------------------------------------------------------------------------------ NEWS for version 0.3.9 ------------------------------------------------------------------------------ o fixed configure to work on FreeBSD o patched configure to correctly set 'HAVE_ATEXIT' and also detect the path of the default dynamically linked libc. This patch fixes some problems with Redhat 7.1 which uses libc6. o '--version' and '--help' for 'ccmalloc' script o finally allowed 'free(0)' ------------------------------------------------------------------------------ NEWS for version 0.3.8 ------------------------------------------------------------------------------ o checking statically allocated dynamic data for C++ seems to work again o better support for C++: increased buffer size, better lexer for parsing demangling info of C++ template symbols from gdb (Henning Moll). ------------------------------------------------------------------------------ NEWS for version 0.3.7 ------------------------------------------------------------------------------ o fixed 'ccmalloc' script to skip files compiled with '-S' and '-c' o fixed 'ccmalloc' script to filter out '-o ' as well o do not log chain that contain a reference to 'ccmalloc_init' ------------------------------------------------------------------------------ NEWS for version 0.3.4 ------------------------------------------------------------------------------ o Started to work on compatibility with Sun's workshop compiler CC. The idea is to compile 'callchain.c' with 'gcc' and the 'C++' wrapper 'ccmalloc.cc' with 'CC'. Then it should be possible to link and compile your own project with 'CC'. Just link as follows CC -o binary your-obj1.o your-obj2.o ccmalloc-CC.o -lccmalloc -ldl I also added a script 'ccmalloc' that hides this additional arguments to the compiler and makes the usage of ccmalloc similar to the usage of 'purify'. Now the standard way of using ccmalloc with C++ becomes ccmalloc CC -o binary your-obj1.o your-obj2.o or with 'gcc' ccmalloc gcc -o binary your-obj1.o your-obj2.o However, for some unknown reason 'CC' still produces link errors and this feature does not work yet. o Added patches by Fritz Elfert for better command line argument detection under Linux (and actually reenabled the whole detection code). ------------------------------------------------------------------------------ NEWS for version 0.3.3 (patches from Henning Moll) ------------------------------------------------------------------------------ o Fixed off by one bug in reporting line number o Used 'set width 0' command for GDB to produce easier to parse one line output. This removes some 'could not read info from gdb' messages, i.e. with long symbol names as in STL. o Skipped 'Current language: ...' message from GDB for better support of multiple language debugging. o Better heuristics for skipping library callchains. ------------------------------------------------------------------------------ NEWS for version 0.3.2 ------------------------------------------------------------------------------ o Initialized 'logfile' properly o Added '-x' for 'nm' on Solaris o Forced 'gdb' not to read initialization file '.gdbinit' ------------------------------------------------------------------------------ NEWS for version 0.3.0 ------------------------------------------------------------------------------ o Removed support for SunOS (only Solaris and Linux left) o Added the ability for ccmalloc to report only the call chains for wasted allocations, i.e. allocations which were not completely freed properly (Ian Ishmael). ------------------------------------------------------------------------------ NEWS for version 0.2.4 ------------------------------------------------------------------------------ o Moved `banner()' out of the initialization code. This fixed the `ctime' bug: Because `banner()' calls `ctime' itself, and `ctime' might call malloc, the memory potentially allocated by ctime must be public memory, since it could be deallocated from a following call to `ctime' originating outside from `ccmalloc'. ------------------------------------------------------------------------------ NEWS for version 0.2.3 ------------------------------------------------------------------------------ new features: ============= o Change the cmp_CallChains_by_*_for_Qsort functions to not return 0 so often by comparing them on address data, too. (This means we have to read in the data from gdb before sorting the callchains.) This allows one to compare resulting log files from different runs of the same program with, e.g. diff or emacs ediff, without getting too many irrelevant differences (Johannes). o Last, but not least, read address data for dynamic libraries from gdb as well. Only works for those dynamic libraries that the program is linked with, not the ones that are dlopen'ed explicitly. Controlled by a flag called `load-dynlibs' (Johannes). o new flag `align-8-byte' to align memory to 8 byte on Linux. bugfixes: ========= o The configure script now tries to find the standard C library (libc.so) and checks if it can be opened with dlopen. This should make ccmalloc run on Redhat 5.0 Linux out of the box. o GPL ;-) o Fix the src/Makefile so that it also installs the config file and so that install depends on ccmalloc.o (Johannes). ------------------------------------------------------------------------------ NEWS for version 0.2.2 ------------------------------------------------------------------------------ bug fixes: ========== o `wrapper.c' did not handle realloc correctly while in state INITIALIZING. This lead to `*** realloc(0x8065f3d) called with non valid argument' before the program was even started. This bug occured when using KDE together with Qt. Now even the static allocator in wrapper.c saves the size of the allocated data. ------------------------------------------------------------------------------ NEWS for version 0.2.1 ------------------------------------------------------------------------------ bug fixes: ========== o returned data is now 8 byte aligned on SunOS and Solaris. This avoids a Bus Error when allocating heap memory for doubles. (a lot of people reported that bug) o shrinking allocated data with realloc works now (a lot of people reported that bug) o fixed dereferencing bug in cmpAddr (Johannes Keukelaar) o calls fflush before killing itself new features: ============= o `only-log-chain' and `dont-log-chain' (Johannes Keukelaar) o `logpid' (Didier Remy) ------------------------------------------------------------------------------ (C) 1997-2000 Armin Biere $Id$ ------------------------------------------------------------------------------