While we generally don’t publish literal logs of our IRC discussions, in some case a discussion was found to be so instructive, that we save a literal transcript and provide an (slightly condensed) excerpt for later referral.
How to control and to what extent to pre-calculate the allocation of processing buffers.
These buffers are used by the processing node to be invoked, to get the “input data”
and to store the “output data”. This topic might get us into some complications,
since some nodes take multiple input/output buffers, while other nodes require to
calculate “in-place”
→IRC June 2008
We need a way of identifying a frame within the cache; this frame might hold an
intermediary result; so the access key needs to include the position in the graph,
as well as the output process and some generation number. This topic was touched
in one of our earliest meetings
→summary
Prompted by a RfC regarding the engine service, this discussion in May 2011 covered
various aspects of architecture and implementation regarding the engine and especially
scheduler and render jobs; aborting and superseding of jobs, questions of resource handling,
waitstates, dedicated time bound operations.
→IRC transcript from 5/2011 dev meeting
This discussion in September 2013 touched various requirements for the scheduler
and the ramifications for a possible implementation. More specifically, from the
design and implementation for the playback and rendering subsystem done meanwhile
we can derive some specific requirements to be imposed on the scheduler from the
client’s point of view.
→IRC transcript from 9/2013 dev meeting