Re: [gamin] gam-poll.c rewrite

> monitors a directory and expects change events on changes
>            of files in the directory.  The SGI documentation is clear
>            that change events are only sent if the stat information
>            changes.  This is not the case if a file changes.

I've changed my position on this: fthe SGI documention says:

  FAMMonitorDirectory() monitors not only changes that happens to the
  contents of the specified directory file, but also to the files in
  the directory. If the directory contains subdirectories,
  FAMMonitorDirectory() monitors changes to the subdirectory files,
  but not the contents of those subdirectories. FAMMonitorFile()
  monitors only what happens to the specified file. Both functions
  return 0 if successful and -1 otherwise.

I've changed the patch to reflect this.  The changes were relatively
small (only in gam_poll.c and gam_subscription.c).  I've included the
patch in its entirety for your convenience.

> does a watch_file but should do a watch_directory

I've also changed by position on this due to the above as well as
nautilus' which often monitors a directory as a file e.g. in
fm-directory-view.c:load_directory.  Sending a delete event (because
the file does not exist) instead of an exist event (because the node
exists) causes nautilus to close the window thinking the directory has
been deleted.  I think that is a valid interpretation and now
understand FAMMonitorFile to mean that the FAM server should watch for
file type changes on the node while FAMMonitorDirectory means the FAM
server should watch for changes in the content of the node. now fails: we were testing that monitoring a file as a
directory returns a deleted event.

> monitors a missing file then creates a directory with the
>            same path and expects a create event.  I think this is
>            incorrect: a create event should only be sent to file
>            subscriptions

now passes

> similar problem as

passes if you add a f.flush after the f.write().

> get an extra change event on temp_dir/a.  I've traced it
>            and it seems legitimate: there is one event for a file size
>            change and one for an mtime change.  I am not entirely
>            clear what is happening, however it may have something to
>            do with the way python writes to the file.

seems tto psas

> there is a timing problem: the thread creates the
>            directories before gamin does the initial scan.  Creating
>            the thread after we get the Exist events on the directory
>            fixes this.

problem still exists
> 4.tst: same problem as
> 9.tst: same problem as

both now pass

And nautilus works for me.


Attachment: gam-poll.diff
Description: Binary data

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]