libcwd-0.99.44 is released.  Anyway, this release implements the possibility of 'redzones': you can configure libcwd, using --with-redzone=x (in bytes), and it will add (at least) x bytes above and below every allocation. See the NEWS file for more info. This has been quite a while in CVS, but now also as tar ball. People who build libcwd from CVS (and therefore need to run autogen.sh) be aware: my autoconf macros are now put in a separate project called cwautomacros (I use them for many different projects). You will need to download and install cwautomacros too. Two bugs have been fixed: one that could cause a core dump during global deinitialization (when the application terminates) and one that could cause a COREDUMP for a self-"collision" of a shared library at start up, namely when one tried to dlopen a library is already linked to the application in the normal way.

libcwd-0.99.43 is released.  This release fixes problems with the calculation of the load address of shared libraries. On some systems, this bug resulted in not finding a shared object for a given address, finding the wrong one, or even core dumping on start up.

libcwd-0.99.42 is released.  Compiler warning fix for g++ 4.x in private_allocator.h.

libcwd-0.99.41 is released.  This release adds support for g++-4.0.2.  Addresses in libraries that are loaded on addresses lower than the executable (yeah, I didn't know that was possible either) are now resolved.  dlopen/dlclose are now caught at link time instead of #defining them, solving a bug that could cause a crash when for example dumping memory allocations done in already dlclosed modules.  Finally, the .spec file was fixed so that 'rpmbuild -ta libcwd-0.99.41.tar.gz' should work again on the latest Fedora release (that uses a new version of rpm).

libcwd-0.99.40 is released.  This release adds support for g++-4.0.0.

libcwd-0.99.39 is released.  This release fixes a few major bugs.  The most important one caused libcwd to crash even before reaching main when the compiler used is configured with --enable-nls, making libcwd completely useless on for example Gentoo.  A second important bug that was fixed is one where the shared libraries that one uses are stripped (as is also the case on Gentoo).  In this case libcwd failed to determine the end of a shared library because the symbol "_end" is missing and it took the end of the first function in the library by accident as end of the library, instead of the last function.  As a result, addresses inside such stripped libraries could not be found by libcwd - and worse, trying to print such locations caused a core dump!

Special thanks go to Michal Grzejszczak for reporting the problems with Gentoo; not many people take the effort to report bugs these days, especially for a package that crashes immediately the first time you try a 'Hello World'! Thanks to him also Gentoo users can use libcwd from now on.

libcwd-0.99.38 is released.  This release adds support for gcc 3.4.2 (needed for the threaded case) and for the latest gcc CVS version 4.0.0 (20040922).  The configuration option --disable-location has been fixed, and when using this together with --disable-alloc you will now be able to use libcwd for debug output on cygwin and mingw32 (no threading yet).

Along with a few other minor bugs, a rather nasty one that appeared first on fedora core 2 machines have been fixed.  If you have been getting the following error when starting an application linked with libcwd:
FATAL : std::ifstream.open(""): ENOENT (No such file or directory)
Then you should certainly upgrade as this release will fix that.

libcwd-0.99.37 is released.  This release concentrates on improving memory leak detection.  Unfortunately it had to contain again an API change: ooam_filter_ct has been renamed to alloc_filter_ct and ooam_format_t has likewise been renamed to alloc_format_t.

With this release it is possible to filter the list of printed allocations on allocation-function (the function from which the allocation is done).  This makes it in principle possible to print a precise overview of leaked memory at process termination by suppressing a list of "leaks" that are not really leaks.  Memory allocations now print the location where they are done at the moment they are done - and not only when they are freed.  The memory allocation filter got a new flag show_function which causes the mangled allocation-function to be shown (even when the sourcefile and line number are known).  I left this deliberately in mangled form because it is more likely to be used in a regular expression search in your debug output log file than to be understood by a human (and you can always use c++filt to demangle it).  Finally, the function list_allocations_on now returns the number of visible, non-filtered allocations; a value larger than 0 can then be interpreted as 'memory leak'.  A pair of functions has been added to more easily control whether third-party allocations are invisible or not: libcwd::set_invisible_on and libcwd::set_invisible_off.  Memory allocated while this flag is set will be invisible and the MALLOC showing the allocation will have '(invisible)' appended.  This line now also always shows the location where the allocator is called from.

This release fixes the --disable-alloc configure flag and adds support for gcc version 3.4.1 (needed for threaded applications).

libcwd-0.99.36 is released.  A very minor bug fix release.

libcwd-0.99.35 is released.  This release fixes a bug related to making memory allocations invisible.  Before it was not possible to make an allocation invisible that was allocated with realloc(2) due to an assertion failure.  Library authors who use libcwd will need to carefully read the NEWS file and The Custom debug.h File / Libraries paragraph in the reference manual, because the way that it has been set up has been changed.

libcwd-0.99.34 is released.  Although this is mainly a bug fix release, it also contains a major API change: the namespace libcw::debug was renamed to namespace libcwd. Libcwd is now an official debian package and see, as a result I got some bug reports (finally!).  It turned out that libcwd had problems with certain shared libraries that use more than 256 entries in the .debug_abbrev section.  Finally, there turned out to exist two global initialization order fiasco bugs for the non-threaded case; one showing on debian and one showing on gentoo.  All reported bugs have been fixed.

libcwd-0.99.33 is released.  Added support for gcc version 3.4.0. The latest gcc CVS version 3.5 is working again.  Two new features have been added: debug_alloc(ptr) and debug_watch(ptr).  Both can only be called from within gdb.  The first prints information about the memory allocation (if any) under the pointer ptr and the second causes gdb to stop as soon as the memory allocation under ptr is deleted or freed.

libcwd-0.99.32 is released.  This release fixes a possible deadlock during initialization, in the threaded case.  The latest gcc CVS version 3.4 and 3.5 are working again.  Two new features have been added: read_rcfile() and attach_gdb().  The first reads initialization from an rcfile (allowing to set which debug channels should be on or off for example) and the second opens an xterm with an attached gdb session in it, allowing you to start to debug the application from that point (especially handy for threaded applications).

libcwd-0.99.31 is released.  This release really works with glibc-2.3.2 and RedHat 9.  I missed the fact that RedHat 9 supports NPTL (Native POSIX Threads Library) and TLS (Thread Local Storage) (TLS needs at least g++ 3.3 or the version of 3.2.3 that comes with RedHat 9).  This release reimplements the way that libcwd deals with TSD (Thread Specific Data) and now really works with NPTL.  Only the version of NPTL that comes with RedHat 9 was tested though (0.35, the current development version of NPTL is 0.50).  You will need at least g++ 3.2.1 in order to get threading to work when you have an NPTL enabled glibc, this is due to a bug in NPTL-0.35 but that bug doesn't cause any problems with g++ 3.2.1 and higher.

libcwd-0.99.30 is released.  This release works with glibc-2.3.2 and RedHat 9.  I found that the glibc that is distributed with rawhide is broken. A few threads related bug have been fixed among which a serious one that caused allocations done between two list_allocations_on()'s using the same filter (or none at all, which uses the default filter) to be randomly filtered (depending on an uninitialized boolean).

libcwd-0.99.29 is released.  This release contains a few large changes.  All the header files have been moved from libcw/ to libcwd/.  Compiler version 3.x is now required (2.95 and 2.96 didn't support stringbuffers well enough).  All allocations inside Dout() et al are now non-internal!  Finally there were a few minor bug fixes.  Please read the NEWS file as usual, for more details.

libcwd-0.99.28 is released.  This release adds support for new(std::size_t, std::nothrow_t const&) allowing it to be used with gtkmm-2.0 among others.  The demangler was fixed to match the mangling of g++ 3.1, which uses a slightly different way to mangle constant member functions than g++ 3.0 did.  The decoding of line numbers has once again been improved and is now perfect (assuming you don't use a compiler that generates broken debug information).  Finally, this release fixes a bug where dlopen() could only read libraries with an absolute path, failing for libraries that reside somewhere in LD_LIBRARY_PATH.

libcwd-0.99.27 is released.  This release fixes a few minor bugs. Well, minor unless you run into them of course: libcwd would core dump when reading a shared library whose last compilation unit did not contain a .debug_line section. And linking with a libcwd that was configured with --disable-alloc would lead to an undefined reference. Finally, demangling of _GLOBAL_* functions have been fixed.

libcwd-0.99.26 is released.  This release contains some bug fixes and a new 'overview of allocated memory' filter class extension: ooam_filter_ct::hide_unknown_locations(bool).

libcwd-0.99.25 is released.  This release fixes a few major bugs: internal memory leaks in libcwd itself (now garanteed leak free!), and the demangle routines weren't thread safe when you were using compiler version 2.95.x or 2.96.  Also, the library was erroneously not compiled with -O, making it about twice as slow (still very fast).  The function list_allocations_on() didn't work at all for other debug objects; only list_allocations_on(libcw_do) would work.  Under certain circumstances (linking with a third party C++ library that got initialized first and whose first allocation is via std::allocator) would lead to a dead lock during initialization.  Finally, channel_ct::get_label() has been changed to be zero terminated.  The fact that it was not (deliberately), wasn't very clearly documented.

libcwd-0.99.24 is released.  This release fixes multi-threading problems among which writing continued debug output: the label of each output line will now really always start on a new line.

Today an excellent review of C++ tools appeared on the front page of kuro5hin.org in which libcwd received positive critics.

libcwd-0.99.23 is released.  This release adds two possibilities to the 'overview of allocated memory' filter class: ooam_filter_ct::hide_sourcefiles_matching(std::vector<std::string> const&) and ooam_filter_ct::hide_untagged_allocations(bool).
You now can pass a filter to a marker_ct to specify which allocations are to be considered outside the marker area as well as how to list the allocations that were leaked.  Three minor bugs have been fixed: two related to the use of dlclose() and one that caused the full path of location source files to be incomplete when using gcc 3.x.

libcwd-0.99.22 is released.  Fixes a horrible bug that was introduced in 0.99.20, causing any realloc() to crash when not having used an AllocTag for it.  Did not run into this before because all of the testsuite uses AllocTag for reallocated blocks, and libstdc++ doesn't use realloc at all.  I now ran into it because I am playing with libgtkmm.

libcwd-0.99.21 is released.  This release adds support for g++ 3.1.  In the previous two versions there was an include missing causing libcwd not to compile on SuSe and Mandrake.  The administration of allocated memory is now kept in an STL map per thread instead of one big global map.  The advantage of that is that in general a call to new/malloc etc. will not cause a thread to have to wait for other threads anymore.

libcwd-0.99.20 is released.  This release fixes three (possible) deadlocks in list_allocations_on, one in dlopen (when an error occured during loading), and one during core dumping (COREDUMP) in pthread_kill_other_threads_np.  Furthermore a bug was fixed in relation to the use of the control flag error_cf as that was using the non-reentrant libc function strerror instead of strerror_r, causing libcwd likely to complain about freeing an 'internal' allocation (which means that the user (or the system) is freeing memory that was allocated by libcwd, and libcwd likes to take care of its own allocations).

Support for non-threaded applications on solaris 2.8 has been added as well.  Please contact me if you are interested in making threads work too on this OS.  The same holds for FreeBSD.

libcwd-0.99.19 is released.  This release fixes a deadlock during the initialization of threaded applications that use g++ version 2.96 or earlier. There are now two libraries compiled and installed: a thread safe version which is called libcwd_r, and the thread unsafe version called libcwd. Finally, a new feature is added that allows one to filter the overview of allocated memory, removing the output related to certain shared libraries; restricting the memory allocations shown to a given time interval in which they were allocated; and showing the full path of a location source file at which the allocation took place, the time at which the allocation was made and/or the name of the shared library an allocation belongs to. Detailed information can be found in the reference manual.

libcwd-0.99.18 is released.  This release contains several minor documentation fixes and again an API change: the configure option related macros (DEBUGMALLOC, DEBUGDEBUG etc) have been renamed and are defined to 0 or 1 instead of being undefined or defined.  As always, read the NEWS file!

A project has been created on sourceforge especially for libcwd.  This means that libcwd also has a new home page now (you're probably looking at it): http://libcwd.sourceforge.net/.
Bookmark this page!

libcwd-0.99.17 is released.  This release fixes a few bugs related to threading.  Also a few typos in the tutorial have been fixed and a new chapter on debugging threaded applications was added.

Errata: There is a typo in the documentation.  The environment variable isn't LIBCWD_ALWAYS_PRINT_LOADING but LIBCWD_PRINT_LOADING.

libcwd-0.99.16 is released.  Five months of non-stop, hard work, but here it finally is: libcwd is now thread-safe!  In order to get a thread-safe version you have to use the configuration option --enable-libcwd-threading.  This release also contains new documentation that was generated with doxygen; no more slow on-line browsing, now you can read the reference manual locally.

There are a few major API changes again (I know, I know - but if you don't like it then wait till release 1.0).  If you are upgrading from a previous version: Read the NEWS file that comes with the source distribution for details about these API changes! 

libcwd-0.99.15 is released.  This release contains one API change: you now must define CWDEBUG when compiling.  Using -DDEBUG will not work any longer.  A few major bugs have been fixed in this release. Two demangle bugs, a bug related to symbol name lookup and a compilation error (gcc-3.0 couldn't find the header file gthr-default.h, which is actually a bug in the header files of the compiler).  A bug in the demangler for gcc-2.96 and earlier caused core dumps for symbols that started with _X....  Since those are used a lot in X libs, libcwd was unusable with X windowing system applications (like those using qt).  This version of libcwd has been tested (for the first time) with a single threaded qt application.  Another major problem was an erronous assertion in the ELF symbol lookup code that failed on glibc-2.2.4, making libcwd unusable with that version.

libcwd-0.99.14 is released.  Nobody did notice it (nobody mailed me about it), but under certain circumstances the previous version could horribly fail in determining the correct line number of a location.  This version should fix that!  There is more news however, we have a new developer for libcwd!  Eric is an experienced coder who I've known for about eight months now (on our IRC channel #g++ on Undernet).  I've already been enjoying many fruitful discussions with him before, and now he joined the coding forces in order to attack the thread-safety problem.  We're hoping to make libcwd thread-safe in the foreseeable future.  In the meantime you can see what we're doing using CVS and the unstable alpha branch: branch-threading.

libcwd-0.99.13 is released.  New in this release is support for dynamically loaded shared object files (dlopen()).  Also new is native support for the DWARF 2 debug format (this is used when you compile with -ggdb or -gdwarf-2.  It is also used for -g when DWARF is the native debug format, as is the case on IRIX 6).  The advantage of DWARF is that it is much faster to load: it is very compact and stores the line number information in a seperate section so that we don't have to load other debug info.  In the case of stabs (the native debug format of linux) we need to load all debug information.  You'll notice that libcwd now prints "Loading debug info from ..." under all circumstances when an application is started.  It uses debug channel dc::bfd and in most cases before main() is reached.  This is achieved by forcing the debug object as well as dc::bfd to on prior to loading the debug information from the object files.  This was disabled in previous versions because it's a bit tricky: at this point the debug object might not be initialized yet!  Previously, only a snap shot version of g++ that I kept around caused a crash because of this (it depends highly on the order in which global objects are initialized), but now I made a few changes that make me feel confident enough to enable it by default.  If you still get a crash, or if you want to disable this printing for other reasons, then you can #undef ALWAYS_PRINT_LOADING in src/libcwd/bfd.cc.

This release contains two API changes - and rather annoying ones I am afraid: Firstly, the header file <sys.h> has been renamed to <sysd.h>.  Please manually remove the old <sys.h> if you install libcwd over an old install: that is the only way to be sure that you are not accidently including it somewhere.  Secondly, the debug channel dc::warning is turned on by default - which will cause a core dump if you try to turn it on yourself at the start of main(): You have to remove an explicit Debug( dc::warning.on() ) if you have one.

libcwd-0.99.12 is released.  This release fixes the bug that libcwd was accessing recently freed memory.  In all that time this never lead to a crash for me, amazing.

libcwd-0.99.11 is released.  Debug channel dc::stabs was renamed back to dc::bfd.  Possible confusion was now solved differently by renaming the configure option --enable-libcwd-bfd into --enable-libcwd-libbfd.

libcwd-0.99.10 is released.  This release comes with two major changes.  malloc(3) and friends are now exported with external "C" linkage instead of being a macro; this means that also allocations done from other shared libraries like libc are caught now.  Most notably, this fixes the bug that strdup(3) couldn't be used (thanks for the feedback I now got!).  This release doesn't depend anymore on GNU's libbfd; it is now capable of understanding stabs debug info read from ELF32 object files directly.  You will need to use the configure option --enable-libcwd-bfd to get the old behaviour back and still link with libbfd.  Note that due to the viral nature of the GPL (that doesn't allow one to link a program with a GPL-ed library and then distribute it under any other license but the GPL (unlike the LGPL and QPL)), you are not allowed to distribute executables that are linked against libbfd because you could only do so under the GPL and that violates the license of libcwd (QPL).  I trust that this is not a problem as these are purely test-executables.

libcwd-0.99.9 is released.  This release comes with a brand new demangler for the new C++ ABI as promised.  I am happy to have learned that there exists interest in libcwd in the Open Source community: the maintainer of gdb wrote me that he translated the new demangler into C and will incorporate it into GNU libiberty and gdb (the current demangler in libiberty doesn't handle qualifiers correctly in all cases).  The maintainer of clisp expressed his interest in the bfd.cc file in order to be able to print backtraces in case of internal errors.

I have very little feedback from users of libcwd itself though.  If there is anything you would like to be added, or when you find a bug, mail me.  It won't get fixed when you don't tell me about it!  But if you are just a happy user without any complaint, I'd really love to hear from you too! Click here and drop me mail!

libcwd-0.99.8 is released.  Finally, after a long delay a new release - and as promised one that works with libstdc++ version 3! This release still lacks a demangler for the new C++ ABI however (that is part of g++-2.97 and higher), which will be added to the next release.

Making libcwd suitable for libstdc++-v3 wasn't an easy task.  One of the major problems is that during the writing of debug output (for example operator<<(ostream&, int)) memory is being allocated under certain circumstances by libstdc++-v3.  While it is unacceptable that due to writing debug output other debug output is generated, there were only two options: 1) turning off the malloc and bfd debug channels but still storing all memory allocations in memory allocation overview, or 2) making all memory allocations done from a Dout internal, disabling any checks.  With in mind that the existance of debug code shouldn't matter (after all, it won't be in the final production version of an application) it follows that no memory should be allocated in the first place as part of a Dout and when it is it isn't important if it leaks or not - I decided to go for solution two because that uses less cpu.

Other changes involve mostly a rewrite of certain parts that were depending on a specific GNU implementation of libstdc++, using the now available standard.  While most of these changes are invisible, there was a small change to the interface of type_info_of resulting at the same time in preserving a top-level const qualifier as well as faster code (with g++-3.0 that is; older versions of g++ are still so much broken that slower work-arounds were needed).  The only kludge left in libcwd at the moment is the detection of when the standard streams are initialized.  I blame the need for this kludge on a bug in libstdc++-v3 and have reported it as such; perhaps also this can be gotten rid of in the future.

At the moment of writing two problems are already known with 0.99.8:

Using a libbfd on a 32bit machine that was compiled with 64bit target support will fail to work with libstdc++-v3 when the latter is compiled without support for long long (long long is not a part of the C++ standard, while it is part of C99.  libbfd being a C library, this results in problems).  I added a test for this to 0.99.8 but accidently forgot to test for libstdc++-v3 causing libcwd to fail always with g++-2.95.x! If you get this error: libbfd is compiled with 64bit support on a 32bit host, but your libstdc++ is not compiled with support for long long. and you are using g++-2.95.x then just remove the #error line from bfd.cc and it should compile just fine.

Finally, it is known that many operating systems can't deal with shared C++ libraries and global objects.  While this means that those operating systems are broken, I decided to fix this problem by removing all global objects in the next release.  So, if libcwd core dumps on you for even very simple programs then please try 0.99.9 when it gets released.

libcwd-0.99.7 is released.  This release includes a work-around for the compiler crash in lockable_auto_ptr.h that occured with g++-2.95.x.  There are a few other changes however; most notably, the support for the macro Dout_vform has been removed because ostream::vform isn't conforming to the Standard and is gone in libstdc++ version 3.  Note that this release still doesn't work with libstdc++ version 3, hopefully the next release will.

libcwd-0.99.6 is released.  This release deals with namespace issues and therefore contains a few major API changes.  The most important changes follow.  The macro DEBUG has been renamed to CWDEBUG because the former was too general and collided with other libraries.  All functions have been moved to namespace libcwd; this will break compilation of programs that use earlier versions of libcwd, unless one includes a "using namespace libcwd;" line in the source files using more than just Dout and Debug.  Four functions have not only been moved to the new namespace but were also renamed.  Apart from this all, also the interface to BFD has been redesigned; among others the structure location_st has been replaced by a class location_ct.  Also here functions have been renamed or even removed.  Please read the NEWS file contained in the source distribution for details.  A new template function has been added: cwprint_using.  This utility allows one to easily print objects using a method of the object rather than operator<<.  Finally, two bugs have been fixed: one in the demangler which still couldn't demangle const member functions, and a bug that could cause a core dump when memory was allocated from the constructor of a global object that was initialized before libcwd was initialized, this has only been reported for solaris.

libcwd-0.99.5 is released.  This release contains mostly bug fixes, especially for the demangler.  Otherwise it mainly contains support for the compiler that is released with RedHat-7.0: gcc-2.96.  Version 0.99.5 contains two API changes: the prototype of libcw_bfd_pc_location was changed in order to get rid of the named return values gcc extension that has been deprecated in g++-2.96 and higher.  Secondly, DoutFatal doesn't have a default debug channel anymore and dc::fatal, which was the default in prior releases, must explicitly be specified.

libcwd-0.99.4 is released.  This release comes with a testsuite and again allows Linux users to just link their application with -lcwd (without the need to also specify -lbfd -liberty).  All reported bugs (two) have been fixed, as well as a third that showed up while I wrote the testsuite: a locked lockable_auto_ptr does not transfer ownership anymore when using the assignment operator.

libcwd-0.99.3 is released.  This release includes an example showing how to use libcwd in an autoconf/automake project.  Please note that, due to constraints of libtool, your programs need to be linked with -lcwd -lbfd -liberty now (as opposed to just -lcwd).  Support for Solaris has been added.  The autoconf/automake configuration setup has been improved; also the configuration on FreeBSD should work out of the box now (assuming you have binutils installed).  Finally, a .spec file for building an RPM has been added to this release.

libcwd-0.99.2 is released.  Added autoconf/automake/libtool for configuration.  This obsoletes the need for the prototype package.  The source file demangle.cc was completely rewritten and now supports demangling of symbols as well as types.

libcwd-0.99.1 is released.  Apart from a few minor bug fixes, support for FreeBSD has been added!

The web site has been improved in order to give new visitors a quicker overview: libcwd now has its own home page.  A page with features and "screenshots" was added.

First public announcement of libcwd on freshmeat.

First public release of libcwd on sourceforge.