v0.6.39

v0.6.39 — graph ids never reused; the PGXN archive builds on Linux and macOS; pgrx 0.19.3

Fixed

• Graph ids are never reused (#150). add_graph(iri) allocated MAX(graph_id) + 1, so dropping the highest-numbered graph gave its id to the next graph created, and a numeric id held across a drop and a create silently named a different graph. Ids now come from a sequence (registered for pg_dump) and are never handed out again; explicitly bound ids advance it too. The IRI is the graph's identity, the id a handle; gaps are normal. Existing ids are unchanged. A library newer than the installed SQL refuses 55000 with HINT ALTER EXTENSION pgrdf UPDATE.
• The PGXN source archive builds again (#149). Since 0.6.35 it could not compile on any platform: shacl_capability() reads two data files at compile time that the archive did not ship. It now carries them and the macOS link setting, and make install takes its paths and library suffix from pg_config and the platform, so it installs on Linux (Debian and RPM layouts) and macOS (Homebrew). CI now builds the archive, installs it from a clean unpack and loads the extension on every run. Do not use the archives attached to 0.6.35–0.6.38.

Changed

• pgrx 0.19.3 (#148): interrupts held while non-ERROR reports are raised, safer varlena handling, PostgreSQL 19 beta4 support. MSRV unchanged (1.96).

Upgrade

• ALTER EXTENSION pgrdf UPDATE in place; creates the id sequence above every id in use. No stable export added or removed; surface() lists 53 stable rows. SHACL capability (47/47 enforced) and path matrix (17 cells, 0 defective) re-measured, unchanged.

What's Changed

• 0.6.39: graph ids never reused; PGXN archive builds on Linux and macOS; pgrx 0.19.3 by @styk-tv in #151

Full Changelog: v0.6.38...v0.6.39