James Clark
James Clark
I don't think the translation should panic. It should do the best approximation it can and give a warning when approximating. We should aim to produce a readable CDuce type.
I agree we should put this in extra
> Graphs are acyclic along forward edges and along backward edges What does this mean? It's the graph formed by the forward edges that is acyclic.
> All blocks in the function code can be reached when navigating the graph Reached from where? Must be reachable via forward edges only.
> Backward branches will only be allowed in a region that is a loop Point is that backward branch must point to a block that is the entry of a...
> Region records are created for each single-entry single-exit region in the function code Really? I don't think so. How are you going to find *every* single-entry/single-exit region.
@ushirask Does it mean that every conditional branch is the entry point of a single-entry, single-exit region? Or does it mean every place that is a join has to be...
> We had a chat yesterday and concluded, yes, every conditional should be a region entry Sounds plausible. > but exit should only appear when all the paths that diverged...
That looks reasonable, but I don't think we need separate files.
In particular handle #1022.