Re: [RFC][PATCH] inotify 0.9.2

[John McCutchan <ttb@tentacle.dhs.org>]

On Wed, 2004-09-22 at 21:46, Ray Lee wrote:
> On Tue, 2004-09-21 at 19:27, John McCutchan wrote:
> > On Tue, 2004-09-21 at 11:43, Chris Friesen wrote:
> > > Edgar Toernig wrote:
> > > > Robert Love wrote:
> > > > 
> > > >> struct inotify_event {
> > > >> 	int wd;
> > > >> 	int mask;
> > > >>-	char filename[256];
> > > >>+	char filename[PATH_MAX];
> > > >> };
> > > > 
> > > > 
> > > > You really want to shove >4kB per event to userspace???
> > > 
> > > Ouch.
> > > 
> > > Maybe make it variable-size?  On average it would likely be shorter.
> > > 
> > > struct inotify_event {
> > > 	int wd;
> > > 	int mask;
> > > 	short namelen;
> > > 	char filename[0];
> > > };
> > 
> > This makes reading events from inotify a pain, first you need to read up
> > to namelen, then read namelen more bytes. 
> > 
> 
> Not the case. A mildly smarter userspace program would merely read
> everything outstanding (or everything up to a fixed buffer length), and
> then unserialize the events based on the boundaries it can figure out
> from the first portion of the structure.
> 
> This is a way common technique.
> 
> At least in code I write, anyway :-).

Of course this is possible, but the inotify kernel driver only allows userspace 
program to read in event sized chunks (So that the event queue
handling is kept simple)

John