1 2004-08-11 Martin Quinson
2 Version 0.6 (protocol not changed; ABI expended)
3 - The parsing macro can deal with the references, provided that you add
4 the relevant annotations (using GRAS_ANNOTE(size,field_name))
6 2004-08-09 Martin Quinson
7 Version 0.5 (protocol not changed; ABI changed)
8 - Allow to off turn the cycle detection code in data exchange at
9 compilation time. It should be at run time, but I'm short of time (and
10 the config stuff is still broken). That way, we keep dict of of the
11 critical path, which is good because the performance is poor:
12 - search not dichotomial yet
13 - dynar give no way to access their content and memcpy everytime
14 - In composed data description (struct, ref and so on), stop foolness of
15 keeping the subtype's ID, but store the type itself. This keeps sets out
16 of the critical path, which is good since they rely on dynar and
17 dictionnaries. The only loose of that is that we cannot detect the
18 redeclaration of a structure/union with another content (but I'm not sure
19 the code detected this error well before anyway). We still can detect
20 the redefinition discrepancy for the other types.
21 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
22 it breaks the code because of type-punning used all over the place).
23 This breaks on all non-gcc architectures (for now).
25 All those changes (plus the buffer of last time) allow me to gain 2 order
26 of magnitude on cruel tests consisting of 800000 array of integers on two
27 level of a hierarchical structure (200 secondes -> 4 secondes)
30 - the selector of reference must now return the type it points to, not
33 2004-08-06 Martin Quinson
34 Version 0.4 (protocol changed; ABI not changed)
35 - Allow to pass --gras-log argument to processes in simulation mode. Really.
36 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
37 - Implement a buffer transport, and use it by default (it relies on tcp in
38 real life and on sg in simulation).
39 That's a bit hackish since I had a new field to the structure to store
40 its data without interfering with the subtype ones. Inheritance
41 is tricky in C. And that's a kind of reverse inheritance with one class
42 derivating two classes. Or maybe a game with java interfaces. Anyway,
43 that's damn hard in C (at least).
44 Moreover, I got tired while trying to ensure plugin separation and
45 genericity in SG mode. MSG wants me to do weird things, so let's go for
46 cruel hacks (temporarily of course ;).
47 See comment in transport_private.h:71
48 - do not include all the _interface headers in private but in the files
49 which really need them (to cut the compilation time when they are
52 2004-07-26 Martin Quinson
53 Version 0.3 (protocol not changed; ABI changed)
54 - Major overhault of the datadesc interface to simplify it:
55 - shorted the function names:
56 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
57 - add a trivial way to push/pop integers into the cbps without malloc.
58 This allows to make really generic sub_type description, which simply
59 pop their size of the stack.
60 - add a function gras_datadesc_ref_pop_arr() which does what users want
61 most of the time: Declare a dynamic array (which pops its size of the
62 stack) and declare a reference to it. Poor name, but anyway.
63 - kill the post-send callback, add a post-receive one
65 2004-07-23 Martin Quinson
66 Version 0.2 (protocol changed; ABI changed)
67 - add some testing for cpbs in the test cases, and fix some more bugs.
68 This invalidate again the little64 data file, since I cannot regenerate
70 - remove an awfull optimization in the logging stuff, allowing me to:
72 - learn gcc how to check that the argument match the provided format
73 - fix all errors revealed by gcc after that
74 - internal keys of dict are not \0 terminated. Deal with it properly in
75 loggings instead of segfaulting when the user want to see the logs :-/
77 2004-07-22 Martin Quinson
78 - Fix some stupid bug preventing cbps (callback postit) from working
80 2004-07-21 Martin Quinson
81 - Some documentation cleanups
82 - remove the useless last argument of msgtype_declare
83 - rename the Virtu functions to fit into the 'os' namespace
84 - move headers src/include -> src/include/gras/ and stop fooling with
85 gras -> . symbolic link
86 - make distcheck is now successful
88 2004-07-19 Martin Quinson
90 - Build shared library also
91 - Install html doc to the right location
92 - stop removing maintainer files in make clean
93 - build tests only on make check
95 2004-07-13 Martin Quinson
97 - No major issue in previous version => change versionning schema
98 - Re-enable little64 convertion test now that Abdou kindly regenerated the
99 corresponding dataset.
101 2004-07-11 Martin Quinson
103 - Get it working with any kind of structure (we can compute the padding
104 bytes remotely for all the architectures I have access to)
105 - Implement the structure parsing macro (still not quite robust/complete)
106 - Improvement to the remote testing toysuite
108 2004-07-10 Martin Quinson
110 - get ride of a bunch of deprecated macros
111 - actually run the test for two-compliment, not only compile it :-/
112 - test whether the structures get packed (and bail out if yes. Damn.
113 Alignment is a serious matter)
114 - test whether the structures get compacted (but respecting the alignment
115 constraints of each types)
116 - test whether the array fields of structures can straddle alignment boundaries
118 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
119 (took me 2 hours to find that bug, looking at the wrong place)
120 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
121 before sending/receiving and are used to compute the offsets of fields
122 - Given that padding size depend even on compiler options, keep track of
123 alignment and aligned_size only for the current architecture. Not a big
124 deal since we send structure fields one after the other (seems
126 - Add the datastructure used for IEEE paper by the PBIO guys to the test
127 program, let it work on linux/gcc/little32. portability todo.
129 2004-07-08 Martin Quinson
130 - import and improve remote compilation support from FAST
131 - make sure make check works on half a dozen of machines out there
133 2004-07-07 Martin Quinson
134 Let's say it's version 0.0.3 ;)
135 - Implement conversions (yuhu!)
136 - Let it work on solaris (beside conversion, of course)
137 - Stupid me, using rand() to generate the conversion datatests in not wise.
139 2004-07-06 Martin Quinson
140 - Let make dist work, since I'm gonna need it to compile on remote hosts
141 - Let Tests/datadesc_usage write the architecture on which the file was
142 generated as first byte.
143 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
144 architecture descriptions.
145 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
146 on those architectures.
147 - Optimization: send/recv array of scalar in one shoot
149 2004-07-05 Martin Quinson
150 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
152 - Plug a whole bunch of memleaks
153 - each process now have to call gras_{init,exit}. One day, their log
154 settings will be separated
155 - Continue the code factorisation between SG, RL and common in Transport.
157 2004-07-04 Martin Quinson
159 - Redistribution between SG and RL.
160 We wanna have to accept in SG, so move accepted related parts of RL in
161 the common part. (more precisely, the dynar of all known sockets is no
162 more a static in transport.c, but part of the process_data)
164 [gras_stub_generator]
165 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
167 2004-07-03 Martin Quinson
168 - Create a new log channel tbx containing dict, set, log, dynar (to shut
169 them all up in one shot)
171 - Fix the ugly case of reference to dynamic array.
172 - New (semi-public) function gras_datadesc_size to allow the messaging
173 layer to malloc the needed space for the buffer.
175 - gras_socket_close now expect the socket to close (and not its address to
176 put NULL in it after it). This is because the socket passed to handlers
177 is one of their argument (=> not writable).
179 - propagate the interface cleanup from last week in datadesc, ie remove a
180 superfluous level of indirection. User pass adress of variable
181 containing data (both when sending and receiving), and not of a variable
182 being a pointer to the data. Let's say that I like it better ;)
183 The price for that is constructs like "int msg=*(int*)payload" in
184 handlers, but it's a fine price, IMHO.
186 - Let it work in RL (yuhu)
188 2004-06-21 Martin Quinson
190 - porting SG plugin and SG select to new standards (works almost).
191 - plug memleaks and fix bugs around.
194 - cleanup the prototype of data recv and force users to specify when they
195 want to handle references to objects. Test case working even for cycles.
196 - plug memleaks. Valgrind is perfectly ok with this.
198 2004-06-12 Martin Quinson
200 - cleanup the separation between plugin and main code in plugin creation
202 2004-06-11 Martin Quinson
204 - Reput hook for raw sockets, needed for BW experiments
205 - kill a few lines of dead code
206 [Data description] Interface cleanup
207 - gras_datadesc_by_name returns the searched type or NULL.
208 That way, no variable is needed to use a type desc once, which makes
210 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
211 two parameters were strings (field name, type name), leading to
213 [Dicos] Interface cleanup
214 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
215 This is consistant with the dynar API.
217 2004-04-21 Martin Quinson
219 - Porting to new standards.
222 There is no bag anymore, no need to take extra provision to mask the
223 pointers behind "ID".
224 Better splitup of functions between files create/exchange/convert.
225 This is still a bit artificial since convert and receive are so
226 interleaved, but anyway.
228 - add a queued message list to procdata (the ones not matching criteria
230 - factorize some more code between SG and RL wrt procdata
232 - use gras_exit in example to track memleaks
233 - get rid of gs_example now that GS is properly integrated into gras
234 - update run_test to integrate the lastest tests (datadesc)
236 - rename WARNINGn macros to WARNn since it prooved error-prone
238 2004-04-19 Martin Quinson
240 - register init/exit functions within gras module mecanism
241 - send/receive function.
242 Convertion is not implemented, but short-cutted if not needed.
243 struct/array elements are sent one by one (instead of block-wise), but
244 nobody really cares (yet). Get a prototype before optimizing.
245 - tests (using a file socket) for DD send/receive on:
246 - base types: int, float
247 - array: fixed size, string (ie ref to dynamic string)
248 - structure: homogeneous, heterogeneous
249 - chained list, graph with cycle
250 Believe it or not, valgrind is not too unhappy with the results. The
251 cycle happily segfaults, but the others are ok. And I'm sick of pointers
255 - Bugfix when using a filename explicitely (instead of '-')
257 2004-04-09 Martin Quinson
259 - factorize more code between RL and SG in socket creation
260 - Complete the implementation and tests of:
262 o file (only in RL, and mainly for debugging)
264 I lost 3 days to design a portable address resolver, and then decided
265 that the prototype mainly have to run on my box.
266 Addressing portability too early may be like optimizing too early :-/
268 - use gras_init in the Tests instead of the crappy parse_log_opt
269 (the latter function is removed)
270 [Conditional execution]
271 - New functions: gras_if_RL/gras_if_SG (basic support for this)
272 [Code reorganisation]
273 - Get rid of libgrasutils.a since it makes more trouble than it solves.
274 Build examples against the RL library, since there is no way to disable
275 its creation for now.