v0.6.0

What's Changed

• Map Spanner gRPC status codes to specific ADBC statuses by @fornwall in #20
• Auto-detect credential type from keyfile JSON type field by @fornwall in #23
• Rename vendor option prefix adbc.spanner.* to spanner.* by @fornwall in #24
• Accept all Arrow timestamp units in bind, preserving nanosecond precision by @fornwall in #22
• Scaffold the ADBC Driver Foundry validation harness (non-gating) by @fornwall in #25
• Read TIMESTAMP at nanosecond precision (matches the bind/write path) by @fornwall in #27
• Add PyPI/build badges to the Python package README by @fornwall in #26
• Tag JSON columns with the arrow.json extension type by @fornwall in #29
• Bind parameters positionally; make the Foundry type/bind suite pass by @fornwall in #28
• Add service-account impersonation (spanner.impersonate.* options) by @fornwall in #30
• Port the Foundry type/select suite to Spanner (dialect overrides + skips) by @fornwall in #31
• Add Polars/pandas dataframe usability tests against the emulator by @fornwall in #32
• Implement create-mode bulk ingest; port the Foundry ingest suite by @fornwall in #33
• Add differential (oracle) test for the Spanner->Arrow type mapping by @fornwall in #34
• Pin Linux wheel builds to glibc 2.35 and enforce the manylinux tag by @fornwall in #35
• Make cancellation sticky so a cancel between stream chunks is not lost by @fornwall in #36
• Fix most Foundry connection/statement metadata cases by @fornwall in #38
• Add Toxiproxy transport-fault resilience harness by @fornwall in #37
• Support ARRAY parameter binding and bulk-ingest by @fornwall in #39
• Error on undecodable wire values instead of fabricating NULLs by @fornwall in #41
• Declare all four ingest modes in the validation quirks; re-gate the two skipped tests by @fornwall in #42
• Report DriverArrowVersion in get_info by @fornwall in #43
• Implement the Foundry sample_ddl_constraints hook; report non-FK constraint_column_usage as NULL by @fornwall in #44
• Return rows from DML with THEN RETURN instead of discarding them by @fornwall in #45
• Lex GoogleSQL raw and triple-quoted strings in the SQL scanners by @fornwall in #46
• Quote identifiers with GoogleSQL escapes; decouple ingest param names by @fornwall in #47
• Port the Foundry type/literal suite to Spanner (dialect overrides + skips) by @fornwall in #48
• Classify statements behind leading @{...} statement hints by @fornwall in #49
• Tolerate trailing semicolons on queries; gate SqlQueryTrailingSemicolons by @fornwall in #51
• Preserve the gRPC code in vendor_code; map ABORTED to IO, not Internal by @fornwall in #50
• Cover the autocommit-toggle commit of buffered DML with a real test by @fornwall in #52
• Download only the library artifacts in the release job by @fornwall in #53
• Pin release-path actions to commit SHAs; least-privilege workflow tokens by @fornwall in #54
• Make the emulator scripts fail fast instead of falling through by @fornwall in #55
• Add an actionlint job to CI by @fornwall in #56
• Bind Utf8View/BinaryView, Int8 and Date64 parameters by @fornwall in #57
• Group statistics columns once; harden the metadata batch builders by @fornwall in #58
• Python packaging polish: license in the wheel, version floors, _version.py sync by @fornwall in #59
• Test/fuzz upkeep: scratch-table cleanup, JSON/FLOAT32 coverage, fuzz-corpus caching by @fornwall in #60
• Enforce the macOS deployment target; bump x86-64 to 10.15 by @fornwall in #61
• Keep the buffered DML when a manual-transaction commit fails by @fornwall in #62
• Align get_table_types with the vocabulary get_objects reports by @fornwall in #63
• Build Utf8 columns with a StringBuilder in the read hot path by @fornwall in #64
• Resolve bind parameter names once per batch, not per row by @fornwall in #65
• Drop comment-only segments in split_statements by @fornwall in #66
• Accept the standard isolation-level connection option by @fornwall in #67
• Map ingest-append failures to the spec-mandated error statuses by @fornwall in #68
• docs: fix README Rust quickstart deps and document all ingest modes by @fornwall in #69
• Accept adbc.ingest.target_db_schema and target_catalog by @fornwall in #71
• perf: group get_objects INFORMATION_SCHEMA batches into HashMaps by @fornwall in #72
• Add spanner.read.staleness / spanner.read.timestamp options (stale reads) by @fornwall in #73
• Trigger bulk ingest through execute(), not just execute_update() by @fornwall in #70
• ci: inspect wheels before publishing to PyPI by @fornwall in #74
• Make build.rs fall back gracefully when Cargo.lock is missing by @fornwall in #76
• Document the adbc.connection.readonly connection option by @fornwall in #78
• Add offline unit tests for statement.rs option-coercion helpers by @fornwall in #77
• Cap streamed chunks by an approximate byte budget, not just row count by @fornwall in #75
• Gate tag→publish on tests and green CI (REVIEW.md P1 #4) by @fornwall in #79
• test: cover adbc.connection.readonly enforcement by @fornwall in #82
• docs: correct CLAUDE.md against ground truth by @fornwall in #80
• test: fail loud instead of self-skipping when target env is missing by @fornwall in #81
• refactor: extract shared src/options.rs for option parsing by @fornwall in #83
• docs: document ingest modes, emulator port-9010, and manual-mode None in README by @fornwall in #84
• docs: document read_partition executable-descriptor trust caveat by @fornwall in #87
• ci: hoist tag-vs-version check into a shared version-gate job by @fornwall in #86
• test: cover create_append ingest and create-on-existing error path by @fornwall in #85
• refactor: move INFORMATION_SCHEMA collectors into feature modules by @fornwall in #88
• fix: drop parse_int64 f64 fallback that lost precision above 2^53 by @fornwall in #92
• docs: single revert-checklist anchor for the two git-pin revs by @fornwall in #90
• fix: bounds-check str_col to avoid panic on zero-column metadata batch by @fornwall in #89
• perf: run get_statistics per-table scans with bounded concurrency by @fornwall in #93
• fix: exclude TOKENLIST/PROTO columns from get_statistics distinct counts by @fornwall in #91
• docs: spell out manual-transaction no-read-your-writes and DML/DDL reordering by @fornwall in #96
• Accept adbc.ingest.temporary=false as a no-op by @fornwall in #95
• Fix is_dml_returning false positive on CASE ... THEN return by @fornwall in #94
• Fix autocommit-enable race that could strand concurrently buffered DML by @fornwall in #97
• Chunk bulk-ingest DML under Spanner's per-commit limits (REVIEW.md P1 #3, stage a) by @fornwall in #98
• ci: run doctests and build docs with --all-features by @fornwall in #99
• Make get_table_schema honor its catalog argument by @fornwall in #100
• Error loudly on undecodable values in ARRAY/STRUCT columns by @fornwall in #101
• Refuse emulator mode combined with explicit credentials by @fornwall in #102
• Reject DML in execute_schema with a clear error by @fornwall in #103
• Add consolidated option tables to both READMEs by @fornwall in #107
• Make adbc.connection.readonly a live flag instead of a creation-time snapshot by @fornwall in #105
• Factor dense-union assembly into a shared nested::dense_union helper by @fornwall in #104
• Skip all INFORMATION_SCHEMA queries in get_objects at Catalogs depth by @fornwall in #106
• Add test coverage for four small untested surfaces by @fornwall in #108
• Stream bound queries in one shared read-only snapshot by @fornwall in #109
• Report SPANNER_TYPE as xdbc_type_name in get_objects by @fornwall in #110
• Add tests proving sliced batches/arrays bind at their offset by @fornwall in #111
• get_statistics: serve exact statistics for approximate=true by @fornwall in #112
• Add criterion benchmarks for the rows-to-Arrow conversion path by @fornwall in #113
• Bind arrow.json-tagged parameters as Spanner JSON by @fornwall in #114

Full Changelog: v0.5.0...v0.6.0