Changes for version 0.42 - 2026-09-12
- Enhancements
- XLS backend: .xls (old binary format) now detected before .xlsx; opened via DBD::Excel (Spreadsheet::ParseExcel internally); all queries go through DBI/SQL; no in-memory slurp; type set to 'Excel'.
- XLSX backend rewritten: .xlsx (modern OOXML format) is now parsed directly via Spreadsheet::ParseXLSX rather than DBD::Excel. The workbook is slurped into an in-memory hash (keyed mode) or array (no_entry mode) — the same fast-path used by CSV, DBM::Deep, and HTML backends. No DBI handle is created; type is set to 'XLSX'. Multi-worksheet access via the 'table' constructor parameter is fully supported.
- Query builder: terminal methods all(), first(), count() now delegate to the in-memory path for type 'XLSX' (matching the existing Deep/BerkeleyDB delegation), via the check ($db->{'type'} // '') =~ /\A(?:Deep|XLSX)\z/.
- Remote backend: .xls added to the remote extension probe list alongside the existing .xlsx entry.
- Bug Fixes
- DF-11.5 (GitHub Actions flake): replaced !-f $temp_path filesystem check with a Scalar::Util::weaken weak-reference check. The old check occasionally failed on CI when the OS or DBI still held the file descriptor open briefly after DESTROY; the weak-reference check directly tests GC, independent of OS filesystem visibility.
- Zero-byte CSV (keyed mode): _open() now stores {} instead of () in $self->{'data'}. The empty-list assignment was equivalent to undef, which caused query methods to fall through to DBI with a 0-column table and croak on the 'entry IS NOT NULL' WHERE clause. The empty hashref mirrors what the newline-only path stores and lets all fast-path methods return correct empty results without touching the DBI handle.
- Tests
- t/xlsx.t: completely rewritten with two independent SKIP sections. Section A — .xls binary format via Spreadsheet::WriteExcel + DBD::Excel, type 'Excel', DBI path (10 tests). Section B — .xlsx OOXML format via Excel::Writer::XLSX + Spreadsheet::ParseXLSX, type 'XLSX', in-memory slurp, query-builder delegation, no_entry mode, and table override (24 tests). Each section is skipped independently when the required modules are absent.
- t/function.t A32, t/unit.t section 21, t/edge_cases.t EC19, t/integration.t section Q: updated to use Excel::Writer::XLSX for .xlsx fixtures and assert type 'XLSX' / $db->{'data'} defined instead of the former type 'Excel' / DBI handle assertions.
- t/data-flow.t DF-11.5: weak-reference check for File::Temp GC (see above).
- t/empty_files.t (new): 8 subtests (EF1–EF8) covering CSV, PSV, XML, and gzip CSV backends with empty (0-byte) and newline-only files. Verifies new() does not crash, count() returns 0, selectall_arrayref/selectall_array return empty, fetchrow_hashref returns undef. XML tests assert graceful croak (not segfault) since empty/newline content is invalid XML. Gzip CSV subtests are skipped when Gzip::Faster is unavailable.
- t/extended_tests.t EX5: updated to reflect that zero-byte keyed CSV now stores an empty hashref (not undef) in $db->{'data'} after open, enabling fast-path query methods to return 0/undef/[] without touching DBI.
Documentation
Modules
Read-only Database Abstraction Layer (ORM)
Fluent, chainable query builder for Database::Abstraction