Re: [PATCH update2] Support for filesystem watching (inotify)

[Rüdiger Sonderfeld]

On Sunday 05 June 2011 18:58:07 Eli Zaretskii wrote:
> > From: Rüdiger Sonderfeld <address@hidden>
> > Date: Sun, 5 Jun 2011 18:14:33 +0200
> > Cc: address@hidden
> > 
> > On Sunday 05 June 2011 17:59:03 John Yates wrote:
> > > Is there a good reason not to provide access to the full functionality
> > > of the inotify API?
> > 
> > Yes, the reason is portability. I want the same API to be available on
> > Linux, *BSD, OS X and Windows. inotify is the most advanced API and
> > therefore I can not support every inotify feature if I want to stay
> > compatible.
> 
> Being compatible does not mean providing only the least common
> denominator.  We have already several features that provide different
> levels of support depending on the underlying facilities.  One example
> is process-attributes (which is the main primitive on which Proced is
> based).  All you need is provide a superset of all the supported
> attributes, and document which ones are supported on which platform.

I could add the IN_OPEN, IN_ACCESS, IN_CLOSE_WRITE and IN_CLOSE_NOWRITE 
feature. But having separate IN_MODIFY, IN_CREATE and IN_DELETE, 
IN_DELETE_SELF would be problematic because they can't be separated for 
kqueue. Maybe I should just export the basic inotify and kqueue API and 
implement a portable layer on top of it in elisp.

Regards,
Rüdiger