highly sequential control flow from its birth until its death.
153 The timers won't stop the current execution to branch elsewhere, they
154 will be delayed until the actor is ready to listen. Likewise, no signal
155 delivery. The goal is to KISS for users.
156 * You can fork a new actor, even on remote hosts.
157 * They are implemented as threads in RL, but this is still a distributed
158 memory *model*. If you want to share data with another actor, send it
159 using the message interface to explicit who's responsible of this data.
160 * data exchange between actors placed within the same UNIX process is
161 *implemented* by memcopy, but that's an implementation detail.
163 [Other, more general issues]
164 * watchdog in RL (ie, while (1) { fork; exec the child, wait in father })
165 * Allow [homogeneous] dico to be sent
166 * Make GRAS thread safe by mutexing what needs to be
173 * add a version guessing the appropriate datasizes automatically
175 * log control, management, dynamic token ring
176 * a way using SSH to ask a remote host to open a socket back on me
184 * select portion of the system that changed instead of recomputing
185 * the whole system solution at each action change