%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%% generic configuration file for %%%% %%%% the ccmalloc memory profiler %%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %-----------------------------------------------------------------% % COPY THIS FILE TO '.ccmalloc' in your project or home directory % %-----------------------------------------------------------------% ############################################################################## ## (C) 1997-2003 Armin Biere, 1998 Johannes Keukelaar ## $Id: ccmalloc.cfg,v 1.6 2003/02/03 08:03:54 biere Exp $ ############################################################################## %%% '%' and '#' are comments !!!!!!! % This file must be called '.ccmalloc' and is searched for in the % current directory and in the home directory of the user. If it % does not exist then the default values mentioned below are used. % It is also the only available user manual yet ;-) So here is a reading % hint. First have a look at the short one line descriptions of each option % ... %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % with 'file' the executable is specified [a.out] % ---------------------------------------------------------------------- % This should not be necessary for Linux and Solaris because the proc % file system can be used to find argv[0]. % % (the rest of this comment only applies to other OS) % % For other OS you should use this option unless the executable is % in the current directory or its name is 'a.out'. % % If you do not specify this then ccmalloc tries to find an executable % in the current directory that matches the running program starting % with 'a.out'. For this process it must call 'nm' on each executable % file in the directory which may be time consuming. With this option % you can speed up this process. % % You can also specify absolute or relative path names. This is % necessary if you do not start your program from the current directory. % But you can also simply link or name your program to 'a.out'. %file FILE %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'log' specify the logfile [stderr] % ---------------------------------------------------------------------- % The default is to use stderr. The argument to 'log' is the name of % the file you want to write to. It can also be 'stdout' or '-' which % sets stdout as logfile. If the logfile is stdout or stderr and is % connected to a terminal then the output is slightly different. % % For big programs the logfile can be really big. To reduce the size % you can use a small chain length (see 'chain-length' below). The other % possibility is to use compressed logfiles. This can be done by % specifying a logfile name with a '.gz' (or a '.Z') suffix. This means % that gnuzip (resp. compress) is used to compress the output. %log FILE %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'logpid' specify the logfile % ---------------------------------------------------------------------- % Can be used alternatively to the 'log' command if you want to use % ccmalloc for debugging parallel applications where several copies of % the program you are debugging must be run simoultaneously. In this % case you can not use 'log' because you do not want to write to the same % log file. Using 'logpid' uses a file name ending with the of % the process which means the name is unique even if several copies of % your program are run simoultaneously. % % If you use the compressing suffixes then the is inserted before % the suffix (e.g. 'logpid ccmalloc.log.gz' uses 'ccmalloc.log..gz' % as the name for the log file). %logpid FILE %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'dont-log-chain' skip info about certain chains [] % ---------------------------------------------------------------------- % This command may be repeated any number of times. The argument to this % command is a comma-separated list of function-or-file-and-line % specifications. Garbage allocated from a callchain that contains this % subchain anywhere will _not_ be logged. % % The ';'-separated list should not contain any spaces. E.g. not: % % main ; foo ; bar % % but: % % main;foo;bar % % A function-or-file-and-line specification is a string followed by an % optional colon and number, for example: main or main:14 or main.c or % main.c:15. Note that the string is compared with both the function and the % file name, if available. If main.c happens to be a function name, that % will cause a match (for that string at least). Not specifying a line % number will match any line number. If line number information is not % available, anything will match! Not specifying a name (e.g. ;;;) will % match an unknown function name. Not giving any parameters at all, will % match a chain containing at least one unknown function. % % Note that if you say 'dont-log-chain wrapper.c' nothing will be logged. %dont-log-chain %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'only-log-chain' skip info about other chains [] % ---------------------------------------------------------------------- % The obvious counterpart to dont-log-chain. In this case, only matching % chains will be reported. Non-matching chains will not be reported. % Can be repeated any number of times; if the chain matches any of the % instances, it will be reported. %only-log-chain ######################################################################## # # # This is the 'flag' section # # # # 'set FLAG' is the same as 'set FLAG 1' # # # # The default values are those set below. If 'silent' is disabled # # then you will find the banner in the log file (or it is listed on # # stdout or stderr). The banner describes the current settings of all # # these flags. # # # ######################################################################## %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % with 'only-count' ccmalloc only counts garbage - no call chains [0] % ---------------------------------------------------------------------- % If only-count is set to one then only one additional pointer for % each allocated data is used and no call chain is generated. This is % the fasted and most space efficient mode ccmalloc can operate % in. In this mode you get at least the size of garbage produced. % % Note that 'check-free-space' does not work at all with 'only-count' % set and over writes ('check-overwrites') are only checked when % calling free. %set only-count 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'load-dynlibs' load dynamic linked libraries into gdb [0] % ---------------------------------------------------------------------- % If your program is linked with dynamic libraries, function and file % name information is not available for addresses in those libraries, % unless you set 'load-dynlibs' to 1. %set load-dynlibs 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'keep-deallocated-data' does not recycle deallocated data [0] % ---------------------------------------------------------------------- % If you enable keep-deallocated-data then all data deallocated with % 'free' (or 'delete' in C++) is not given back to the free store % but stays associated with the call chain of its allocation. This is % very useful if your program does multiple deallocation of the % same data. %set keep-deallocated-data 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'check-overwrites' detect overwrites [0] % ---------------------------------------------------------------------- % If you want to detect 'off by n bytes' errors you should set % 'checking-overwrites' to n/4 (on 32-Bit machines). % % ccmalloc inserts a boundary above allocated data. This boundary % consists of 'check-overwrites' words. If your program writes to % this area then ccmalloc can detect this (see also check-start % and check-interval). 'ccmalloc' also does checking for overwrites % at non word boundaries (e.g. strcpy(malloc(strlen("hello")),"hello");) set check-overwrites 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'check-underwrites' detect underwrites [0] % ---------------------------------------------------------------------- % same with writes below allocated data. You do not have to set this % option if you only want detect 'off (below) by one' errors because % ccmalloc keeps a magic value just before the user data. set check-underwrites 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'check-free-space' can be used to find dangling pointers. [0] % ---------------------------------------------------------------------- % A very serious type of bug is to write on data that has already been % freed. If this happens the free space management of malloc is in % trouble and you will perhaps encounter non deterministic behaviour of % your program. To test this first enable 'keep-deallocated-data' and % restart your program. If the problem goes away and ccmalloc does not % report anything then you should *also* enable 'check-free-space'. Now % ccmalloc checks already deallocated data for corruption. % % Note that to perform this check 'keep-deallocated-data' also must % be enabled and 'only-count' disabled. set check-free-space 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'check-interval' can be used to speed up checks [0] % ---------------------------------------------------------------------- % If check-overwrite, check-underwrites or check-free-space is set then % the default is to do 'write checks' when data is deallocated and % to do 'free space checks' when reporting together with % 'write checks' for garbage. When you want these checks to be % performed more often then you should set 'check-interval' to a % positive number. This number is the interval between the number of % calls to free or malloc without performing the checks. %set check-interval 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'check-start' can be used to speed up checks [0] % ---------------------------------------------------------------------- % The flag 'check-start' delays the start of checks until the given % number of calls to free and malloc have occured. Together with % 'check-interval' you can use a binary search to find an aproximation % when a corruption occured! If you simply set check-interval to 1 and % check-start to 0 then this will slow done your program too much. %set check-start 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'silent' disables banner [0] % ---------------------------------------------------------------------- % If you don't want to see the banner of ccmalloc then set % 'silent' to 1 (f.e. when logging to stderr) %set silent %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'file-info' en/disables file and line number information [1] % ---------------------------------------------------------------------- % If your program was compiled with debugging information (-g) then % ccmalloc can generate line number and file info for call chains opening % a pipe to gdb. For very big programs this method is slow. In this case % you can set 'file-info' to zero and you will only get the function % names. For SunOS 4.3.1 'nm' does not 'demangle' C++ identifiers % very well. So gdb is called instead but only if 'file-info' is % not set to 0. %set file-info 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'continue' if ccmalloc aborts when something weired happened [0] % ---------------------------------------------------------------------- % If the free function of ccmalloc is called with an argument that does % not make sense to ccmalloc or that has already been freed then you % probably want the program to stop at this point. This is also % the default behaviour. But you can force ccmalloc also to ignore % this if you set 'continue' to 1. This flag also controls the behaviour % of ccmalloc when free space is found to be corrupted or a write % boundary has been overwritten. %set continue 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'chain-length' is the length of the maximal call chain [0 = infinite] % ---------------------------------------------------------------------- % You can restrict the length of call chains by setting 'chain-length' % to a number greater than zero. If 'chain-length' is zero (the default) % then chains are as long as possible (on a non x86 system only call % chains with a finite maximal length can be generated). For big % programs especially if keep-deallocated-data is enabled this can % reduce the size of the log file from over 100MB to several MB! %set chain-length 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'print-addresses' of data [0] % ---------------------------------------------------------------------- % If you want to see the addresses of the allocated data (and % deallocated data if keep-deallocated-data is set to 1) set % 'print-addresses' to 1. %set print-addresses 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'print-on-one-line' shortens log file [0] % ---------------------------------------------------------------------- % The default is to print function names and file/line number info % on separate lines. With 'print-on-one-line' set 1 all are printed % on one line. %set print-on-one-line 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'additional-line' enlarges readability [1] % ---------------------------------------------------------------------- % When printing call chains an empty line is printed between to % call points. Set 'additional-line' to 0 to disable this feature. %set additional-line 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 'statistics' enables more accurate profiling [0] % ---------------------------------------------------------------------- % Calculate number of allocations and deallocations and bytes also on % a per call chain basis. This uses 4 additional pointers for each % call chain. set statistics 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % set order for sorting of call chains [1] [1] % ---------------------------------------------------------------------- % When printing the report to the log file the call chains are sorted by % default with respect to the largest accumulated garbage produced by % that call chain. This can be changed with setting 'sort-by-wasted' % to 0. In this case they are sorted by the number of allocated bytes. % If you want the number of allocations (only possible if 'statistics' % is enabled) as sorting criteria instead then set 'sort-by-size' to 0. %set sort-by-wasted 1 %set sort-by-size 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % report library chains [0] % ---------------------------------------------------------------------- % Some external libraries (like libg++) have memory leaks. On some % systems even a call to printf produces a leak. ccmalloc tries to % detect this (only heuristically!) and with this flag you can control % if leaks produced by such library calls are reported. % % Since version 0.2.1 some similar effect can be achieved by using % 'dont-log-chain' with no argument. %set library-chains 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % print debugging information [X] (compile time dependend) % ---------------------------------------------------------------------- %set debug X %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % align memory on 8 byte boundary [0] (no effect on SunOS or Solaris) % ---------------------------------------------------------------------- %set align-8-byte 0 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % only report allocations which ended up being wasted (i.e don't report % allocations which were completely freed properly. ) [1] % ---------------------------------------------------------------------- %set only-wasting-alloc 1