nx_varcensus_lib.nx

nx_varcensus_lib.nx -- the VaM corpus census's DATA-DRIVEN pieces, importable by the CLI (nx_varcensus) and its gate. (1) a growable byte ARENA -- tables store OFFSETS into it, so growth (a bigger mapping plus a copy) never leaves a dangling pointer; an i64 vector rides the same arena in 8-byte cells. (2) a growable string COUNTER -- FNV-1a, open addressing, the slot table doubled at half load; entries are kept in FIRST-SEEN ORDER so a report reads in the order the corpus was walked (and the walk itself is sorted). (3) the RULES conf (the classifier is DATA, never code). Rows, TAB separated, '#' comments: role|format <TAB> <class> <TAB> <kind> <TAB> <pattern> [<TAB> <kind> <TAB> <pattern>] both conditions must hold screen <TAB> <reason letter A-Z> <TAB> class <TAB> <role class> a member of that class sets the reason screen <TAB> <reason letter A-Z> <TAB> namecontains <TAB> <token> the package id contains the token deptag <TAB> <tag> <TAB> prefix <TAB> <dependency id prefix> a direct dependency with that prefix kinds: dirent (name ends with a slash) | exact | prefix | suffix | contains. Matching is on LOWERCASED names and lowercased patterns. Within one axis the FIRST matching rule wins, and a member no rule matches lands in the implicit class `other` of that axis -- so each axis is a PARTITION of the members by construction, and the census asserts the sum. A row this parser does not understand is an ERROR counted with its line number: a malformed rules file cannot produce a census (the CLI refuses), because a silently skipped rule moves members into `other` with no trace. license_tier: ORIGINAL No hw writes (Rule 26).