Changes in version 0.2.0                        

Performance

  - Row matching for duplicate-key strategies keep_all and all_pairs
    (ks_match_rows_keep_all(), ks_match_rows_all_pairs()) now
    pre-allocates the per-group parts list and caches the level factor
    used for splitting, removing O(n^2) list-copy churn on data sets
    with many unique keys.
  - The main comparison loop in ks_compare() now pre-allocates the
    value_parts accumulator and uses order(..., method = "radix") for
    the final stable sort of value_diff, materially speeding up wide
    comparisons (many matched column pairs) and large diff tables.
  - ks_schema_diff() no longer rescans meta_b$name / meta_c$name for
    every column pair; it builds a name-indexed lookup once and reuses
    per-row metadata, and it hoists the compare_labels / compare_formats
    option checks out of the per-column loop.
  - ks_value_diff_one() now slices the unequal cells once and reuses the
    result for ks_explain_diffs(), format_cell(), and ks_na_flow(),
    avoiding three redundant subsets on large columns; the type-mismatch
    branch likewise slices base_col / comp_col only once.

Internal

  - No user-facing API changes; all optimisations are
    behaviour-preserving.

                        Changes in version 0.1.0                        

  - Initial development release. ksCompare is a tidyverse-native engine
    for comparing two data frames in the spirit of SAS PROC COMPARE,
    with extensions for clinical/pharma workflows. Highlights:
      - ks_compare() orchestrator with explicit by keys, by = "auto"
        uniqueness inference, and explicit column mapping.
      - ks_tol() with absolute, relative, and ULP tolerance plus
        per-column overrides.
      - ks_comp_options() for NA semantics, SAS special-missing
        handling, label/format comparison, string trim/case/normalize
        toggles, and an optional path that pins every downstream
        ks_report_*() artefact to a single output folder. Every field
        also falls back to a getOption("ksCompare.<arg>") global, and
        ks_set_comp_options() forwards to options() for project-wide
        defaults.
      - ks_comparison S3 with print(), summary(), as_tibble(),
        ks_tidy(), ks_glance().
      - Per-type cell diff for numeric (with ULP-aware compare),
        character, factor, date/datetime, and haven_labelled columns;
        encoding-safe string comparison.
      - Schema diff including label and SAS-format mismatches; format
        comparison is trailing-dot- and case-tolerant so haven import
        artefacts do not surface as false diffs.
      - Manifest with input hashes and package version for QC
        traceability.
      - as_outbase(), as_outcomp(), as_outdif(), as_outnoequal() helpers
        reshape diffs into PROC COMPARE-style tibbles, plus ks_sysinfo()
        for a SAS-compatible bitmask.
      - Smart features: auto-key inference, fuzzy column rename
        suggestions (Suggests-gated on stringdist), duplicate-key
        strategies (first, last, keep_all, all_pairs, error), and
        pattern detection (constant_offset, constant_scale, sign_flip,
        integer_round, whitespace_only, trim_only, case_only,
        factor_recoded).
      - Reports: ks_report_html() (self-contained htmltools + reactable
        page with sticky TOC, KPI cards, status pill, themed tables, an
        optional group_by_key mode that renders one collapsed block per
        key value, and a print-friendly stylesheet) and ks_report_xlsx()
        (openxlsx2 multi-sheet workbook with conditional formatting on
        OUT_DIF). The HTML report inlines every JS/CSS dependency into a
        single file, so no sibling lib/ folder is written next to the
        report.
      - I/O: ks_read_sas() / ks_read_xpt() (haven wrappers preserving
        labelled metadata) and ks_read_arrow() (Parquet / Feather).
      - Ecosystem: ks_assert_clean() pipeline gate with
        ks_pointblank_step() adapter; as_ks_comparison() generic with
        method for arsenal::comparedf.
      - Documentation: pkgdown site, five long-form vignettes
        (getting-started, from-proc-compare, smart-features, reports,
        pipeline-gates), and CI workflows for R-CMD-check, pkgdown, and
        test-coverage.