[Unison-hackers] fsmonitor.py & ignore paths on Linux

[Jay Levitt]

I'm trying out the most excellent -repeat watch functionality in SVN, 
syncing between a Mac and a Linux machine; the directory has about 
260,000 files in it, but I'm ignoring all but 5,000.  The Mac detects 
changes instantly, but the Linux side has a few problems:

- At startup, it takes about five minutes of 100% CPU crunching to 
populate the watcher paths - and this is five minutes on a machine with 
a hybrid-SSD drive! It then takes another 10-15 minutes of seeming idle 
before it starts working normally.

It looks like it's adding every single file as its own inotify_watch.  
If I understand inotify correctly, you don't need to do that - all you 
need to do is watch each directory in the tree, and any contained files 
will also be notify.  I'm guessing that doing a stat on each file (to 
see if it's a directory) would be more efficient than adding each watcher.

- That's made significantly worse by the fact that it doesn't honor any 
ignore paths in the unison config; this probably would be a minor 
annoyance if it only had to iterate over 5K of files instead of 250K.

- Now that I think about it, fsmonitor.py on Linux (server) is assuming 
that its ~/.unison directory contains the same profile as the Mac 
(client).  In my case, it doesn't, only because I hadn't thought to copy 
it over.

- But even worse: unison obeys fsmonitor's instruction and syncs up 
changed files - even if they were previously ignored.  So, f'rinstance, 
the server's .unison archive files are getting synced to the client.

Do you need any further info to reproduce these? I know the linux code 
is especially untested.  I'd offer a patch, but I don't know Python, so 
that'd be more threat than offer.

Jay Levitt