[coyotos-dev] Endpoints and replies

[Charles Landau]

At 12:59 PM -0500 1/17/06, Jonathan S. Shapiro wrote:
>On Tue, 2006-01-17 at 09:52 -0800, Charles Landau wrote:
>
>>  There is a design option that eliminates case 2. The openness of the
>>  wait could be a property of the endpoint. Most commonly, servers will
>>  wait on open endpoints and callers will wait on closed endpoints. You
>>  couldn't use the same endpoint for both purposes, but I don't see a
>  > need for that.
>
>If the endpoint type is specialized in this way, it seems likely that
>two checks need to be done in the invocation path: one on the endpoint
>type and the other on the application-provided option bits for the
>invocation.

More precisely, the option bits provided by each of the threads that 
may be waiting.

>If a single endpoint type can semantically handle all cases,
>and performs reasonably for the sensible cases, it seems preferable to
>do one check rather than two.

You would be doing N checks rather than N+1, where N is the number of 
threads waiting, typically one.