The optimization(s) that prevents IOP::ServiceContexts from being added to the 
outgoing request IOP::ServiceContextList upon a connection retry (e.g. 
LOCATION_FORWARD or a transport retry) also prevents the 
PortableInterceptor::ClientRequestInfo::add_request_service_context() method 
from working properly.

That method may potentially attempt to add a IOP::ServiceContext during each 
connection retry.  However, the optimization in question prevents the 
IOP::ServiceContextList from being reset.  As such, 
add_request_service_context) may fail due to the duplicate already in the 
service context list.

In my opinion, we should drop this optimization.  Besides, how often do 
LOCATION_FORWARDs or transport retries occur compared to successful 
invocations?  I think that we can simply set the IOP::ServiceContextList length 
to zero.  That way, we can reset the contents of the request 
IOP::ServiceContextList without causing additional allocations due to service 
context list growth since the underlying buffer will not be deallocated when 
reducing the length of the list.