[e-lang] E and FUSE-j integration

[Mark Miller]

On 9/26/07, Toby Murray <toby.murray at comlab.ox.ac.uk> wrote:
> I take it that the introduction of deadlock dangers cannot be avoided if
> one is going to implement a synchronous interface, not matter how it's
> done. Is that right?

Not necessarily. One could allow synchronous calls only in situations
that were known to be safe from deadlock. For example, one could
assign each thread a layer number and adopt the rule that a thread may
only block on a thread with a lower layer number than itself. KeyKOS
style imposes such a layering discipline informally by convention.
IIUC, likewise for CapROS and Coyotos.

Or, in a transactional system like Argus, one could always do
synchronous calls within speculative transactions. If one finds
oneself in a deadlock (or if one seems to, because a timeout expires),
then one can abort the enclosing transaction and try again. This
trades deadlock dangers for livelock dangers.

Erlang has no primitive support for speculative transactions, but
Erlang style -- computing a turn's outcome purely functionally from
its inputs -- gives much the same effect. IIUC Erlang style for
inter-process synchronous calls is indeed to use timeouts, aborts, and
retries to avoid deadlock. I'd be curious to know whether this has led
to livelock bugs in practice.

-- 
Text by me above is hereby placed in the public domain

    Cheers,
    --MarkM