xref: /src/lib/libpuffs/TODO

	$NetBSD: TODO,v 1.1 2007/05/19 12:03:30 pooka Exp $

Document some possible user-visible changes that may take place.
For a complete list, please dump my brain and excavate.

  * figure out what do to with struct vattr, maybe introduce some
    vattr-like puffs-specific structure (translation costs?) instead
    of direct exposure

  * get rid of pid argument on most ops - virtually nobody (not the nfs guy)
    uses it and we should provide the lwp also.  introduce puffs_caller(pcc)?

  * make puffs_node opaque outside the library

  * make it possible to cache symlink names and getattr results in
    the kernel ... although I'm not as critically concerned with the
    kernel caching as I am with userlevel caching

  * try to implement a kernel policy for VOP_ACCESS, it's called
    very very often - but this requires some kernel caching

    + also try to figure out how to implement it in cases where it's
      more difficult to do, e.g. how can we know readdir on sshfs will
      fail without actually trying to read the directory?  If we fail
      readdir itself, it's treated as success

  * implement file system layering .. this will most likely bring
    massive changes to how the ops vector is handled, for instance.
    pcc/pu division should stay quite like the way it is now, but
    maybe there will be layer specific ops.

  * make puffs more like kernel vfs - or make kernel vfs more like puffs

  * decide what to do about setback operations.  they kind of violate
    the transparency of FAF for op handling.  also decide if the refs
    parameter in inactive is still needed

  * fix incoming requests to not require memcpy with continuations
    (not user-visible?)

  * make continuations play with libpthread, lib/36011 (not user-visible?)

  * clean up the request dispatching / continuation code (not user-visible?)