On 04/18/2013 06:10 PM, Pádraig Brady wrote: > On 04/17/2013 06:39 AM, Bernhard Voelker wrote: >> Somehow, I don't have a good feeling for using inotify in the >> --follow=descriptor case because inotify works with file names. >> >> Once tail opens the file successfully after inotify says it's >> there, tail would have to go back to polling, because all further >> inotify events are only relevant for the file name, not the >> descriptor. >> >From playing with the inotify code, I even had the impression that >> there are not all events delivered to tail reliably. Is this true? > > I've not noticed such issues TBH. > What kernel we you using for reference?
That was when I ran tail via GDB. For the deletion of a file, there usually seem to be 2 events generated (IN_ATTRIB and IN_DELETE_SELF), but while stepping through the code handling the first event, the latter never happened then. That's on a openSUSE-12.1 with: $ uname -r 3.1.10-1.19-desktop >> Additionally to the above, I noticed that tail now would no longer >> exit in the case the file appears untailable, e.g. when it appears >> as directory. I fixed this, too. > > Excellent thanks. > > You may want to consider using the existing retry_delay_() > rather than wait4lines() in the tests, but otherwise it > looks good to push. Thanks for the review. I'll have a look at retry_delay_() again tomorrow, and see if it makes waiting for tail's output easier. Have a nice day, Berny
