Experiences w/ Garbage Collection and C/C++?

dberger queries: "Java has helped garbage collection enter the mainstream programmer's life - but it's certainly not new or unique to java. There have been (and are) several ways to add garbage collection to C/C++ - the most active seeming to be Hans Boehm's free libgc. I'm curious if any of the Slashdot crowd has used this (or any other) C++ garbage collector in non-trivial commercial applications. If so - what were your experiences? If not, why not? (Before you ask, yes - I know that GC isn't the only difference between C++ and Java, but 'automagic memory management' is certainly part of Java's marketing luster)"

Comments

Re:There's another way.

[swillden]

You repeat some common myths about GC; allow me to counter them.

The obvious: CPU & memory overhead for the checking and tracking. I can't comment on the amount here, but it is a generalized solution, so you forego the optimization opportunities that you'd otherwise have.

Malloc/free and new/delete (without pooling) are also generalized solutions, and they also consume CPU and memory overhead for checking and tracking. There is good reason to believe that in the right type of language (which C and C++ are not) that GC can actually be much more efficient than manual deallocation, mainly because it can do its work in larger batches, and because it can reorganize objects in memory to make allocating more efficient. Contrast a simple single-heap malloc implementation, which has to scan a free list looking for a sufficiently large block against a copying garbage-collected system where the allocation pool is simply a large contiguous block from which you just grab the first 'n' bytes.

If you look on Boehm's web site, you can find a few papers comparing the performance of conservative GC for C with optimized malloc/free implementations. malloc/free wins, but not by as much as you'd expect.

The subtle: Memory allocation can become a major bottleneck in multithreaded systems. Garbage collection has similar issues.

Actually, GC *eases* the issues associated with recovering memory in multithreaded systems. Why? In a multi-threaded program with manual deallocation, both allocation and deallocation occur in every thread context. In a GC system, all deallocation is typically concentrated in a single thread, the GC thread. Allocation is still spread across threads but the required interlocking is hugely reduced since the GC thread can do all of the reclaiming, block coalescing and free list construction (if that's the technique used) without any interference from the other threads. It will have to acquire a mutex to place the recovered blocks back where the active threads can get them, of course.

Generational, copying GCs can do even better, but not for C or C++.

The irritating: you don't know when your destructors are called.

As experience with finalize methods in Java has shown, you should really treat GC as a way of having infinite memory. The problem with finalizers/destructors is that not only do you not know when they'll be called, you have no way of knowing that they'll *ever* be called. That means that they're effectively useless and add significant complexity and overhead for little or no return.

IMO, if you want to use C++ with GC, you should make sure that objects that have non-trivial destructors (those that do something besides memory management) get destructed normally, and just let GC handle the memory.

Reference Counting Smart Pointer (RCSP for short)

They're useful, but I'd hardly call them great. Reference counting is *far* more compute-intensive than scanning-type garbage collection. And then there's the problem of circular references, which will never be reclaimed. Of course, it's not that hard to avoid those situations most of the time, but with GC you don't have to care.

Owning Smart Pointer (OSP) ... auto_ptr

These are very useful, and conscientious use of them will eliminate 95% of memory leaks and dangling pointers. OTOH, they don't work when things get complex enough that ownership isn't simple and clear.

Also, you can optimize your smart pointers for individual types (through template specialization). A great example is to give the no-longer-needed object back to a pool for later reuse.

Generally, I would do this through specialized new/delete, rather than specialized smart pointers. Regardless of the mechanism, though, pooled allocation is the absolute best thing you can do to minimize the cost of memory management in your application. The reason is, of course, that you build the pooling based on your knowledge of the actual usage characteristics of the objects; knowledge that no general-purpose memory manager can possibly have.