1 2004-11-07 Martin Quinson
2 - Speed up dynar lookup operation a bit.
4 gras_dynar_get is dead.
6 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
7 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
8 (which gives you the address of the stored data).
10 gras_dynar_get_as is an helpful macro which allows you to retrieve a
11 copy of the data using an affectation to do the job and not a memcpy.
13 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
14 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
16 It does not really speedup the dynar test because they are
17 setting elements all the time (and look them seldom). But the dict does
18 far more lookup than setting.
20 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
22 2004-11-04 Martin Quinson
23 - Allow to (en/dis)able the cycle detection at run time.
25 Whether we should check for cycle or not is now a property of each
26 datatype. When you think there may be some cycle, use datadesc_cycle_set.
27 datadesc_cycle_unset allow to remove this property when previously set.
29 Note that the cycle detection is off by default since it impacts the
30 performance. Watch the data you feed GRAS with ;)
32 This property is hereditary. Any element embeeded in a structure having it
33 set have it set for the time of this data exchange.
35 You should set it both on sender and receiver side. If you don't set it on
36 sender side, it will enter an endless loop. If you forget on receiver
37 side, the cycles won't be recreated after communication.
39 - Header reorganization.
40 Kill gras_private.h, each submodule must load the headers it needs.
42 2004-10-04 Martin Quinson
43 - Interface revolution: do not try to survive to malloc failure.
45 Now, gras_malloc and friends call gras_abort() on failure.
46 As a conclusion, malloc_error is not a valid error anymore, and all
47 functions for which it was the only gras_error_t return value are
48 changed. They now return void, or there result directly.
49 This simplify the API a lot.
51 2004-09-29 Martin Quinson
52 - Re-enable raw sockets.
53 Created by gras_socket_{client,server}_ext;
54 Used with gras_raw_{send,recv}
57 It should allow to kill the last bits of gras first version soon.
59 This is not completely satisfactory yet (dupplicate code with
60 chunk_{send,recv}; a bit out of the plugin mecanism), but it should
63 - Simplify transport plugin (internal) interface by not passing any
64 argument to _server and _client, but embeeding them in the socket
67 2004-09-28 Martin Quinson
68 - Finish the port to AIX.
69 autoconf was my problem (segfault within the malloc replacement
70 function. No idea why)
72 2004-09-16 Martin Quinson
73 - Fix some size_t madness on 64bit architectures.
75 2004-09-08 Martin Quinson
76 - Reduce the number of system headers loaded, overload some more system
77 calls (such as malloc to cast the result of the system one, and work
79 - Fix and reintroduce the config support
81 2004-09-07 Martin Quinson
82 - Source code reorganization to allow Arnaud to surf all over there.
83 - Allow to document the logging categories.
84 - Remove all uppercase from logging categories and useless cleanup in names.
86 2004-08-18 Martin Quinson
87 Version 0.6.2 (protocol not changed; API changed)
88 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
89 gras_error_t), and NULL when not found. Functions expecting a msgtype
90 as argument (msg_wait; msg_send) deal with NULL argument by providing a
91 hopefully usefull message.
92 - Portability to prehistoric sparcs again
94 2004-08-17 Martin Quinson
95 Version 0.6.1 (protocol not changed; ABI not changed)
96 - prealloc some buffers to speed things up
98 2004-08-11 Martin Quinson
99 Version 0.6 (protocol not changed; ABI expended)
100 - The parsing macro can deal with the references, provided that you add
101 the relevant annotations (using GRAS_ANNOTE(size,field_name))
103 2004-08-09 Martin Quinson
104 Version 0.5 (protocol not changed; ABI changed)
105 - Allow to off turn the cycle detection code in data exchange at
106 compilation time. It should be at run time, but I'm short of time (and
107 the config stuff is still broken). That way, we keep dict out of the
108 critical path, which is good because the performance is poor:
109 - search not dichotomial yet
110 - dynar give no way to access their content and memcpy everytime
111 - In composed data description (struct, ref and so on), stop foolness of
112 keeping the subtype's ID, but store the type itself. This keeps sets out
113 of the critical path, which is good since they rely on dynar and
114 dictionnaries. The only loose of that is that we cannot detect the
115 redeclaration of a structure/union with another content (but I'm not sure
116 the code detected well this error before anyway). We still can detect
117 the redefinition discrepancy for the other types.
118 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
119 it breaks the code because of type-punning used all over the place).
120 This breaks on all non-gcc architectures (for now).
122 All those changes (plus the buffer of last time) allow me to gain 2 order
123 of magnitude on cruel tests consisting of 800000 array of integers on two
124 level of a hierarchical structure (200 secondes -> 4 secondes)
127 - the selector of reference must now return the type it points to, not
130 2004-08-06 Martin Quinson
131 Version 0.4 (protocol changed; ABI not changed)
132 - Allow to pass --gras-log argument to processes in simulation mode. Really.
133 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
134 - Implement a buffer transport, and use it by default (it relies on tcp in
135 real life and on sg in simulation).
136 That's a bit hackish since I had a new field to the structure to store
137 its data without interfering with the subtype ones. Inheritance
138 is tricky in C. And that's a kind of reverse inheritance with one class
139 derivating two classes. Or maybe a game with java interfaces. Anyway,
140 that's damn hard in C (at least).
141 Moreover, I got tired while trying to ensure plugin separation and
142 genericity in SG mode. MSG wants me to do weird things, so let's go for
143 cruel hacks (temporarily of course ;).
144 See comment in transport_private.h:71
145 - do not include all the _interface headers in private but in the files
146 which really need them (to cut the compilation time when they are
149 2004-07-26 Martin Quinson
150 Version 0.3 (protocol not changed; ABI changed)
151 - Major overhault of the datadesc interface to simplify it:
152 - shorted the function names:
153 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
154 - add a trivial way to push/pop integers into the cbps without malloc.
155 This allows to make really generic sub_type description, which simply
156 pop their size of the stack.
157 - add a function gras_datadesc_ref_pop_arr() which does what users want
158 most of the time: Declare a dynamic array (which pops its size of the
159 stack) and declare a reference to it. Poor name, but anyway.
160 - kill the post-send callback, add a post-receive one
162 2004-07-23 Martin Quinson
163 Version 0.2 (protocol changed; ABI changed)
164 - add some testing for cpbs in the test cases, and fix some more bugs.
165 This invalidate again the little64 data file, since I cannot regenerate
167 - remove an awfull optimization in the logging stuff, allowing me to:
168 - understand it again
169 - learn gcc how to check that the argument match the provided format
170 - fix all errors revealed by gcc after that
171 - internal keys of dict are not \0 terminated. Deal with it properly in
172 loggings instead of segfaulting when the user want to see the logs :-/
174 2004-07-22 Martin Quinson
175 - Fix some stupid bug preventing cbps (callback postit) from working
177 2004-07-21 Martin Quinson
178 - Some documentation cleanups
179 - remove the useless last argument of msgtype_declare
180 - rename the Virtu functions to fit into the 'os' namespace
181 - move headers src/include -> src/include/gras/ and stop fooling with
182 gras -> . symbolic link
183 - make distcheck is now successful
185 2004-07-19 Martin Quinson
187 - Build shared library also
188 - Install html doc to the right location
189 - stop removing maintainer files in make clean
190 - build tests only on make check
192 2004-07-13 Martin Quinson
194 - No major issue in previous version => change versionning schema
195 - Re-enable little64 convertion test now that Abdou kindly regenerated the
196 corresponding dataset.
198 2004-07-11 Martin Quinson
200 - Get it working with any kind of structure (we can compute the padding
201 bytes remotely for all the architectures I have access to)
202 - Implement the structure parsing macro (still not quite robust/complete)
203 - Improvement to the remote testing toysuite
205 2004-07-10 Martin Quinson
207 - get ride of a bunch of deprecated macros
208 - actually run the test for two-compliment, not only compile it :-/
209 - test whether the structures get packed (and bail out if yes. Damn.
210 Alignment is a serious matter)
211 - test whether the structures get compacted (but respecting the alignment
212 constraints of each types)
213 - test whether the array fields of structures can straddle alignment boundaries
215 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
216 (took me 2 hours to find that bug, looking at the wrong place)
217 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
218 before sending/receiving and are used to compute the offsets of fields
219 - Given that padding size depend even on compiler options, keep track of
220 alignment and aligned_size only for the current architecture. Not a big
221 deal since we send structure fields one after the other (seems
223 - Add the datastructure used for IEEE paper by the PBIO guys to the test
224 program, let it work on linux/gcc/little32. portability todo.
226 2004-07-08 Martin Quinson
227 - import and improve remote compilation support from FAST
228 - make sure make check works on half a dozen of machines out there
230 2004-07-07 Martin Quinson
231 Let's say it's version 0.0.3 ;)
232 - Implement conversions (yuhu!)
233 - Let it work on solaris (beside conversion, of course)
234 - Stupid me, using rand() to generate the conversion datatests in not wise.
236 2004-07-06 Martin Quinson
237 - Let make dist work, since I'm gonna need it to compile on remote hosts
238 - Let Tests/datadesc_usage write the architecture on which the file was
239 generated as first byte.
240 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
241 architecture descriptions.
242 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
243 on those architectures.
244 - Optimization: send/recv array of scalar in one shoot
246 2004-07-05 Martin Quinson
247 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
249 - Plug a whole bunch of memleaks
250 - each process now have to call gras_{init,exit}. One day, their log
251 settings will be separated
252 - Continue the code factorisation between SG, RL and common in Transport.
254 2004-07-04 Martin Quinson
256 - Redistribution between SG and RL.
257 We wanna have to accept in SG, so move accepted related parts of RL in
258 the common part. (more precisely, the dynar of all known sockets is no
259 more a static in transport.c, but part of the process_data)
261 [gras_stub_generator]
262 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
264 2004-07-03 Martin Quinson
265 - Create a new log channel tbx containing dict, set, log, dynar (to shut
266 them all up in one shot)
268 - Fix the ugly case of reference to dynamic array.
269 - New (semi-public) function gras_datadesc_size to allow the messaging
270 layer to malloc the needed space for the buffer.
272 - gras_socket_close now expect the socket to close (and not its address to
273 put NULL in it after it). This is because the socket passed to handlers
274 is one of their argument (=> not writable).
276 - propagate the interface cleanup from last week in datadesc, ie remove a
277 superfluous level of indirection. User pass adress of variable
278 containing data (both when sending and receiving), and not of a variable
279 being a pointer to the data. Let's say that I like it better ;)
280 The price for that is constructs like "int msg=*(int*)payload" in
281 handlers, but it's a fine price, IMHO.
283 - Let it work in RL (yuhu)
285 2004-06-21 Martin Quinson
287 - porting SG plugin and SG select to new standards (works almost).
288 - plug memleaks and fix bugs around.
291 - cleanup the prototype of data recv and force users to specify when they
292 want to handle references to objects. Test case working even for cycles.
293 - plug memleaks. Valgrind is perfectly ok with this.
295 2004-06-12 Martin Quinson
297 - cleanup the separation between plugin and main code in plugin creation
299 2004-06-11 Martin Quinson
301 - Reput hook for raw sockets, needed for BW experiments
302 - kill a few lines of dead code
303 [Data description] Interface cleanup
304 - gras_datadesc_by_name returns the searched type or NULL.
305 That way, no variable is needed to use a type desc once, which makes
307 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
308 two parameters were strings (field name, type name), leading to
310 [Dicos] Interface cleanup
311 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
312 This is consistant with the dynar API.
314 2004-04-21 Martin Quinson
316 - Porting to new standards.
319 There is no bag anymore, no need to take extra provision to mask the
320 pointers behind "ID".
321 Better splitup of functions between files create/exchange/convert.
322 This is still a bit artificial since convert and receive are so
323 interleaved, but anyway.
325 - add a queued message list to procdata (the ones not matching criteria
327 - factorize some more code between SG and RL wrt procdata
329 - use gras_exit in example to track memleaks
330 - get rid of gs_example now that GS is properly integrated into gras
331 - update run_test to integrate the lastest tests (datadesc)
333 - rename WARNINGn macros to WARNn since it prooved error-prone
335 2004-04-19 Martin Quinson
337 - register init/exit functions within gras module mecanism
338 - send/receive function.
339 Convertion is not implemented, but short-cutted if not needed.
340 struct/array elements are sent one by one (instead of block-wise), but
341 nobody really cares (yet). Get a prototype before optimizing.
342 - tests (using a file socket) for DD send/receive on:
343 - base types: int, float
344 - array: fixed size, string (ie ref to dynamic string)
345 - structure: homogeneous, heterogeneous
346 - chained list, graph with cycle
347 Believe it or not, valgrind is not too unhappy with the results. The
348 cycle happily segfaults, but the others are ok. And I'm sick of pointers
352 - Bugfix when using a filename explicitely (instead of '-')
354 2004-04-09 Martin Quinson
356 - factorize more code between RL and SG in socket creation
357 - Complete the implementation and tests of:
359 o file (only in RL, and mainly for debugging)
361 I lost 3 days to design a portable address resolver, and then decided
362 that the prototype mainly have to run on my box.
363 Addressing portability too early may be like optimizing too early :-/
365 - use gras_init in the Tests instead of the crappy parse_log_opt
366 (the latter function is removed)
367 [Conditional execution]
368 - New functions: gras_if_RL/gras_if_SG (basic support for this)
369 [Code reorganisation]
370 - Get rid of libgrasutils.a since it makes more trouble than it solves.
371 Build examples against the RL library, since there is no way to disable
372 its creation for now.