Re: tail --recursive

[Jim Meyering]

Matthew Woehlke <address@hidden> wrote:
> Simon Windows wrote:
>> Jim Meyering wrote:
>>> I haven't looked at it at all yet,
>>> but wonder if you've considered using inotify?
>>> That seems like that mechanism would be much more appropriate,
>>> when tailing many files in an arbitrary hierarchy.
>>
>> I have looked at the inotify library as you suggested. I agree it would
>> be a useful tool to use in tail as it removes the need to poll files and
>> would make a recursive version much more efficient. I can also see there
>> has been some discussion of using inotify in coreutils for this purpose.
>>
>> I think that issue is orthogonal to the recursive option I am proposing.

It's not orthogonal.
The implementation of any new option like the one you've proposed
must be robust, maintainable, and reasonably efficient.  Portability
to non-Linux would be nice, but isn't an immediate priority.
Doing all of that is not possible with a conventional directory
traversal.  I suppose that's at least part of why inotify came about.

> To point out the obvious... 'tail' works on most POSIX systems (and
> perhaps some that aren't). inotify works on... Linux. So an
> enhancement to tail might benefit users of AIX, Irix, OSF, etc. that
> don't have inotify, whereas iwatch is useless on such systems.

I am very reluctant to add any enhancement to tail
that makes it perform a directory traversal.

IMHO, the proper solution is to use inotify in a new tool, or _maybe_
in tail.  If the right people decide they want that functionality on
non-Linux systems, maybe they'll write a gnulib module to provide some
minimal approximation.