** ** Source tree organization ** ****************************************************** There is at least 4 sub-projects in the tree: - XBT: eXtended Bundle of Tools (low-level toolbox: logging, datatypes). - SURF: a SimUlation aRtiFact. This is the simulation kernel. - MSG: originally MetaSimGrid, MSG is a simple distributed application simulator. - SMPI: Simulated MPI, to run MPI application using emulation technics. They are all in the same tree because they are complementary tools and having all of them in the same package makes the installation easier for end-users. Moreover, it enables to share the compilation chain and eases the development. The tree is not split on projects, but on file finality: include/ -> all *public* headers include/xbt/*.h -> one file per module src/include -> another location for protected headers. Used by SURF, and other should be converted, since this is the Right Thing. testsuite/ -> The more test the better. Same organization than src/ and include/ Tests are allowed to load some headers of the module they test. All tests should be listed in run_test.in so that they get run on 'make check'. examples/ -> Supposed to be copy/pastable by the user, so keep it clear and avoid any kind of trick. In particular, do only include the public headers here. teshsuite/ -> The more test the better. Put in there any strange test doing things that the users are not supposed to do, just to see if our framework is robust to incorrect and unusual behaviors. All tests written in this section should leverage our tesh(1) utility. testsuite/ -> Old test suite, that should be converted to tesh and moved to teshsuite at some point. ** ** Indentation standard ** ***************************************************** Most files use the Kernighan & Ritchie coding style with 2 spaces of indentation. The indent program can help you to stick to it: indent -kr -l80 -nut -i2 -lps -npcs -br -brs -ce -cdw -bbo -npsl The script ./tools/indent runs indent with the appropriate options. FIXME: this list of arguments is still to be discussed, maybe ** ** Type naming standard ** ***************************************************** It may sound strange, but the type naming convention was source of intense discussion between da SimGrid posse members. The convention we came to may not be the best solution, but it has the merit to exist and leave everyone work. So please stick to it. - ???_t is a valid type (built with typedef) - s_toto_t is a structure (access to fields with .) - s_toto is a structure needing 'struct' keyword to be used - e_toto_t is an enum - u_toto_t is an union - u_toto is an union needing 'union' keyword to be used - toto_t is an 'object' (struct*) Please to not call toto_t something else than an 'object' (ie, something you have to call _new and _free on it). Example: typedef struct s_toto {} s_toto_t, *toto_t; typedef enum {} e_toto_t; Moreover, only toto_t (and e_toto_t) are public. The rest (mainly s_toto_t) is private. If you see any part of the code not following this convention, this is a bug. Please report it (or fix it yourself if you can). ** ** Random bits about coding standards and portability ** ***************************************************** MALLOC: Don't use it, or you'll have to check the result (and do some dirty stuff on AIX). Use xbt_malloc (or even better, xbt_new) instead. SIZE_T (FIXME: obsolete?) If possible, avoid size_t and use unsigned long instead. If not, #include in all files manipulating size_t do cast it to unsigned long before printing (and use %lu), or use %zu. PRINTF pointer difference (FIXME: advertise %td instead?) printf ("diff = %ld\n", (long) (pointer2 - pointer1)); INLINE functions The definition of a inline function must be visible when it is used. As such, an inline function should be defined (an not only declared) in header file (.h) with attributes 'static XBT_INLINE'. It should not be defined in source file (.c). ** ** Commenting the source: doxygen ** **************************************************** The global structure of the documentation is in doc/modules.doc The structure of each module (xbt, msg, etc) is in doc/module-.doc The structure of a module is in its public header. This way, you're sure to see all the public interface (and only it). The different parts of the interface are grouped using the @name construct, even if it's buggy. Since parts often get reordered, it's better to add numbers to the parts (so that users can see the intended order). The documentation of each type and macro are also in the public header since this is were they live. The documentation of each function must be in the C file were it lives. Any public element (function, type and macro) must have a @brief part. ** ** XBT virtualization mechanism (FIXME: this section is deprecated) ** **************************************************** There is some functionalities that we want to virtualize in XBT. We want xbt_time to give the simulated clock when running on top of the simulator, and the host clock when running on a real system. This could be placed in GRAS (and was, historically), but there is some reason to lower it down to XBT. Here is the used naming scheme: - xbt__(): functions working both in SG and RL - xbt_os__(): RL functions usable even in simulator That way, in libsimgrid, we still can use native functions if we want to. It may for example be useful to get the real time when implementing the simulator. Think of the SIGINT handler, which wants to see if the user pressed the key twice in a 5 seconds interval. This is of little use to check the simulated time here. Here is the file layout: - xbt_rl_.c: native implementation (xbt__()). Simply call the corresponding xbt_os__. Part only of libgras.so - xbt_sg_.c: SIMIX implementation xbt__()). Simply call the corresponding SIMIX implementation. Part only of libsimgrid.so - xbt_os_.c: body of the functions implementing natively the stuff (xbt_os__()). Part of both libgras.so and libsimgrid.so Since there is almost nothing in xbt_rl_module.c and xbt_sg_module.c, it'd be better to use symbol aliasing here (to declare in the object code that the same function have two names), but I'm still investigating the portability of the thing to windows. * * SimGrid Hacker Survival Guide (FIXME: should be betterly placed) ******************************** * Before pushing any change, don't forget to check if the compilation passes with compiler optimizations and warnings turned on: cmake -Denable_compile_optimizations=ON \ -Denable_compile_warnings=ON * Your commit message should follow the git habits, explained eg here: http://tbaggery.com/2008/04/19/a-note-about-git-commit-messages.html * When you add/remove files, and/or make changes in the lists of files to build, please check that "make distcheck" still succeeds. This is needed to ensure that the generated archive is consistent. * If you want to debug memory allocation problems, here are a few hints: - disable compiler optimizations, to have better backtraces; - disable the mallocators, or it will be hard to match malloc's with free's; - disable model checking, unless your problem lies in the model checker part of SimGrid (MC brings its own malloc implementation, which valgrind doesn't understand). All this is configured with: cmake -Denable_model-checking=OFF \ -Denable_mallocators=OFF \ -Denable_compile_optimizations=OFF * If you break the logs (for example while hacking in the dynars), you want to define XBT_LOG_MAYDAY at the beginning of log.h. It will deactivate the whole logging mechanism, switching to printfs instead. SimGrid becomes incredibly verbose when doing so, but it you let you fixing the dynars.