Re: [gpsd-dev] What is our current bug state?

[Eric S. Raymond]

Hal Murray <address@hidden>:
> 
> address@hidden said:
> > My current belief is that solving that problem will require gpsd to go to
> > using blocking I/O rather than nonblocking, because setting VTIME is only
> > effective when blocking I/O is enabled.  The trouble is that I don't know
> > how to make that change in a way that won't cause device opens to hang hard.
> 
> The context is opening things like /dev/tty.  Right?
> 
> I had similar problems ages ago.  I don't remember the exact context.
> 
> My solution was to use O_NONBLOCK to open the tty, then turn on CLOCAL, then 
> close and reopen without O_NONBLOCK.
> 
> That may be a wild goose chase, YMMV, etc.  That was on Linux.  I don't know 
> how portable it is.

Interesting.  I tried just removing O_NOBLOCK from gpsd's tty opens, and
it turns out that both gpsd and gpsmon work fine without that flag.

I thought they were more broken than they are because the first thing I tried
was the regression tests - and something inside gpsfake hangs.  My next step
will be to groval through an strace dump trying to figure out why. 
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>