I wrote a test program which allocates objects and registers finalizers for them. Only a few (or no) objects are finalized. What's wrong?
Probably nothing. Finalizers are only executed if all of the following happen before the process exits:
Does this mean that the collector might leak memory?
In the short term yes. But it is unlikely, though not impossible, that this will result in a leak that grows over time. Under normal circumstances, short term, or one time leaks are a minor issue. Memory leaks in explicitly managed programs are feared because they almost always continue to grow over time.
For (a lot) more details see:
``Bounding Space Usage of Conservative Garbage Collectors'', Proceedings of the 2002 ACM SIGPLAN-SIGACT Symposium on Principles of Programming Languages, Jan. 2002, pp. 93-100. Official version. Technical report version.
How can I get more of the finalizers to run to convince myself that the GC is working?
Invoke GC_gcollect a couple of times just before process exit.
I want to ensure that all my objects are finalized and reclaimed before process exit. How can I do that?
You can't, and you don't really want that. This would require finalizing reachable objects. Finalizers run later would have to be able to handle this, and would have to be able to run with randomly broken libraries, because the objects they rely on where previously finalized. In most environments, you would also be replacing the operating systems mechanism for very efficiently reclaiming process memory at process exit with a significantly slower mechanism.
You do sometimes want to ensure that certain particular resources are explicitly reclaimed before process exit, whether or not they become unreachable. Programming techniques for ensuring this are discussed in
``Destructors, Finalizers, and Synchronization'', Proceeedings of the 2003 ACM SIGPLAN-SIGACT Symposium on Principles of Programming Languages, Jan. 2003, pp. 262-272. Official version. Technical report version. HTML slides. PDF slides.
I wrote a memory allocation loop, and it runs much slower with the garbage collector than when I use malloc/free memory management. Why?
Odds are your loop allocates very large objects and never initializes them. Real programs generally don't behave that way. Garbage collectors generally perform appreciably worse for large object allocations, and they generally initialize objects, even if you don't.
What can I do to maximize allocation performance?
Here are some hints:
If my heap uses 2 GB on a 32-bit machine, won't every other integer or other random data be misinterpreted as a pointer by the collector? Thus won't way too much memory be retained?
Maybe. Probably, if the collector is used purely conservatively, with no pointer layout information (such as use of GC_MALLOC_ATOMIC).
With a gigabyte heap, you are clearly much better off on a 64-bit machine. Empirical evidence seems to suggest that some such applications work on a 32-bit machine, and others don't perform acceptably.
Simple probability calculations for pointer misidentifications are generally incorrect. The probability of misinterpreting an integer is typically reduced significantly by a number of collector features and fortunate accidents. Most integers are small, and small integers can generally not be heap addresses. The collector black-listing mechanism avoids allocating areas that are prone to be targets of misinterpreted references. The collector can be told to ignore some or all pointers to object interiors. I have a different question that isn't answered here, nor in the other GC documentation. Where else can I go?
You may want to check the the GC mailing list archives. If you can't find the answer there, please post your question to the GC mailing list. If you are not subscribed to the mailing lists, your posting may take a while to appear, since such postings are moderated.
Why can't I get to the mailing list administrative web site to subscribe?
It's behind the HP firewall. (This is unintentional, but not easily repaired.) Send mail to gc-request or gc-announce-request at linux.hpl.hp.com instead.