Re: [gpsd-dev] test failures with shmget

[Eric S. Raymond]

Gary E. Miller <address@hidden>:
> > So I boosted the number of NTP segments allocated to twice the max
> > number of devices, which by default equals 8.
> 
> Yes.  That is good.  However what the owner:group and file permissions 
> is still a matter of discussion within the ntpd commmunity.  So 
> nothing settled yet.
>  
> > This is a recent change and can easily be rolled back if I have
> > misunderstood thwe interface.
> 
> Nope, what you have is good, just expect it to change a bit.  Given
> how fast ntpd moves do not hold your breath.

I wouldn't be worried in any case.  The change under discussion would be
easy even for me, and I don't know that interface as well as you do.

As a separate matter, the code is not in my opinion ideally factored.
I would prefer an implementation with these properties:

(1) only the preallocated root-permission segment pointers
    live in the context structure, being put there before 
    privileges are dropped.

(2) segment pointers in use are part of the per-device structure,
    allocated ad-hoc as devices are opened

(3) the allocation logic gets a root segment if it can, and
    creates a non-root segment if it must.

But this is not for before 3.12 is out, and should probably wait until
the NTP interface re-stabilizes.
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>

signature.asc
Description: Digital signature