> "François Revol" <revol@xxxxxxx> wrote: > > > BFS could use this internally, and although the original one > > > doesn't, > > > my implementation actually does. It's not really needed though, > > > so > > > we > > > could think about removing it. > > Doesn't tell me what it does... > > Ah, okay. I should have said if the directory/container in question > supports duplicate entries or not. For indices this is generally > true, > for other types, it's not :) > > > > > Also I'd reconsider using S_FOO_INDEX in favor of S_INDEX, > > > > since > > > > it's not very scalable... we have only 2 bits left in that > > > > field. > > > I don't understand what you're trying to say here. > > Are the different bits really needed or could we use just one to > > say > > S_INDEX without type distinction ? > > They are actually used by BFS to maintain the index type. Although it > also duplicates this information in the B+tree header, we now also > use > it to report the types when calling stat() on an index. > Or what about using the bits as value instead of bitmask ? would give more room, S_IFMT does it (see LNK, and SOCK... wait, where are my sockets gone ? I need sockets for santa to put things in !!!!!) /me goes add S_IFSOCK back (Yes we do can use that type for sockets even in R5. It's just about not using a device driver which publishes only char devices. sockfs works great here with BONE, I'll port it to our stack) François.