On Fri, Nov 17, 2023 at 10:13:32PM +0000, Matthew Wilcox wrote: > On Fri, Nov 17, 2023 at 02:04:37PM -0800, Darrick J. Wong wrote: > > On Fri, Nov 17, 2023 at 09:58:23PM +0000, Matthew Wilcox (Oracle) wrote: > > > It is hard to find where mapping->private_lock, mapping->private_list and > > > mapping->private_data are used, due to private_XXX being a relatively > > > common name for variables and structure members in the kernel. To fit > > > with other members of struct address_space, rename them all to have an > > > i_ prefix. Tested with an allmodconfig build. > > > > /me wonders if the prefix ought to be "as_" for address space instead of > > inode. Even though inode begat address_space, they're not the same > > anymore. > > It'd be the first thing in fs.h to ase an as_ prefix. Right now, we > have i_pages, i_mmap_writable, i_mmap, i_mmap_rwsem. We have a_ops > (which differs from f_op, i_op, s_op, dq_op, s_qcop in being plural!). > Everything else doesn't have anything close to a meaningful prefix -- > host, invalidate_lock, gfp_mask, nr_thps, nrpages, writeback_index, > flags, wb_err. > > So i_ was the most common prefix, but if we wanted to go with a different > prefix, we could go with a_. Maybe we'll rename a_ops to a_op at > some point. Um... One obvious note: there is only one point in the cycle where such renaming can be done - rc1. And it has to be announced and agreed upon in the previous cycle. IMO we need to figure out a policy for that kind of stuff; I _think_ that ->d_subdirs/->d_child conversion I have in my tree does not quite reach the threshold for that (relatively small footprint), but the thing you are suggesting is almost certainly crosses it.
