Occasional seg fault in make check at dispatch.cc

[Michael D. Godfrey]

>
> but I don't see why the newly created files are not found immediately.
>   

What I know from having written file locking code for an application
a while ago is that the Linux (Unix) filesystem can behave in quite
mysterious ways (even without NFS, which makes it much worse).
There are various ways in which the filesystem can get into transiently
(or, of course, permanently) inconsistent states.  It might be worth
trying code after the filename create that tests on filename existence
and waits for some number of iterations and then raises an error if
the name does not show up.

You likely know that one of the major nightmares in Linux is applications
which  support controlled file sharing for multiple users.  Everyone has
trouble with this and many different "solutions" end up being used.  I
just recently got to debug a failure of file locking in the Cadence software
system which we use for VLSI design.  They, of course, used different
locking techniques in different parts of their system.  Not a happy story,
but it was finally possible to fix it, at least for now.

Which causes me to ask: has anyone ever asked for a "protected file
sharing" capability for Octave?  Or, did we discuss this once quite a
long time ago?  In any case, if this gets thought about it needs very
careful thought if it is going to end up working reliably.

Michael