imv only works on Linux about imv

[jdiez17]

Platform-specific implementations of the reload API sound good to me. Another option would be to replace `reload.{c,h}` with a library that provides the functionality on all platforms.

[eXeC64]

My preference would be to avoid adding new libraries when possible, to try and keep the requirements of imv small. Compatibility with other unices and unix-likes is definitely something I'd like to maintain support for.

[czarkoff]

I see three possible ways to provide this functionality on non-Linux platforms:

• kqueue(2) is available on BSDs (including OS X and probably Minix).
• libinotify-kqueue adds inotify support to kqueue-enabled systems.
• stat(2) is available on every Unix-like system.

My initial urge was to do exactly as @eXeC64 says: move reload.c to reload-inotify.c, add reload-kqueue.c and some magic to Makefile to pick right thing. There are dragons here, as kqueue requires keeping descriptors for all watched files around, which has potential to cause descriptor starvation. Also, kqueue won't work on some filesystems. Same issues apply to libinotify-kqueue.

I'd rather go with the following option:

• Call stat(2) and remember mtime when adding files to navigator.
• Call stat(2) again before displaying file and compare actual mtime to stored mtime.

This should not have significant overhead. In fact it may turn out to be faster then creating and destroying inotify watches (one stat(2) operation instead of one write(2) and two read(2)s), and it is more reliable in some scenarios. Eg. inotify will not report changes on NFS.

If this plan sounds sane, I'll come up with pull request.

[czarkoff]

P.S.: while testing my approach I've noticed that libinotify-kqueue doesn't seem to work for me at all...

[eXeC64]

Fixed by #68