-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Wednesday 10 June 2009, Christian Schladetsch wrote:
monotonic::storage<100*sizeof(int)> storage; //< on the stack! not on the heap, as others have suggested monotonic::vector<int> v(storage); for (int n = 0; n < 100; ++n) v.push_back(n); // will fail at n = 34 with an exhausted storage
but this succeeds: monotonic::storage<100*sizeof(int)> storage; monotonic::vector<int> v(100, storage); // create with an initial size and storage for (int n = 0; n < 100; ++n) v[n] = n;
As the vector in the first case expands, it resizes. But the old memory is not freed - that is why it is explicitly called a "monotonic allocator".
Have you considered also supporting a slightly modified allocator that falls back on the heap once the initial storage is exhausted? I'm more interested in a small buffer optimization while still supporting larger buffers, rather than restricting usage to a fixed storage size. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.9 (GNU/Linux) iEYEARECAAYFAkoxAMEACgkQ5vihyNWuA4XcjQCgihXSDr6ZpAOqeq9wZVxLGXV3 hpcAniShoS7yU11EtR5pFO8CYv2ZF/nL =DNgA -----END PGP SIGNATURE-----