Expand description
Flat scans — the passes that read the tree as a directory, not as a graph.
Three of the four things resolution needs cannot themselves be found by
following links, because they are what makes following links possible. The
TitleIndex is the clearest case: nominal references ([[My File]])
resolve through it, and a nominal reference may itself be spanning
(contents: alias), so building the index by walking the tree would need the
index to walk the tree. It is a flat filesystem scan for exactly that reason
— a derived cache (DESIGN §5), rebuilt on demand and never persisted.
scan_ids and content_documents
are the same shape: enumerate what is on disk, decide nothing about it.
direct_child_files is the bounded variant the
census uses — the directories the graph already reaches, and no others, so a
vendored subtree or a nested workspace is never swept in (DESIGN §8).