Changes for version 1.10 - 2026-08-11
- Portability -- Windows reserved device names. CPAN Testers reported t/1021_integrity.t failing its three K3 subtests on MSWin32, on both perl 5.8.9 and perl 5.42.2, while every other platform passed. The cause was not the NULL handling those subtests exercise: the fixture table was called "nul", and Windows resolves con, prn, aux, nul, com0-com9 and lpt0-lpt9 to a character device whatever extension follows them. "nul.sch" was therefore the bit bucket. create_table saw no file, wrote a schema that went nowhere, and the first INSERT reported that the table did not exist.
- Fixed: t/1021_integrity.t names its NULL fixture nl1. This alone is what the FAIL reports were about.
- Fixed: create_database, create_table and create_index now reject the device names above, in any letter case. Before this a user who named a table "nul" on Windows got a table that appeared to be created, swallowed every write and read back as empty. The names are rejected on every platform, not only on MSWin32, so that a data directory built on one system stays usable on another. A name that merely begins with one, such as "nul1", is unaffected.
- Incompatible: a database, table or index already carrying one of those names is unreachable from 1.10. Windows could never hold one, but a system that could -- a Unix host, say -- may have one built by 1.09 or earlier, and every method that takes a name now refuses it, drop_database and drop_table included. There is no way to reach or remove it through the API; use the tools of the operating system. Noted in BUGS AND LIMITATIONS and in README.
- Added: t/1022_names.t gains the N9 group (10 assertions) covering the rejection through both the low-level API and the SQL layer.
- Documented: the rule in the identifier restriction section, in DIAGNOSTICS under "Invalid database name", and in README.
- Test suite -- assertions that could never fail. ok() and is() take a plain list, so their arguments are evaluated in list context, and a match in list context returns the empty list when it fails rather than a false scalar. ok($str =~ /wanted/, 'name') therefore handed 'name' to ok() as the condition and nothing as the description whenever the match failed: the assertion printed itself as a nameless "ok" and could not fail. With a capture group the error ran the other way, a match capturing '0' reporting a failure. The module's own ok() carries an ($;$) prototype and was never affected; the harnesses the t/1NNN files define for themselves were.
- Fixed: 30 such assertions across twelve test files now force scalar context. All of them pass once made real, so nothing was hiding behind them.
- Added: t/lib/INA_CPAN_Check.pm 0.44 adds K4, which reports a bare match handed to an ok()/is()/isnt() declared without a prototype in the file under test. pmake.bat 0.50 spells the same rule in its own selfcheck, so the two implementations agree.
- Changed: K1 (a comma is followed by whitespace) now runs over t/ as well as lib/, which it has since INA_CPAN_Check 0.43; the test files were brought into line with it, and 48 stray double commas in t/1004_subquery.t and t/1006_cpan_api.t were removed.
Documentation
Modules
Pure-Perl flat-file relational database with DBI-like interface