From: Eric V. H. <er...@gm...> - 2004-12-12 16:42:52
|
I'm going to need some time to look over things in detail (unfortunately time is one thing I don't have a lot of these days). I've attached a few comments based on your description without reviewing the code. On Sun, 12 Dec 2004 10:27:46 -0500, Latchesar Ionkov <lu...@io...> wrote: > I would like to describe the changes in 'lucho-1' in the 9p and conv layers. > > I. 9P Layer > > 1. fcall_pkt is removed. There is a local struct in conv.c that is used for > similar functionality. > I'm good with removing fcall_pkt, but I'm not so sure on using the local structure. One thing I wanted to move towards was removing the copy for reads and writes from the fcall_pkt and fcall. My plan was to collapse the fcall_pkt structure into the fcall - then serialize/deserialize could work on the header but the data pointer could just point to the data without having to copy it. Not sure if your approach prevents this, but its something to consider. > 2. I got rid of type specific structs (Tversion, Rversion, ...). IMHO having > all possible fields in a single structure is easier. This is a preference thing, I could care one way or another, but I think Ron liked having stuff in the union. > > 5. All t_* functions receive v9fs_fcall** as parameter, don't have any other > output parameters. > In principal I don't see a problem with this - I don't think it'll interfere with my zero copy plans. However, I think its strange that neither Russ nor Ron went down this route, so I'm wondering if there's something we are missing. -eric |