1 2004-08-09 Martin Quinson
2 Version 0.5 (protocol not changed; ABI changed)
3 - Allow to off turn the cycle detection code in data exchange at
4 compilation time. It should be at run time, but I'm short of time (and
5 the config stuff is still broken). That way, we keep dict of of the
6 critical path, which is good because the performance is poor:
7 - search not dichotomial yet
8 - dynar give no way to access their content and memcpy everytime
9 - In composed data description (struct, ref and so on), stop foolness of
10 keeping the subtype's ID, but store the type itself. This keeps sets out
11 of the critical path, which is good since they rely on dynar and
12 dictionnaries. The only loose of that is that we cannot detect the
13 redeclaration of a structure/union with another content (but I'm not sure
14 the code detected this error well before anyway). We still can detect
15 the redefinition discrepancy for the other types.
16 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
17 it breaks the code because of type-punning used all over the place).
18 This breaks on all non-gcc architectures (for now).
20 All those changes (plus the buffer of last time) allow me to gain 2 order
21 of magnitude on cruel tests consisting of 800000 array of integers on two
22 level of a hierarchical structure (200 secondes -> 4 secondes)
25 - the selector of reference must now return the type it points to, not
28 2004-08-06 Martin Quinson
29 Version 0.4 (protocol changed; ABI not changed)
30 - Allow to pass --gras-log argument to processes in simulation mode. Really.
31 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
32 - Implement a buffer transport, and use it by default (it relies on tcp in
33 real life and on sg in simulation).
34 That's a bit hackish since I had a new field to the structure to store
35 its data without interfering with the subtype ones. Inheritance
36 is tricky in C. And that's a kind of reverse inheritance with one class
37 derivating two classes. Or maybe a game with java interfaces. Anyway,
38 that's damn hard in C (at least).
39 Moreover, I got tired while trying to ensure plugin separation and
40 genericity in SG mode. MSG wants me to do weird things, so let's go for
41 cruel hacks (temporarily of course ;).
42 See comment in transport_private.h:71
43 - do not include all the _interface headers in private but in the files
44 which really need them (to cut the compilation time when they are
47 2004-07-26 Martin Quinson
48 Version 0.3 (protocol not changed; ABI changed)
49 - Major overhault of the datadesc interface to simplify it:
50 - shorted the function names:
51 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
52 - add a trivial way to push/pop integers into the cbps without malloc.
53 This allows to make really generic sub_type description, which simply
54 pop their size of the stack.
55 - add a function gras_datadesc_ref_pop_arr() which does what users want
56 most of the time: Declare a dynamic array (which pops its size of the
57 stack) and declare a reference to it. Poor name, but anyway.
58 - kill the post-send callback, add a post-receive one
60 2004-07-23 Martin Quinson
61 Version 0.2 (protocol changed; ABI changed)
62 - add some testing for cpbs in the test cases, and fix some more bugs.
63 This invalidate again the little64 data file, since I cannot regenerate
65 - remove an awfull optimization in the logging stuff, allowing me to:
67 - learn gcc how to check that the argument match the provided format
68 - fix all errors revealed by gcc after that
69 - internal keys of dict are not \0 terminated. Deal with it properly in
70 loggings instead of segfaulting when the user want to see the logs :-/
72 2004-07-22 Martin Quinson
73 - Fix some stupid bug preventing cbps (callback postit) from working
75 2004-07-21 Martin Quinson
76 - Some documentation cleanups
77 - remove the useless last argument of msgtype_declare
78 - rename the Virtu functions to fit into the 'os' namespace
79 - move headers src/include -> src/include/gras/ and stop fooling with
80 gras -> . symbolic link
81 - make distcheck is now successful
83 2004-07-19 Martin Quinson
85 - Build shared library also
86 - Install html doc to the right location
87 - stop removing maintainer files in make clean
88 - build tests only on make check
90 2004-07-13 Martin Quinson
92 - No major issue in previous version => change versionning schema
93 - Re-enable little64 convertion test now that Abdou kindly regenerated the
94 corresponding dataset.
96 2004-07-11 Martin Quinson
98 - Get it working with any kind of structure (we can compute the padding
99 bytes remotely for all the architectures I have access to)
100 - Implement the structure parsing macro (still not quite robust/complete)
101 - Improvement to the remote testing toysuite
103 2004-07-10 Martin Quinson
105 - get ride of a bunch of deprecated macros
106 - actually run the test for two-compliment, not only compile it :-/
107 - test whether the structures get packed (and bail out if yes. Damn.
108 Alignment is a serious matter)
109 - test whether the structures get compacted (but respecting the alignment
110 constraints of each types)
111 - test whether the array fields of structures can straddle alignment boundaries
113 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
114 (took me 2 hours to find that bug, looking at the wrong place)
115 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
116 before sending/receiving and are used to compute the offsets of fields
117 - Given that padding size depend even on compiler options, keep track of
118 alignment and aligned_size only for the current architecture. Not a big
119 deal since we send structure fields one after the other (seems
121 - Add the datastructure used for IEEE paper by the PBIO guys to the test
122 program, let it work on linux/gcc/little32. portability todo.
124 2004-07-08 Martin Quinson
125 - import and improve remote compilation support from FAST
126 - make sure make check works on half a dozen of machines out there
128 2004-07-07 Martin Quinson
129 Let's say it's version 0.0.3 ;)
130 - Implement conversions (yuhu!)
131 - Let it work on solaris (beside conversion, of course)
132 - Stupid me, using rand() to generate the conversion datatests in not wise.
134 2004-07-06 Martin Quinson
135 - Let make dist work, since I'm gonna need it to compile on remote hosts
136 - Let Tests/datadesc_usage write the architecture on which the file was
137 generated as first byte.
138 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
139 architecture descriptions.
140 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
141 on those architectures.
142 - Optimization: send/recv array of scalar in one shoot
144 2004-07-05 Martin Quinson
145 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
147 - Plug a whole bunch of memleaks
148 - each process now have to call gras_{init,exit}. One day, their log
149 settings will be separated
150 - Continue the code factorisation between SG, RL and common in Transport.
152 2004-07-04 Martin Quinson
154 - Redistribution between SG and RL.
155 We wanna have to accept in SG, so move accepted related parts of RL in
156 the common part. (more precisely, the dynar of all known sockets is no
157 more a static in transport.c, but part of the process_data)
159 [gras_stub_generator]
160 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
162 2004-07-03 Martin Quinson
163 - Create a new log channel tbx containing dict, set, log, dynar (to shut
164 them all up in one shot)
166 - Fix the ugly case of reference to dynamic array.
167 - New (semi-public) function gras_datadesc_size to allow the messaging
168 layer to malloc the needed space for the buffer.
170 - gras_socket_close now expect the socket to close (and not its address to
171 put NULL in it after it). This is because the socket passed to handlers
172 is one of their argument (=> not writable).
174 - propagate the interface cleanup from last week in datadesc, ie remove a
175 superfluous level of indirection. User pass adress of variable
176 containing data (both when sending and receiving), and not of a variable
177 being a pointer to the data. Let's say that I like it better ;)
178 The price for that is constructs like "int msg=*(int*)payload" in
179 handlers, but it's a fine price, IMHO.
181 - Let it work in RL (yuhu)
183 2004-06-21 Martin Quinson
185 - porting SG plugin and SG select to new standards (works almost).
186 - plug memleaks and fix bugs around.
189 - cleanup the prototype of data recv and force users to specify when they
190 want to handle references to objects. Test case working even for cycles.
191 - plug memleaks. Valgrind is perfectly ok with this.
193 2004-06-12 Martin Quinson
195 - cleanup the separation between plugin and main code in plugin creation
197 2004-06-11 Martin Quinson
199 - Reput hook for raw sockets, needed for BW experiments
200 - kill a few lines of dead code
201 [Data description] Interface cleanup
202 - gras_datadesc_by_name returns the searched type or NULL.
203 That way, no variable is needed to use a type desc once, which makes
205 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
206 two parameters were strings (field name, type name), leading to
208 [Dicos] Interface cleanup
209 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
210 This is consistant with the dynar API.
212 2004-04-21 Martin Quinson
214 - Porting to new standards.
217 There is no bag anymore, no need to take extra provision to mask the
218 pointers behind "ID".
219 Better splitup of functions between files create/exchange/convert.
220 This is still a bit artificial since convert and receive are so
221 interleaved, but anyway.
223 - add a queued message list to procdata (the ones not matching criteria
225 - factorize some more code between SG and RL wrt procdata
227 - use gras_exit in example to track memleaks
228 - get rid of gs_example now that GS is properly integrated into gras
229 - update run_test to integrate the lastest tests (datadesc)
231 - rename WARNINGn macros to WARNn since it prooved error-prone
233 2004-04-19 Martin Quinson
235 - register init/exit functions within gras module mecanism
236 - send/receive function.
237 Convertion is not implemented, but short-cutted if not needed.
238 struct/array elements are sent one by one (instead of block-wise), but
239 nobody really cares (yet). Get a prototype before optimizing.
240 - tests (using a file socket) for DD send/receive on:
241 - base types: int, float
242 - array: fixed size, string (ie ref to dynamic string)
243 - structure: homogeneous, heterogeneous
244 - chained list, graph with cycle
245 Believe it or not, valgrind is not too unhappy with the results. The
246 cycle happily segfaults, but the others are ok. And I'm sick of pointers
250 - Bugfix when using a filename explicitely (instead of '-')
252 2004-04-09 Martin Quinson
254 - factorize more code between RL and SG in socket creation
255 - Complete the implementation and tests of:
257 o file (only in RL, and mainly for debugging)
259 I lost 3 days to design a portable address resolver, and then decided
260 that the prototype mainly have to run on my box.
261 Addressing portability too early may be like optimizing too early :-/
263 - use gras_init in the Tests instead of the crappy parse_log_opt
264 (the latter function is removed)
265 [Conditional execution]
266 - New functions: gras_if_RL/gras_if_SG (basic support for this)
267 [Code reorganisation]
268 - Get rid of libgrasutils.a since it makes more trouble than it solves.
269 Build examples against the RL library, since there is no way to disable
270 its creation for now.