When analyzing the overhead of specialization, sometimes you want
to distinguish the first run (or first run with concrete signature)
from later runs. Consequently it can be useful to extract the
flattened times in the order in which inference happened.
This also switches to a depth-first algorithm for searching the graph,
as required to support sorted=false.
When analyzing the overhead of specialization, sometimes you want to distinguish the first run (or first run with concrete signature) from later runs. Consequently it can be useful to extract the flattened times in the order in which inference happened.
This also switches to a depth-first algorithm for searching the graph, as required to support
sorted=false.