[GRASS-dev] Re: mkdir function

[Glynn Clements]

Hamish wrote:

> > > >  lib/raster/io_fifo.c
> > > 
> > > Can't we just kill this? I don't even know if it still works, given
> > > that almost no-one actually uses --enable-fifo.
> > > 
> > > In any case, I doubt that file will work on Windows.
> > 
> > If we don't need it, let's remove it. Unused code is only
> > confusing.
> 
> just for curiosity, what's the (dis)advantage of using FIFO vs sockets?

Ugly code, not particularly robust, needs a global mode 1777
(world-writable, sticky-bit) directory for lock files (and thus the
need to deal with stale lock files), need to pre-create the FIFOs.

Also the need to store their paths in etc/monitorcap, which means
modifying it if you move the GRASS installation:

http://grass.itc.it/pipermail/grass-dev/2003-February/007083.html

[Currently, we store relative paths, which don't actually work.]

The "obvious" mechanism for client-server communication is a socket. 
The FIFO code existed for systems which don't have sockets. Nowadays,
such systems are about as common as 8-inch floppies or punched cards.

I only know of one person (Bev Wallace) who was actually using the
FIFO option, and that appears to be no longer relevant:

http://grass.itc.it/pipermail/grass-dev/2003-February/007095.html

-- 
Glynn Clements <glynn at gclements.plus.com>