A few years ago I had an issue with a few of the larger carriers rejecting my routes (from a natural Class B space) because their prefix length was too short (at one point I simply had the /16 divided into two /17's and this still got rejected in some places). I can't remember which carriers exactly, but it may have been some larger transit providers like AboveNet/etc.
Anyone know what the current attitude is by carriers about this? Nowadays with ever-growing memory and CPU it doesn't seem like it's as much of an issue. In an environment where we're all trying to conserve address space watching natural boundries doesn't seem all that smart.
It is rare that providers filter on classful boundaries. What is common is filtering on RIR allocation boundaries. It just happens that in 128/2 nothing longer than /16 has ever been allocated, ttbomk. Providers should encourage their customers to always originate their largest aggregate, and _then_ announce (if necessary) any more-specifics to those they need to and who agree to accept them. If networks always originated their largest aggregates there wouldn't be an issue with filtering out long prefixes. The issue is only when a network announces only the long prefix, and in effect shoots themselves in the foot by intentionally limitting their own reachability.