Changes in version 0.1.3                        

  - Updated the embedded polyglot-sql engine from 0.12.0 to 0.13.2
    (upstream changelog:
    https://github.com/tobilg/polyglot/blob/main/CHANGELOG.md).
    Highlights inherited by the R API: two new dialects, "hana" (SAP
    HANA, aliases "saphana"/"sap_hana") and "vertica", bringing
    sql_dialects() to 36 entries; set-operation output types
    (UNION/INTERSECT/EXCEPT, including BY NAME and CORRESPONDING) are
    now resolved by sql_analyze() and the type annotation; Oracle output
    renders LIMIT as OFFSET ... FETCH FIRST ... ROWS ONLY and compound
    queries keep a trailing FETCH FIRST; branch-local row limits inside
    set operations are wrapped in parentheses instead of limiting the
    whole union; DuckDB // integer division and ORDER BY ALL translate
    correctly; BigQuery named parameters, decimal precision, interval
    arithmetic and DATE(timestamp, time_zone) convert to DuckDB
    faithfully; Snowflake PIVOT/UNPIVOT over CTEs no longer loops in
    lineage analysis; schema validation covers VALUES sources, LATERAL
    and INSERT ... SELECT.
  - On macOS x86_64 (Intel), the embedded engine crate is now compiled
    without optimization whenever NOT_CRAN is unset, which is how the
    CRAN binaries are built (configure/tools/config.R). The CRAN
    builders for that platform stop an installation after 30 minutes and
    interrupted the 0.1.1 and 0.1.2 builds. The build is about five
    times faster; the engine runs 2-4 times slower there, which is still
    a few milliseconds per call, and the shared object is larger.
    NOT_CRAN=true R CMD INSTALL . gives the fully optimized build. Other
    platforms are unchanged.

                 Changes in version 0.1.2 (2026-10-08)                  

  - Lighter Rust release profile (opt-level = "s", no LTO, 16 codegen
    units) to cut the cost of building from source: about half the
    compile time and roughly a quarter less peak memory for the largest
    crate. This targets the CRAN macOS x86_64 builders, which stopped
    the installation at their 30-minute limit, and the linux-arm64
    check, where the compiler ran out of memory. On macOS and Linux the
    shared library now exports only its R entry point and leaves out
    local symbols, which keeps the installed size at the level of 0.1.1.
    The R API and results are unchanged.
  - On Linux and macOS the shared object no longer references the C
    library's abort(). The Rust standard library calls it as a last
    resort (a failed memory allocation, a panic raised while another
    panic is being handled); those calls are now redirected at link time
    to a function that raises an R error instead, so compiled code can
    never terminate the R session. This removes the Found 'abort' NOTE
    of the CRAN linux-arm64 check. Ordinary Rust panics were already
    caught and turned into R errors.

                 Changes in version 0.1.1 (2026-09-27)                  

  - Updated the embedded polyglot-sql engine from 0.6.2 to 0.12.0
    (upstream changelog:
    https://github.com/tobilg/polyglot/blob/main/CHANGELOG.md).
    Highlights inherited by the R API: parser-depth protection enabled
    by default (deeply nested SQL now raises a polyglot_guard_error
    instead of risking a stack overflow); grouping, aggregate and window
    placement errors in semantic validation (E230-E232); sql_analyze()
    now reports scoped non-projection columnUses and propagates
    nullability and cast types through CTEs and derived tables; NULLS
    FIRST/NULLS LAST are preserved when formatting; 128-bit and unsigned
    integer types; name-aligned UNION BY NAME handling; many dialect
    fixes (DuckDB, Snowflake, ClickHouse, BigQuery, T-SQL).
  - sql_validate(semantic = TRUE) now also reports semantic correctness
    errors E230-E232 (ungrouped columns, misplaced aggregates and window
    functions), which invalidate the statement; quality hints remain
    warnings. Default syntax-only validation is unchanged.
  - Errors of the new upstream kinds invalid_input and column_resolution
    are surfaced as plain polyglot_error conditions.
  - macOS: the Rust build now uses the deployment target of R's C
    compiler (including a -mmacos-version-min flag set in CC) instead of
    the version of the running system. This removes the ld warning
    "object file was built for newer 'macOS' version than being linked"
    seen in the CRAN M1mac checks.

                 Changes in version 0.1.0 (2026-08-04)                  

Initial release, wrapping the polyglot-sql Rust crate 0.6.2.

  - Core: polyglot_version(), sql_dialects() (34 dialects with aliases).
  - Transpilation and formatting: sql_transpile(), sql_format(),
    sql_generate() with complexity guards and configurable handling of
    unsupported constructs.
  - Parsing and validation: sql_parse(), sql_tokenize(), sql_validate()
    (strict syntax, semantic warnings, schema-aware checks).
  - Analysis and lineage: sql_source_tables(), sql_lineage(),
    sql_analyze(), sql_optimize(), sql_diff(), sql_annotate_types(),
    sql_openlineage().
  - Classed error conditions (polyglot_error, polyglot_parse_error,
    polyglot_transpile_error, polyglot_validation_error,
    polyglot_guard_error, ...); Rust failures never crash the R session.
  - Fully offline source builds via vendored crates (cargo vendor).