Breaking Changes

• Remove token/ package; all token types, constants, and Tokenize() now live in the new syntax/ package
• Module.Definitions() now returns iter.Seq[Symbol] instead of a slice
• Remove DropModules method
• Remove Module.Definitions field from internal IR (replaced by typed slices: Objects, Types, Notifications, Groups, Compliances, Capabilities)

Added

• Add syntax/ package with lossless CST, token types, and source-level query APIs
  • Parse() produces a full lossless concrete syntax tree (ModuleFile)
  • Tokenize() lexes source into token stream (moved from token/)
  • CursorContextAt() classifies syntactic context at a byte offset (module, definition, clause, imports, comment, string)
  • TokenAt() finds the CST token at a byte offset (for precise identifier extraction)
  • BuildLineTable() / LineTable.LineCol() / LineTable.Offset() for line/column mapping
  • ClauseKind enum for SYNTAX, STATUS, DESCRIPTION, INDEX, etc.
  • MacroName() returns the macro keyword for a definition node
  • SyntaxBaseTypeNames lists valid SYNTAX clause base type names
  • All CST node types exported (modules, definitions, clauses, type syntax, OID components, conformance)
  • SyntaxToken and Trivia types for whitespace/comment-preserving round-trips via ReconstructText()
• Add SymbolKind type and constants to mib/ package (Object, Type, TextualConvention, Notification, Group, NotificationGroup, Compliance, Capability, Node)
• Add SpanContext and SpanContextKind types to mib/ package (Definition, Import, OidRef, Syntax)
• Add Module.SpanContext() for offset-based span classification
• Add Module.Offset() for line/col to byte offset conversion
• Add RFC-sourced descriptions and references to all base module types and OID nodes
• Add Node.Descendants() iterator
• Add per-item spans in resolved model for INCLUDES and CREATION-REQUIRES clauses
• Add source context to load errors from fsSource and loadAllModules

Changed

• Replace direct-to-IR parser with lossless CST pipeline (parse -> lower -> validate)
• Replace hand-constructed base module IR with embedded SMI source files
• Split mib/ into internal/model and internal/resolver (public re-exports in mib/ preserved)
• Eliminate AST package and lowering pass (CST lowers directly to Module IR)
• Replace OidComponent interface hierarchy with flat struct
• Downgrade conformance diagnostics (DiagGroupMemberUnresolved, DiagGroupObjectsNotification, DiagGroupNotificationsObject) from Error to Minor severity
• Use per-clause spans for resolver diagnostics (ACCESS, DEFVAL, SYNTAX, DISPLAY-HINT, STATUS, INDEX now point at the specific clause, not the whole definition)
• Use per-member spans for group and notification diagnostics (individual member names instead of whole definition)
• Convert SNMP administrative OID nodes to OBJECT-IDENTITY with RFC descriptions
• Use windowed 2-digit year interpretation in date parsing (>= 70 maps to 19xx, < 70 maps to 20xx)
• Thread stdout/stderr through CLI struct, removing global mutation from tests
• Unify Node() and Symbol() name-collision priority logic
• Simplify import format to flat storage

Fixed

• Fix parser recovery to stop at lowercase type assignments (was skipping valid definitions during error recovery)
• Fix CLI pipe deadlock on Windows (concurrent pipe draining instead of sequential)
• Fix resolveUltimateDefiner silently returning wrong module on broken import chains
• Fix span loss during lowering for OCTET STRING and OBJECT IDENTIFIER type syntax
• Fix Notification.TrapInfo() returning internal struct instead of clone
• Fix hex/binary string error spans not including the bad suffix character
• Fix BaseSequence missing from schema export

Added

• Add inspect subcommand for deep-dive symbol inspection (type chains, import provenance, group membership, scoped diagnostics)
• Add NodesByName for multi-result name lookup
• Add Type.EffectiveTC to walk type chain for nearest textual convention
• Add IndexEntry.FixedSize for fixed-width encoding queries
• Add index suffix decoding for table instance OIDs (DecodeSuffix, IndexValue, DecodedIndex types)
• Extend Object.EffectiveIndexes to delegate to parent row on columns
• Add global fallbacks to linkObjectIndexes in permissive mode (INDEX and AUGMENTS resolution)
• Add tests for global fallback code paths and corpus test helpers

Changed

• Simplify Node.Module to return OID-phase owner directly, removing entity-cascading priority chain
• Move module list out of newResolverContext, making registerModules the single owner of ctx.modules
• Deduplicate find command search loops with findFilter struct and collectMatches helper
• Clean up minor CLI and resolver duplication (reportOpt helper, oidComponentNameAndModule extraction)
• Document intentional moduleOid drop in lowering for MODULE-COMPLIANCE and AGENT-CAPABILITIES

Fixed

• Fix scanModuleNames skipping files with comments before DEFINITIONS
• Fix dump command exit code for hard load failure (return exitError instead of exitIssue)
• Fix silent test degradation in corpus helpers (fatal on failure instead of returning empty)
• Fix exit codes for fs.Parse errors in list, paths, dump commands

Added

• Add display hint parsing and formatting API (ParseDisplayHint, FormatInteger, FormatOctets, ScaleInteger)
• Add parsed display hint accessors on Object and Type
• Add LookupInstance method and OidLookup type for combined node + instance suffix lookup
• Add File and Files source constructors for loading individual MIB files by path
• Add CI, pkg.go.dev, and Go Report Card badges to README

Changed

• Refactor FormatOID to use LookupInstance internally
• Move display hint validation from resolver into public displayhint API

Breaking Changes

• Remove v1 prefix from export schema types and functions (V1Foo -> ExportFoo, v1Foo -> exportFoo)
• Remove --all flag from CLI commands, default to all when no modules given
• Swap CLI exit codes: 1 for negative result, 2 for operational error
• Move CLI load diagnostics from stdout to stderr
• Switch CLI get output from header+indent to key-value format
• Change CLI get and find to use -m only for module selection, no positional args

Added

• Register semantic nodes (OBJECT-TYPE, NOTIFICATION-TYPE, groups, compliances, capabilities) on module in resolver, fixing qualified name resolution
• Add qualified name round-trip test across all nodes (643 nodes)
• Expand CLI find kind filter from 5 to 11 values (add node, group, compliance, capability, module-identity, object-identity)
• Expand CLI find search scope to groups, compliances, capabilities, tree
• Add --format (text, json) to CLI find and list commands
• Add --report and --no-descriptions to CLI dump command
• Add examples/modules and examples/notifications
• Add doc.go files with expanded godoc for gomib, mib, smiwrite, token packages

Changed

• Default CLI get and find to permissive+silent strictness
• Show both custom and system paths with annotations in CLI paths command
• Rename CLI list --json to --format json
• Expand README with module metadata, conformance queries, TRAP-TYPE, and CLI strictness/reporting sections
• Add severity level table and strictness note to cmd/gomib/README.md

Fixed

• Fix description nulling inconsistency in dump export (use *string for nullable descriptions)
• Fix path traversal in normalize output path
• Fix godoc link to mib.DiagnosticConfig