On Sun, Nov 12, 2006 at 08:31:42PM -0700, Paul E Condon wrote: > On Sun, Nov 12, 2006 at 09:55:52AM -0800, John L Fjellstad wrote: > > Paul E Condon <[EMAIL PROTECTED]> writes: > > > > > Thanks for stack profiling info. I wonder if this _is_ a reportable > > > bug. After all, there is a lot of information on the 'bad_alloc' > > > exception in various sources. If GNU C++ library doesn't try to throw > > > this exception until it is too late for the throw to succeed, what > > > good is it? And, how was its operation verified during testing on a > > > Debain box? At the point where it kill happens, I think the program is > > > allocating space for the contents of large STL vector of vectors. It > > > doesn't seem to me that this should be done on the stack, but I > > > haven't thought deeply on the issue. > > All my speculation as to the quality of GNU C++ compiler is > inoperative. The problem is in the kernel, as pointed out by John F.: > > > > > One thing you might not be aware of is that Linux overcommit the > > memory. That is, even if there is no memory left, it might tell the > > program that it has enough, since it expect that by the time you use the > > memory, it will be available. It's controlled by > > /proc/sys/vm/overcommit_memory. > > > > Check the man page for malloc (look under BUGS), and do some search in > > your favorite search engine. > > > > This is really interesting. man malloc calls the default behavior > of the kernel a really bad bug, with which I agree. kernel-docs > sysctl/vm.txt justifies the default by claiming it is helpful to > programmers who 'malloc() huge amounts of memory "just-in-case" and > don't use much of it.' I had thought such programmers were rare in the > Linux world, as they are all happily writing garbage code for > BillG. But, oh well. Now I know how to make my code work under Linux.
It is reasonable to reserve large blocks of address space, if you need unknown amounts of contiguous space -- and then populate it later. This happens, for example, with some memory allocation algorithms. FOr specific applications, such techniques may be more efficient or otherwise appropriate than using malloc. But I do believe that if such a feature is provided, it should *not* be called malloc, which has other important uses. In any case, it sould be possible to ask for storage allocation in such a way that the allocation request fails if the storage is not going to be available when it is needed, -- hendrik -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
