>The obvious solution here is to add a tail pointer to the socket buffer
>and use it when you can.

Jason,

Wow, thanks. Please just do it. I recall raising the same issue some 2
years ago (after profiling DU/Tru64 kernels); i thought this got
addressed then.

Skimming the code: sbappend_simple() is just syntactic sugar for
sbcompress(). Why not just call sbcompress() directly (or via an
inline/macro)? Or at least call it sbcompresstail() or sbappendtail()
-- which is what it does, after all?


>I have done this, and added an sbappend_simple() that can be used for
>the simple case of stream data (e.g. TCP).  I have not yet tried to
>use the tail pointer in the pre-existing sbappend*() functions, since
>I need to get my head wrapped around the intracacies of how each one
>works (they're full of special cases for end-of-record and whatnot,
>and I'd prefer not to break them because I missed something subtle :-)


Is it legit for someone who still has access to Tru64 source to
double-check any M_EOR-related changes after the fact?