Icod.TermInfo.BerkeleyDb 1.15.0

Icod.TermInfo.BerkeleyDb

Icod.TermInfo.BerkeleyDb is the optional managed package for read-only acquisition from ncurses-compatible Berkeley DB hashed terminfo stores.

1.15 release closure status

1.15.0 is the stable coordinated release. The complete public surface is frozen at nine exported types, dependency direction remains Runtime-only, and the deterministic sample plus acquisition, compatibility, security/resource, ecosystem, and release authorities are complete. The stable promotion changed no acquisition behavior or public API.

HDB00 selected a dependency-free managed reader for the reviewed Berkeley DB Hash on-disk format version 9 subset required by ncurses acquisition. Native Berkeley DB remains a CI interoperability oracle and is not a production dependency.

The internal ncurses resolver follows bounded marker-2 index chains over one acquired database image and extracts opaque marker-0 payloads. Empty records, unsupported markers, dangling targets, cycles, and excessive hops fail explicitly. Only an absent initial key is a clean miss.

BerkeleyDbTerminalDescriptionProvider reads one caller-selected database path. Lookup tries the exact ordinal UTF-8 key first and, only after a clean miss, one distinct exact Latin-1 key when every requested character is representable. A found malformed or identity-invalid UTF-8 record never falls through. BerkeleyDbTerminalDescriptionProviderOptions snapshots parser, database-size, and index-hop limits. BerkeleyDbDatabaseFormatException identifies malformed or unsupported containers and ncurses record envelopes while parser and I/O failures retain their existing exception types.

Hashed terminal catalog

BerkeleyDbTerminalCatalogReader reads one explicit database path and returns a fresh immutable snapshot on every call. Its options snapshot parser, database-size, record-count, and index-hop limits.

The catalog emits only marker-2 logical publications. It classifies each publication as Canonical or Alias by decoding a valid UTF-8 key first and using Latin-1 only when the raw key is not valid UTF-8. Distinct raw keys that decode to the same ordinal logical name are rejected. Entries resolving to the same marker-0 record share one parsed terminal instance within the snapshot. Marker-0 storage records, including unreferenced orphans, are still parsed and validated but their internal keys are not emitted.

Raw records are validated in unsigned byte-key order. Successful entries are ordered by ordinal publication name, then kind, then canonical terminal name. Physical Hash-page order is never exposed. Reads are uncached, bounded, cancellable, independent, and release the database handle before parsing. Production path reads compare two complete observations through one open handle and reject unequal content or length. This detects mutation between observations but does not provide an atomic snapshot or arbitrary writer coordination.

Malformed Berkeley DB structures and ncurses envelopes use BerkeleyDbDatabaseFormatException. Malformed compiled entries retain CompiledTermInfoFormatException; identity mismatches, I/O failures, and cancellation retain their established exception types.

Package boundary

The package:

• targets net8.0, net9.0, and net10.0;
• depends only on the matching Icod.TermInfo version;
• has no native Berkeley DB dependency;
• does not bundle Berkeley DB binaries;
• does not implement writes, transactions, environments, recovery, or general-purpose Berkeley DB APIs; and
• keeps compiled terminfo semantics in Icod.TermInfo Runtime.

The reviewed subset supports unencrypted, non-checksummed Hash-v9 files with sorted Hash pages (type 13), inline items, and overflow items. Other access methods, revisions, duplicate/subdatabase features, and legacy type-2 Hash pages are unsupported. Validation covers encountered records and does not guarantee an atomic snapshot during external writes.

Dedicated CI compares the production reader, providers, catalog, and tools with native Berkeley DB stores on Linux and macOS; Windows reads Linux-generated fixtures without Berkeley DB installed. Native big-endian Hash-v9 containers are produced by reloading byte-exact ncurses records with big-endian metadata. Native tic also produces exact Latin-1 canonical and alias keys. Lookup is UTF-8-first with the bounded clean-miss-only Latin-1 fallback above, while compiled identity fields retain Runtime's byte-preserving Latin-1 interpretation. Universal non-ASCII encodings, normalization, transliteration, and best-fit mapping are not claimed.

HDB08 qualification requires an isolated package-only consumer on net8.0, net9.0, and net10.0 on Windows, Linux, and macOS. A managed verifier enforces the exact package identity, dependency, assembly, symbol, Source Link, and no-native-asset contract.

Hashed-aware system discovery

BerkeleyDbSystemTerminalDescriptionProvider is the opt-in HDB04 system provider. It snapshots Runtime's existing discovery inputs and policy at construction while leaving Runtime's frozen SystemTerminalDescriptionProvider unchanged. Discovery preserves encoded TERMINFO, TERMINFO, user, TERMINFO_DIRS, and platform-default precedence, including empty default components and equivalent-location deduplication.

For each logical location, an existing exact directory or exact hashed file wins. When the exact location is absent, a regular .db companion is considered, matching ncurses discovery behavior. Missing sources continue; malformed reached sources fail explicitly. Successful results are cached, while misses and failures remain retryable.

HDB07C qualification passed 422 BerkeleyDb unit cases and 50 native-store interoperability cases per target framework on Windows, Linux, and macOS. In addition to HDB07's native 64-entry ASCII matrix, Linux and macOS independently verified a big-endian Hash-v9 container with byte-exact ncurses records and an exact Latin-1 canonical/alias fixture; Windows consumed the Linux fixtures without Berkeley DB installed. Real permission denial and restoration also passed on all three hosts. The optional package's public API remains unchanged: command executables classify explicit paths and then call the accepted provider or catalog reader. Explicit infocmp -A/-B files and explicit human toe file operands are supported; Inspection stays provider-neutral, ambient toe discovery and JSON stay conventional, and tic remains directory-write-only. Installed-tool smoke passed on all three hosts and direct-command smoke passed for all six archive RIDs. The accepted Alpha-8 feature/API source and its HDB09 evidence-record head passed the normal 12-job and genuine HDB00 3-job workflows, including the managed exact-package verifier and three-host net8/net9/net10 package consumption. Atomic snapshots, arbitrary writer coordination, and non-UTF-8/non-Latin-1 producer encodings remain outside the qualified contract. Stable 1.15.0 preserves that exact behavior and surface.