A Garbage Collector for C and C++
http://www.hpl.hp.com/personal/Hans_Boehm/gc/
GPS: I've been impressed by how well it works. I think that it might be interesting to build a Tcl-prototype that use the Boehm-GC for Tcl_Obj management. It could possibly eliminate thousands of lines of code and improve the quality of Tcl/Tk.
There was an excellent article in C/C++ Users Journal about the Boehm-GC.
Jacob Levy July 18, 2003: Having used a commercial version of this garbage collector, I would strongly recommend against using it for Tcl. The free version is likely to be worse than the commercial version with which I have experience. Here are some points against using it: First of all, the GC does not work on all the platforms on which Tcl needs to run. Its allocator is about ten times slower than straight malloc. The Boehm-Weiser GC is very conservative -- it doesn't know anything about Tcl's data types, so it cannot determine where all the pointers are in a piece of memory. It also does not know where all the pointers are on the stack and what their real type is, so the end result is that it retains much more memory than actually needed by the application. The GC interacts very poorly with threads on several platforms, so its going to be a binary choice between threaded Tcl builds and builds that support this GC. Tcl is also used in soft-real-time situations where predictability of performance is crucial, so Tcl cannot tolerate arbitrary unpredictable pauses. The allocator violates the C++ semantics since it cannot run destructors for cleanup (it knows nothing about them) so it'd become impossible to write Tcl extensions in C++, if we used it. Finally, it's a cop-out: we should get the refcounting mechanism (and in general, all memory management) to "just work", instead of abdicating this responsibility.