Changes

Revision history for pgTAP
==========================

1.3.4
--------------------------

* Fixed the upgrade from v1.2.0 to v1.3.0. Thanks to @runqitian for the pull
  request (#338).

* Added `name, name` variants for `has_composite` and `hasnt_composite` Thanks
  to Jim Nasby for reporting the issue (#234) Rodolphe Quiédeville for for the
  pull request (#339).

1.3.3 2024-04-08T13:44:11Z
--------------------------

* Improved the correctness and performance of `col_type_is` on Postgres 17
  thanks to the introduction of the `to_regtypemod()` function.
* Fix bug introduced in v1.3.2 where `col_type_is` throws an error when the
  type isn't in the search path. Thanks to Erik Wienhold for the PR (#332)!
* Removed the reference to PGUnit from the docs, as the project seems to have
  disappeared. Also the link to the GitHub repo. Thanks to Vick Khera for the
  report (#333)!

1.3.2 2024-02-04T18:59:37Z
--------------------------

* Replaced the C `parse_type()` function added in v1.3.1 with a PL/pgSQL
  function, eliminating the need for the the installing use to be a superuser.
  Aliases still work for all data types passed to `col_type_is()` except for
  intervals, which must be specified exactly as rendered by Postgres itself,
  without aliasing. Thanks to @spencerbryson for the report (#328) and to
  Erik Wienhold for the PL/pgSQL technique necessary to properly format
  type strings (#315).
* Added tests to ensure that that the function-testing assertions also support
  procedures and noted the fact where appropriate in the documentation.
  Thanks to @hettie-d for the call-out (#329)!

1.3.1 2023-09-24T15:29:42Z
--------------------------
* Revamped the handling of data type declarations in `col_type_is()` to always
  accurately normalize data type representations, whether using aliases, such
  as `varchar` for `character varying`, or more complicated types that failed
  to work in v1.3.0, such as `interval second(0)`. This is done by delegating
  the parsing of the type declaration to a core PostgresSQL function. Many
  thanks to Erik Wienhold for the report and for ultimately uncovering and
  implementing an SQL interface to the `parseTypeString()` core function
  (#315).
* Removed the documentation for `pg_typeof()`, which was removed from pgTAP
  when support for PostgreSQL 8.3 was dropped, since the same function has
  been available in the PostgreSQL core since then.
* Improved the support for type name aliases to the following functions:
    * `has_cast()`
    * `hasnt_cast()`
    * `cast_context_is()`
    * `domain_type_is()`
    * `domain_type_isnt()`
    * `has_operator()`
    * `hasnt_operator()`
* Added the `has_pk( :schema, :table )` variant, which requires that `:table` be
  cast to `name` to avoid confusion with the `has_pk( :schema, :description )`
  variant. Thanks to Adrian Klaver for the report (#287)!

1.3.0 2023-08-14T22:14:20Z
--------------------------
* Fixed an issue with xUnit tests where they would exit immediately on
  unexpected errors (aside from PL/pgSQL RAISE errors) rather than exit just the
  current test function. Thanks to @kzathey for the report and to @fjf2002 for
  the analysis.
* Fixed a pluralization error reporting xUnit test failures to report "failed 1
  test" instead of "failed 1 tests".
* Removed the indentation for diagnostic comments at the start of subtests to
  make it easier to see where they start and end. This changes the output
  format, which will not affect tests run by `pg_prove`, but will break tests
  that depend on diffing files, as `pg_regress` does. Thanks to Matt DeLuco for
  highlighting the visual confusion of the indented diagnostic (#264).
* Added variants of `col_is_pk()` with schema and without description (variants
  without schema or description already existed). Thanks to Luca Ferrari and
  Matteo Ferrando for the PRs (#178 & #199)!
* Added a note on the use of temporary tables to the `fk_ok()` documentation, as
  well as a test to ensure the suggested expressions work. Thanks to Jim Nasby
  for highlighting the special treatment of the temporary schema (#109).
* Removed redundant `DROP` statements from the 1.2.0 upgrade script that
  prevented upgrades from working properly. Thanks to @robins for the PR
  (#300)!
* Fixed a test failure caused by the removal of support for select rules in
  postgres/postgres@b23cd18 (expected in Postgres 16). Thanks to @Deltaus for
  the report (#309)!
* Fixed `LIKE` expression in `schemas_are()` to escape the underscore wildcard,
  making it `pg\_%` instead of `pg_%`. Thanks to Jehan-Guillaume (ioguix) de
  Rorthais for the PR (#311)!
* Updated function-testing functions to allow short names for argument types, so
  that they can be specified as, e.g., `int` and `bool` rather than `integer`
  and `boolean`. The return value argument to `function_returns` and and data
  type tests such as `col_type_is` and `cast_context_is` must still use full
  type names. Thanks to @wphilips53 for the suggestion (#292).
* Updated type-testing functions to allow short names for types, and updated
  function-testing functions to allow short names for argument types. This means
  that common aliases for standard types can be specified as, e.g., `int` and
  `bool` rather than `integer` and `boolean`. Thanks to @wphilips53 for the
  suggestion (#292).
* Tweaked the installation docs for Docker, binary Linux distributions, and
  downloading the source from PGXN. Thanks to @jed-walker-icd (#275) and
  @machineghost (#301) for the suggestions!

1.2.0 2021-12-05T18:08:13Z
--------------------------
* Made the description field optional in `has_view()` and `hasnt_view()` when
  specifying schema (#230). Thanks to Godwottery for the patch!
* Added support for the date and time keywords `CURRENT_DATE`, `CURRENT_TIME`,
  `CURRENT_TIMESTAMP`, `LOCALTIME`, and `LOCALTIMESTAMP` to `col_default_is()`.
  Thanks to Kevin Brannen for the report (#244)!
* Fixed failure to run tests on Postgres 13. Thanks to Christoph Berg for the
  report (#248).
* Documented that mixed-case column names created with double quotes
  must be double-quoted when passed to `has_index()` or `is_indexed()`,
  sadly unlike other column arguments in pgTAP. Thanks to Keith Fiske for
  the report (#247).
* Fixed test failures where the current username was not being quoted as an
  identifier. Thanks to Matt DeLuco for the report (#216) and pull request
  (#259)!
* Fixed the `col_not_null()` drop statements in the uninstall script. Thanks
  to Kyle L. Jensen for the report (#252).
* Removed straggler references to Postgres 9.0 and earlier, including
  conditional tests.
* Updated all relevant URLs to use https instead of http.
* Added instructions to install the extension on a host with PostgreSQL running
  and with the contrib modules installed. Thanks to F. Eugene Aumson for the
  report (#276).
* Fixed test failures due to changes in PostgreSQL 14. Thanks to Christoph Berg
  for the report (#277).
* Added `isnt_member_of()` (#38, #204). Thanks to Wolfgang Walther for the
  pull request (#249).
* Added docker files for local test environment. Thanks to Wolfgang Walther for
  the pull request (#250).
* Added `hasnt_operator()`, `hasnt_leftop()`, `hasnt_rightop()` (#38). Thanks to
  Wolfgang Walther for the pull request (#251).
* Improved the output of `results_eq()` and `results_ne()` when the number of
  columns or their types are different (#37). Thanks to Wolfgang Walther for
  the pull request (#255).
* Fixed testing issue where tests for unsupported features were not properly
  skipped on Postgres 9.6 and earlier. Only appeared when tests were run
  serially (#279).
* Fixed an issue where tests might not run in parallel even when the server
  supported it (#279).

1.1.0 2019-11-25T19:05:38Z
--------------------------
* Remove support for PostgreSQL older than 9.1. Note that some automated tests
  don't run on versions older than 9.4, so it's recommended you run PostgreSQL
  9.4 at minimum.
* Add a `throw_exception` argument to `finish()`. `finish(true)` will raise an
  exception if any tests failed. Thanks to Stephen Paul Weber for the idea
  (#80), and Rodolphe Quiédeville for most of the work (#104)!
* Make description optional for `col_not_null()` (#221).
* Fix syntax of test_user function example in the docs (#208). Thanks to
  PeteDevoy for the patch!
* Fix issue with `pgtap-core` referencing `pg_proc.proisagg` on PostgreSQL 11+
  (#197).
* Add PostgreSQL 12.0 to automated tests (#223).
* Fix `pgtap_version()`, which incorrectly returned 0.99 when upgrading from
  version 0.99.0 to 1.0.0 (#214).
* Fix issue with using `pg_upgrade` to PostgreSQL 11+ with pgTAP installed
  (#201, #215).
* Start using `DEFAULT` in functions.
* Add automated testing of extension upgrade scripts (#128), as well as testing
  of `pg_upgrade` with pgTAP installed (#218, #222).

1.0.0 2019-02-21T22:39:42Z
--------------------------
* Fixed uninstall script generation to properly emit `DROP TYPE` statements.
  Thanks to Kamen Naydenov for the PR (#181)!
* Added policy test assertion functions for PostgreSQL 9.5 and higher:
    + `policies_are()`
    + `policy_roles_are()`
    + `policy_cmd_is()`
  Thanks to Kamen Naydenov for the PR (#185).
* Fixed an issue with `has_function()` where a test would fail if arguments
  are schema-qualified even though the schema is in the search path. Thanks to
  Sandro Santilli for the bug report (#179) and PR (#180).
* Added table inheritance testing functions:
    + `has_inherited_tables()`
    + `hasnt_inherited_tables()`
    + `is_ancestor_of()`
    + `isnt_ancestor_of()`
    + `is_descendent_of()`
    + `isnt_descendent_of()`
  Thanks to Luca Ferrari for the inspiration.
* Fixed `col_default_is()` to properly report special SQL syntax functions
  like `CURRENT_USER` and `CURRENT_CATALOG` instead of evaluating them on
  PostgreSQL 10 and higher.
* Changed the uninstall script to use `DROP IF EXISTS` rather than just `DROP`
  so that it works even with partial or failed installs. Thanks to Sandro
  Santilli for the PR (#194).
* Officially end support for PostgreSQL 8.2 and older.

0.99.0 2018-09-16T20:55:41Z
---------------------------
* Updated introductory documentation. A lot has changed over the years!
* Added the "Secrets of the pgTAP Mavens" section of the documentation.
  Contributions wanted!
* Fixed `pg_version_num()` to always return the value of `server_version_num`
  on PostgreSQL 8.3 and higher, thus fixing a test failure on 10.1. Thanks to
  Christoph Berg for the report (#148)!
* Changed the diagnostic output of extra and missing values from the `*_are()`
  functions to sort the extra and missing values. Fixes a test failure on
  PostgreSQL 10, and generally makes the diagnostics more deterministic and
  easier to read. Thanks to Michael Glaesemann for the report (#158)!
* Fixed an issue on PostgreSQL 11, where the `pg_class.relhaspkey` column has
  been removed. The primary key testing functions now refer to
  `pg_index.indisprimary`, instead. Thanks to Keith Fiske for the report
  (#174)!
* Fixed an issue on PostgreSQL 11, where the `pg_proc.proisagg` column has
  been replaced with the more generically useful `pg_proc.prokind`. Thanks to
  Keith Fiske for the report (#174)!
* Documented that a `RECORD` can be passed to `row_eq()` on PostgreSQL 11,
  and fixed the tests to match this new functionality.
* Removed tests that only ran on 8.0, support for which was discontinued in
  v0.90.0.
* Removed `test/sql/run*` tests, which intentionally fail, from the `test`
  target in the `Makefile`. It's not meaningful to test them via `pg_prove`;
  they must instead be tested by `make installcheck`, so their output can be
  compared against output that varies by PostgreSQL version. Thanks to Jim Nasby
  for the report (#172).
* Fixed a bug where `have_fk()` would fail if the table had no primary key.
  Thanks to @bigmodem for the report! (#171).

0.98.0 2017-11-06T22:30:54Z
---------------------------
* `make install` once again properly installs the versioned SQL file required
  by `CREATE EXTENSION`. Thanks to Keith Fiske for the reports (Issue #131,
  Issue #143) and Jim Nasby for the fix (Issue #132).
* Removed references to the Epic test framework, since its domain
  (`epictest.org`) has expired and the project appears to no longer be
  available online. Thanks to Andreas Brekken for the report (#140).
* Fix failing version tests on PostgreSQL v10. Thanks to Erik Rijkers
  for the patch.
* Improved foreign tables test to account for varying message order for
  diagnostic output. Thanks to Erik Rijkers for the patch.
* Clarified the use of the `:schema` parameter to the extension-testing
  functions in the documentation. Required since extensions do not belong to
  schemas, but their objects may be specifically created in a schema. Thanks to
  Jim Nasby for the report (#121)!
* Updated the table-testing functions to also recognize partitioned tables:
    + `has_table()`
    + `hasnt_table()`
    + `tables_are()`
    + `table_owner_is()`
* Added partition-testing assertion functions:
    + `is_partitioned()`
    + `isnt_partitioned()`
    + `is_partition_of()`
    + `partitions_are()`
* Added the materialized view-testing assertion functions to the v0.95.0 upgrade
  script; they were inadvertently omitted in the v0.95.0 release.
* Fixed failing tests on Postgres 8.1.
* Added function type testing functions to complement `is_aggregate()` and
  `isnt_aggregate()`:
    + `is_normal_function()`
    + `isnt_normal_function()`
    + `is_window()`
    + `isnt_window()`
    + `is_procedure()`
    + `isnt_procedure()`
* Made the diagnostic output of `is_aggregate()` and `isnt_aggregate()`
  consistent. Previously, when the function did not exist, some instances would
  generate diagnostic output and some would not.

0.97.0 2016-11-28T22:18:29Z
---------------------------
* Fixed the default description for `hasnt_opclass()` to say "should not
  exist" instead of "should exist", thanks to Rodolphe Quiédeville (PR #99).
* Added `is_indexed()`, which tests to see if specific table columns are
  indexed. It's effectively the same as `has_index()` except that it doesn't
  require an index name and does require one or more column names or index
  expressions. Thanks to Rodolphe Quiédeville (PR #103).
* Fixed `pg_version_num()` to work with the new version format coming in
  PostgreSQL 10.
* Fix `lives_ok()` and `throws_ok()` to also trap `ASSERT_FAILURE` (PR #119).
* Add negation functions `isnt_aggregate()` and `isnt_definer()`, which return
  the opposite of `is_aggregate()` and `is_definer()`. Thanks to Rodolphe
  Quiédeville (PR #106).
* Add `has_extension()` and `hasnt_extension()`. Thanks to Rodolphe Quiédeville
  (PR #101).

0.96.0 2016-05-16T20:53:57Z
---------------------------
* Added an optional `:exclude_pattern` parameter to `findfuncs()` to prevent
  matching of some functions. This is useful to prevent matching
  startup/setup/teardown/shutdown functions. Patch from Jim Nasby.
* Changed `runtests()` to use the new `:exclude_pattern` to prevent the
  running of startup/setup/teardown/shutdown functions as test functions.
  Patch from Jim Nasby.
* Fixed `database_privs_are()`, `schema_privs_are()`, `tablespace_privs_are()`
  `fdw_privs_are()`, and `server_privs_are()` to properly handle
  case-sensitive object names. Thanks to Ingo for the report!
* Updated the Makefile to handle the case when PGXS sets the PERL variable to
  `missing perl`.
* Fixed a test failure due to the use of a hash index on PostgresSQL 9.5.
* Added `extensions_are()`, prompted by a pull request from Grégoire HUBERT
  (Issue #98).
* The `sql/unininstall_pgtap.sql` file is now generated by `make` from the
  `sql/pgtap.sql` file. This should make it much more up-to-date and accurate
  in the future.
* Added full stack diagnostic output to the `lives_ok()` failures, as well as
  xUnit test function errors. Patch from Jim Nasby.

0.95.0 2015-03-20T20:31:41Z
---------------------------
* Added materialized view-testing assertion functions:
    + `materialized_views_are()`
    + `has_materialized_view()`
    + `hasnt_materialized_view()`
    + `materialized_view_owner_is()`
  Thanks to Charly Batista for the contribution!
* Added `performs_within()`, a function that tests the average performance of
  an SQL statement over a number of iterations. Contributed by Ryan Duryea.
* Changed `is_member_of()` to work with roles instead of users. Patch from Jim
  Nasby.
* Granted SELECT access to the views created by pgTAP to PUBLIC, to avoid
  permission errors when the tests are run by a user other than the owner of
  those views. Thanks to Jim Nasby for the report.
* Added `foreign_tables_are()` functions. Thanks to Chad for the pull request!
* Updated `enum_has_labels()` to properly sort labels added by `ALTER TYPE ADD
  VALUE`.
* Added `diag_test_name()` to `pgtap-core`, since it's called by `_runner()`
  in that extension. Thanks to Jim Nasby for the spot.
* Added `isnt_strict()` to complement `is_strict()`. Requested by Jim Nasby.
* Added schema/table/column variations of `col_is_unique()`. Contributed by
  Jim Nasby.
* The `runtests()` function now executes each test as a TAP subtest, with its
  own plan and test count. This allows each test function to function more
  independently of the others. It's easier to read the TAP, too. Users of
  xUnit-style tests with `pg_regress` will need to update their expected
  output files, however.
* An exception thrown by a test no run by `runtests()` no longer halts
  execution. It is instead properly rolled back, reported as a subtest
  failure, and the error message emitted as a diagnostic message. Thanks to
  Vitaliy Kotov for the report.
* Eliminated the temporary table `__tresults__`. Test results are no longer
  written to a table, but simply emitted. This elimination of unnecessary I/O
  should make tests run a bit faster.
* Added missing functions `has_sequence(:schema, :sequence)` mentioned in the
  docs (Issue #58):

0.94.0 2014-01-07T01:32:36Z
---------------------------
* Fixed the Makefile to allow `PG_CONFIG=/path/to/pg_config` to actually work.
  Patch from Aaron W. Swenson.
* Eliminated superfluous tabs in the `Makefile` that could cause it to fail in
  some situations. Patch from Eric Neault.
* Fixed `function_privs_are() to work with functions with long names and/or
  signatures. Thanks to Jörg Beyer for the detailed report.
* Added missing functions mentioned in the docs (Issue #54):
    + `has_table(:schema, :table)`
    + `hasnt_table(:schema, :table)`
    + `has_foreign_table(:schema, :table)`
    + `hasnt_foreign_table(:schema, :table)`
* Fixed the format of SKIP output to match the
  [TAP spec](https://testanything.org/tap-specification.html#skipping-tests).
* Fixed incorrect handling of leading space when comparing diagnostic output
  in `check_test()`.
* Fixed an installation issue on PostgreSQL 9.3.2.

0.93.0 2013-01-28T20:14:58Z
---------------------------
* Added additional ownership-testing functions:
  + `tablespace_owner_is()`
  + `schema_owner_is()`
  + `index_owner_is()`
  + `language_owner_is()`
  + `opclass_owner_is()`
  + `type_owner_is()`
* Fixed misselling of "constraint" in constraint test diagnostic output.
* Fixed constraint, trigger, index, opclaass, and foreign key test functions
  so that they don't find objects outside the search path unless the schema is
  specified.
* Fixed `col_type_is()` so that it won't report the column as not existing
  when the column's data type is not in the search path.
* Removed `format_type()` and switched to the core PostgreSQL type formatting
  function so that type names always include the schema name if the schema is
  not in the search path. You might have to update data type tests to include
  the schema name if the type is not in the search path.
* `ALTER EXTENSION` now properly updates the value returned by
  `pgtap_version()`.
* Removed `CREATE EXTENSION` support for `pgtap-core` and `pgtap-control` (by
  deleting `pgtap-core.control` and `pgtap-schema.control`). There were never
  upgrade scripts for them, and upgrade scripts would be difficult to maintain
  anyway. These files are really designed for copying into other projects for
  inline testing support.

0.92.0 2013-01-16T00:41:07Z
---------------------------
* Fixed some misspellings of `SELECT` in documentation examples.
* Added note about the lack of typemods to the documentation for `has_cast()`.
* Added check to ensure a table is visible in tests for column defaults.
  Thanks to Henk Enting for the report.
* Added `has_relation()` and `hasnt_relation()`.
* Added `has_foreign_table()` and `hasnt_foreign_table()`.
* Added `has_composite()` and `hasnt_composite()`.
* Added new ownership assertion functions:
  + `relation_owner_is()`
  + `table_owner_is()`
  + `view_owner_is()`
  + `sequence_owner_is()`
  + `composite_owner_is()`
  + `foreign_table_owner_is()`
  + `function_owner_is()`
* Changed the uninformative error 'relation "__tcache__" does not exist' to
  the more useful 'You tried to run a test without a plan! Gotta have a plan'.
  Thanks to Steven Samuel Cole for [the
  report](https://github.com/theory/pgtap/issues/22).
* Changed `triggers_are()` so that it ignores internal triggers on PostgreSQL
  9.1 and higher. Thanks to Sherrylyn Branchaw for the report!
* Changed the diagnostic output of `results_eq()` for data type differences to
  reduce confusion when the rows have the same values of different types.
  Thanks to Peter Eisentraut and Taylor Ralston for [the
  nudging](https://github.com/theory/pgtap/issues/15).
* Added `isnt_empty()` to ensure that a query does not return the empty set.
* Fixed `has_index()` to properly handle indexes composed of both columns and
  expressions.
* Added new functions for testing a user's access to database objects:
  + `database_privs_are()`
  + `tablespace_privs_are()`
  + `schema_privs_are()`
  + `table_privs_are()`
  + `sequence_privs_are()`
  + `any_column_privs_are()`
  + `column_privs_are()`
  + `function_privs_are()`
  + `language_privs_are()`
  + `fdw_privs_are()`
  + `server_privs_are()`

0.91.1 2012-09-11T00:26:52Z
---------------------------
* The OS name is restored to the call to `os_name()`. The `getos.sh` shell
  script was not getting called during `make`. Thanks to Peter Eisentraut for
  [the report](https://github.com/theory/pgtap/issues/16).
* Noted `pg_prove --verbose` for showing test descriptions and fixed a few
  typos in the documentation. Thanks to Jay Levitt for the pull request!
* Fixed broken examples in the `row_eq()` documentation. Thanks to @set5think
  for the report!
* Fixed `has_tablespace()` to work on PostgreSQL 9.2, where its failure was
  preventing pgTAP from even being installed.

0.90.0 2011-12-03T20:18:16Z
---------------------------
* Removed the "Supported Versions" section of the documentation. Suffice it to
  say that pgTAP supports PostgreSQL 8.1 or higher.
* Added a build target to create a portable copy of pgTAP that can be included
  in any distribution and should just work for tesing on PostgreSQL 8.3+. The
  new files are `pgtap-core.sql`, which contains the core functionality, and
  `pgtap-schema.sql`, which depends on pgtap-core and adds in the schema
  testing assertion functions.
* Added abstract to the `provides` section of `META.json`.
* Removed use of `relistemp` column in an internal function, as it has been
  removed in PostgreSQL 9.1. As a bonus, it now works on PostgreSQL 8.3 and
  older without needing to be patched. Thanks to Tom Lane for the alternate
  syntax.
* Added PostreSQL 9.1 `CREATE EXTENSION` support.
* Removed `TAPSCHEMA` option to `make`. Use `PGOPTIONS=--search_path=tap` with
  `psql`, instead.
* Added `db_owner_is()`. Based on a patch by Gerd Koenig.
* Dropped support for PostgreSQL 8.0, as the new directory structure
  introduced in 0.25.0 is incompatible with the 8.0 version og `pg_regress`,
  making testing all but impossible. 8.0 has been EOLed for some time, so it's
  time for pgTAP to finally drop its support, too.

0.25.0 2011-02-02T03:21:55Z
---------------------------
* Minor documentation tweaks.
* The sequence created to keep track of test numbers is now a temporary
  sequence. This will prevent it from persisting beyond the current
  connection, which could lead to erorrs in future connections. Thanks to
  Richard Huxton for the report.
* Fixed `is_superuser()` and `isnt_superuser()` so that they properly detect
  superuser roles as well as users.
* Fixed examples of single-quoted string values stored in `psql` variables in
  `README.pgtap`.
* Fixed a bug in `plan()` where a sequence wasn't getting created as a
  temporary sequence. This wasn't a problem if all testing was done inside a
  transaction, but if not, you could only run tests once without an error.
* Fixed bug where `has_column()`, `col_type_is()`, `domain_type_is()`,
  `cast_exists()`, `cast_context_is()`, and `has_operator()`, were forcing the
  type argument to lowercase, which wasn't correct when the type is actually
  mixed case or uppercase (that is, named in double quotes when it was
  created).
* The expected type passed to `index_is_type()` is no longer compared
  case-insensitively, since it might be possible in the future to create a
  type with mixed case by using double quotes when naming it.
* Documented that double-quotes must be used for types created with
  double-quoted names in the strings passed to `casts_are()`. This should be
  the only part of the API that does require this sort of quoting.
* Added `diag(anyelement)` (for 8.3 and higher) and `diag(VARIADIC anyarray)`
  (for 8.4 and higher) to make it easier to emit diagnostics for whatever data
  you have at hand.
* The `_cleanup()` function now properly drops the temporary sequences as well
  as the temporary tables. These are usually automatically dropped when the
  tests finish and the client disconnects, but it's best to try to keep things
  tidy ourselves, too. Spotted by Peter Eisentraut.
* Fixed a bug in `has_pk()` so that it properly checks that a table is visible
  in the search path when a schema is not specified, and doesn't check that
  when the schema *is* specified. Patch from Cédric Villemain.
* Fixed a bug in `col_is_unique()` and `col_has_check()` where they considered
  *all* columns in a table that were unique or involved in a check constraint
  were assocated with only one constraint. Since one can have multiple unique
  or check constraints on a table, each using different columns, this was
  sub-optimal. Thanks to Cédric Villemain for the spot!
* Removed `pg_prove` and `pg_tapgen` from the distribution. They must now be
  installed separately from CPAN. If it's not installed, the intaller will
  issue a warning.
* Added `META.json` and distributed on [PGXN](https://pgxn.org/).
* Rearranged the source directory layout to more closely match the [preferred
  PGXN layout](http://manager.pgxn.org/howto#new-order).

0.24 2010-05-24T23:33:22Z
--------------------------
* Got `sql/artap.sql` tests passing again when building with `$TAPSCHEMA` set.
* Changed to nicer source URL in `contrib/pgtap.spec`.
* Fixed a bug in `has_member()` where it failed to confirm that users were
  actually members of a group.
* Fixed a bug where `display_type()` would include the schema name in its
  output for types not found in the current path, despite its assertion that
  it wouldn't.
* Changed the diagnostic output for row type mismatches from `set_eq()`,
  `set_ne()`, `bag_eq()`, and `bag_ne()` to include the schema name for types
  not found in the search path.
* Fixed bug in the diagnostic output of `set_eq()`, `set_ne()`, `bag_eq()`,
  and `bag_ne()` where a column's type would not be displayed if the column
  was not in the search path. Thanks to Rod Taylor for the spot!
* Made the implementation of `row_eq()` always return a value, even if the
  query it executes returns no rows.
* Fixed bug in the default description for `col_type_is()` where the column
  name was listed as the type!
* Updated the `Makefile` and the test suite to support PostgreSQL 9.
* Removed the primary keys from the temporary tables used internally by pgTAP
  and thus eliminated the logging of NOTICEs to the PostgreSQL log when pgTAP
  creates those tables.
* Eliminated the need to set `USE_PGXS=1` when building outsde the contrib
  directory of the PostgreSQL distribution. If `pg_config` isn't in the
  `$PATH` environment variable, one can simply set `PG_CONFIG` to point to it.
  If building inside the contrib directory, set `NO_PGXS=1`.
* Fixed typo in the statement to drop the `tap_funky` view in
  `uninstall_pgtap.sql`. Thanks to Erik Rijkers.
* Removed deprecated `can_ok()` functions. Use `has_function()`, instead.

0.23 2009-12-18T23:03:39Z
-------------------------
* Fixed broken Perl detection in `Makefile`.
* Copied OS detection from Perl's `Configure` script to new script,
  `getos.sh`. OS detection should now be much more accurate.
* Fixed `pg_prove` to exit with non-zero status if tests failed, as it works
  better in automated testing environments and consistent with `prove` (fix by
  Darrell Fuhriman).
* Updated documentation of supported environment variables in `bin/pg_prove`.
* Fixed a bug in the function tests where the return value of a function was
  not always consistently formatted. For example, `function_returns()` would
  find "bool" instead of "boolean".
* Added `isa_ok()`.
* Added support for building against 8.5 [Dan Colish].
* Fixed failing test on alpha releases [Jonathan Leto]
* Added `is_empty()` to test that a query returns an empty set (no results).
* Added `collect_tap()`, which makes it easier to run multiple tests in a
  single `SELECT` statement in combination with `skip()`.
* Added `throws_like()`, `throws_ilike()`, `throws_matching()`, and
  `throws_imatching()` to test that an SQL statement throws an error message
  matching a `LIKE` pattern or a regular expression.
* Added `roles_are()`, `types_are()`, `domains_are()`, `enums_are()`,
  `triggers_are()`, `casts_are()`, `operators_are()`, and `columns_are()`.
* Fixed the diagnostic output from `isnt()` and `col_isnt_pk()` when they
  fail. They now more closely matches what Test::More's `isnt()` outputs and
  makes much more sense.
* Added `domain_type_is()` and `domain_type_isnt()`. Based on an
  implementation by Bob Lunney.
* Added variants of `col_is_type()` to allow a schema name to be specified for
  the type.
* Added `display_type()` and replaced all internal use of `format_type()` with
  it. This is so that schemas are always displayed properly.
* Added `row_eq()`. It's a 90% solution that requires that `record` values be
  cast to existing composite values. But it should work well even at that.
* Changed the default descriptions output by `has_operator()`, `has_leftop()`,
  and `has_rightop()` to format the operator names more like the display of
  `regoperator` types.
* Fixed issue with using SET SESSION AUTHORIZATION among tests [Kevin Field]
* Added `display_oper()` so that schemas are always displayed properly when
  comparing operators.

0.22 2009-07-31T00:26:16Z
-------------------------
* Fixed failing test on 8.4rc2.
* Added result set testing functions. These allow testers to write queries in
  pure SQL and check that their results are as expected. The new functions
  are:
  + `results_eq()`
  + `results_ne()`
  + `set_eq()`
  + `bag_eq()`
  + `set_ne()`
  + `bag_ne()`
  + `set_has()`
  + `bag_has()`
  + `set_hasnt()`
  + `bag_hasnt()`
* Changed some internal queries to use explicit `JOIN`s.
* Fixed capitalization of `TAP::Harness` in the `Makefile`. Thanks to
  Quinn Weaver.
* Fixed a syntax error that caused `uninstall_pgtap.sql` to fail on older
  PostgreSQL versions (fix by Quinn Weaver).
* Modified the summary documentation for the `--schema` and `--match` options
  to `pg_prove` to better reflect that they are for finding xUnit tests,
  not for finding pgTAP functions.

0.21 2009-05-29T00:04:31Z
-------------------------
* Fixed a bug in the processing of the `--schema` and `--match` options that
  only shows up in Getopt::Long 2.38 or higher.
* A few doc and test typos fixed thanks to Gabrielle Roth.
* Fixed failing test on Solaris 10 on Intel thanks to Gabrielle Roth.
* Fixed a failing test for the version number string on 8.4 beta.
* Added `performs_ok()`.
* Added `has_language()`, `hasnt_language()`, and `language_is_trusted()`.
* Added `has_opclass()` and `hasnt_opclass()`.
* Added `has_rule()`, `hasnt_rule()`, `rule_is_instead()`, and `rule_is_on()`.
* Added `tablespaces_are()`, `schemas_are()`, `tables_are()`, `views_are()`,
  `sequences_are()`, `functions_are()`, `indexes_are()`, `users_are()`,
  `groups_are()`, `opclasses_are()`, `languages_are()`, and `rules_are()`.
* Added a `has_trigger(table, trigger, description)`. Note that this means
  that if you were previously using `has_trigger(schema, table, trigger)`, you
  will need to cast the third argument to `NAME` to get it working again.
* Changed `can_ok()` to `has_function()`, so that it's named like all of the
  other functions that check for the presence of database objects. The old
  `can_ok()` function is still available as an alias, but it emits a warning
  and will be removed in a future version of pgTAP.
* Added `hasnt_trigger()`, `hasnt_index()`, and `hasnt_function()`.
* Added `function_lang_is()`, `function_returns()`, `is_definer()`,
  `is_aggregate()`, `is_strict()`, and `volatility_is()`.

0.20 2009-03-29T19:05:40Z
-------------------------
* Changed the names of the functions tested in `sql/do_tap.sql` and
  `sql/runtests.sql` so that they are less likely to be ordered differently
  given varying collation orders provided in different locales and by
  different vendors. Reported by Ingmar Brouns.
* Added the `--formatter` and `--archive` options to `pg_prove`.
* Fixed the typos in `pg_prove` where the output of `--help` listed
  `--test-match` and `--test-schema` instead of `--match` and `--schema`.
* Added `has_cast()`, `hasnt_cast()`, and `cast_context_is()`.
* Fixed a borked function signature in `has_trigger()`.
* Added `has_operator()`, `has_leftop()`, and `has_rightop()`.
* Fixed a bug where the order of columns found for multicolumn indexes by
  `has_index()` could be wrong. Reported by Jeff Wartes. Thanks to Andrew
  Gierth for help fixing the query.

0.19 2009-02-21T02:09:26Z
-------------------------
* Added an alternate version of `col_default_is()` to better handle the common
  case when a default is specified as a string, such as a text or expression
  default. This means that you have to do a lot less casting of default values
  specified in the arguments to `col_default_is()`. It also restores the
  signature that 0.17 recommended be dropped.
* Added `has_role()` and `isnt_role()`.
* Added `has_user()`, `isnt_user()`, `is_superuser()`, and `isnt_superuser()`.
* Added `has_group()`, `isnt_group()`, and `is_member_of()`.
* Fixed syntax in the `Makefile` to make it compatible with older versions of
  `make`. Reported by Aaron Kangas.
* Improved the documentation of `runtests()` to make it clearer that all tests
  it runs are run in transactions that are rolled back after each test.
  Suggested by Aaron Kangas.
* Added the `--runtests`, `--schema`, and `--match` options to `pg_prove` so
  that it can be used to run xUnit-style test functions without an SQL script.
* Fixed a bug in `can()` where it could sometimes report that functions
  were available in a schema when in fact they were not.
* In the schema testing functions, removed the use of `quote_ident()` to
  compare all identifiers, such as table names, schema names, column names,
  etc, added in 0.17. This is because `quote_ident(a) = quote_ident(b)` can't
  give a different result than `a = b`, and besides, made things slower and
  prevented the use of indexes. Thanks to Andrew Gierth for the spot.
* The output from `pg_prove` now includes a list of failing tests when it is
  not run with `--verbose`. When using TAP::Harness 3.17 and later, it also
  shows comments and diagnostics in the non-verbose output. Verbose output
  still outputs everything.

0.18  2009-02-06T20:06:00Z
--------------------------
* Fixed `pg_version_num()`. It was broken in 0.16; sorry about that!
* Fixed a bug in `col_type_is()` where it would die if it was looking
  for a type in a table that had dropped columns. Thanks to depesz and
  RhodiumToad on #postgresql for the help nailing that one down!
* Fixed a test failure in `sql/coltap.sql` on PostgreSQL 8.3.
* Fixed a bug in the `Makefile` where it did not properly point to
  `pg_config` when building from the `compat` directory in the
  PostgreSQL source tree.
* Fixed a bug in the `Makefile` where the `test_setup.sql` file, which
  is required for tests, was not always getting created.

0.17  2009-02-06T17:51:52Z
--------------------------
* Fixed `col_default_is()` so that it works properly in cases where the
  default value is `NULL`.
* Fixed `col_default_is()` to gracefully handle different data types,
  columns without defaults, and nonexistent columns.
* Fixed the three-argument form of `col_default_is()` to accept any data type
  for the default, not just text. *NOTE:* You must `DROP FUNCTION
  col_default_is ( NAME, NAME, TEXT );` if you have previous versions of this
  function installed, so as to eliminate the incompatible version.
* Added `col_has_default()` and `col_hasnt_default()`.
* Changed default descriptions for column testing functions to refer to
  the columns as the more SQL-standard `schema.table.column` instead of
  `schema.table(column)`.
* Modified all diagnostic messages and test descriptions that reference
  database objects to use `quote_ident()` to quote the names of those
  objects, rather than emitting them with or without double-quotes in an
  ad-hoc way as had hitherto been the case.
* Modified `col_not_null()`, `col_is_null()` and `col_type_is()` to emit
  diagnostics if they fail because the column in question does not
  exist.
* Added `has_tablespace()` and `hasnt_tablespace()`.
* Added `has_sequence()` and `hasnt_sequence()`.
* Changed `has_index()` so that expressions are compared case-
  sensitively.
* In the schema testing functions, now using `quote_ident()` to compare
  all identifiers, such as table names, schema names, column names, etc.
  It had been a mix of straight-forward string comparison, and
  case-insensitive matching. This means that tests should use only
  lowercase strings to specify object names, unless mixed case objects
  were created with double quote characters.
* Added version of `todo()` with the `why` and `how_many` arguments
  reversed, so that I don't have to remember a specific order.

0.16  2009-02-03T17:37:03Z
--------------------------
* Switched from a crazy trinary logic in `is()` and `isnt()` to the use
  of `IS NOT DISTINCT FROM` and `IS DISTINCT FROM`. Swiped from PGUnit.
* Fixed documentation for the short version of `--help` in `pg_prove`.
  It's `-H`, not `-h`.
* Fixed a bug in the makefile that prevented OS detection from working
  properly on many platforms (including Linux, and probably many
  others). This allows `os_name()` to actually work on those platforms.
* The definition of the `pg_typeof()` function is removed when
  installing on PostgreSQL 8.4 and no C code is compiled, since the only
  C code is for this function. This is because `pg_typeof()` is included
  with PostgreSQL 8.4.
* Added `has_schema()` and `hasnt_schema()`.
* Added `has_type()`, `hasnt_type()`, `has_domain()`, `hasnt_domain()`,
  `has_enum()`, and `hasnt_enum()`.
* Added `enum_has_labels()`.
* Updated the `Makefile` to fix the shebang line in `pg_prove` if
  Perl is found, and to emit a warning if the `$PERL` variable is not
  set. - Updated the `Makefile` so that if `$PERL` is set and
  TAP::Harness is not installed, it warns the user to install it in
  order to use `pg_prove`. - Updated the `Makefile` to set the location
  of the `psql` binary to the bin directory specified for PostgreSQL
  when it was built.
* Fixed `pg_version_num()` to work on PostgreSQL 8.4devel.
* Fixed test failures on PostgreSQL 8.4devel.

0.15  2009-01-20T18:41:24Z
--------------------------
* Changed `pg_typeof()` from immutable to stable, in compliance with its
  configuration in the forthcoming PostgreSQL 8.4.
* Added `do_tap()`, which finds test functions and runs them, allowing
  tap tests to be embedded in functions and then executed all at once.
* Added `runtests()`, which introduces xUnit-style test running, for
  those who prefer to put their tests in a slew of test functions and
  then run one function to run all of the test functions (plus startup,
  setup, teardown, and shutdown tests).
* Added `findfuncs()`, which is used by `do_tap()` to find the functions
  to execute.
* The `col_type_is()` function now requires that the type be visible in
  the search path when it is called without a schema argument.
* The `plan()` function no longer resets the `client_min_messages`
  setting to its default value, but leaves it set to whatever it was set
  to before the function was called.
* Fixed a typo in the documentation of the short version of the
  `--version` option to `pg_prove`, which is `-V`, not `-v`.
* Added `pgtap_version()`.
* Added the "html" make target to generate the HTML documentation for
  the Web site, including a table of contents and documentation for
  `pg_prove`.

0.14  2008-10-27T22:43:36Z
--------------------------
* Added `SET search_path` statements to `uninstall_pgtap.sql.in` so that
  it will work properly when TAP is installed in its own schema. Thanks to
  Ben for the catch!
* Added commands to drop `pg_version()` and `pg_version_num()`
  to`uninstall_pgtap.sql.in`.
* Added `has_index()`, `index_is_unique()`, `index_is_primary()`,
  `is_clustered()`, and `index_is_type()`.
* Added `os_name()`. This is somewhat experimental. If you have `uname`,
  it's probably correct, but assistance in improving OS detection in the
  `Makefile` would be greatly appreciated. Notably, it does not detect
  Windows.
* Made `ok()` smarter when the test result is passed as `NULL`. It was
  dying, but now it simply fails and attaches a diagnostic message
  reporting that the test result was `NULL`. Reported by Jason Gordon.
* Fixed an issue in `check_test()` where an extra character was removed
  from the beginning of the diagnostic output before testing it.
* Fixed a bug comparing `name[]`s on PostgreSQL 8.2, previously hacked
  around.
* Added `has_trigger()` and `trigger_is()`.
* Switched to pure SQL implementations of the `pg_version()` and
  `pg_version_num()` functions, to simplify including pgTAP in module
  distributions.
* Added a note to `README.pgtap` about the need to avoid `pg_typeof()`
  and `cmp_ok()` in tests run as part of a distribution.

0.13  2008-10-13T19:09:46Z
--------------------------
* Added `pg_version()` and `pg_version_num()`.
* Documented `pg_typeof()`.

0.12  2008-10-11T04:02:42Z
--------------------------
* Updated `plan()` to disable warnings while it creates its tables.
  This means that `plan()` no longer send NOTICE messages when they run,
  although tests still might, depending on the setting of
  `client_min_messages`.
* Added `hasnt_table()`, `hasnt_view()`, and `hasnt_column()`.
* Added `hasnt_pk()`, `hasnt_fk()`, `col_isnt_pk()`, and
  `col_isnt_fk()`.
* Added missing `DROP` statements to `uninstall_pgtap.sql.in`.

0.11  2008-09-24T20:41:42Z
--------------------------
* Simplified the tests so that they now load `test_setup.sql` instead of
  setting a bunch of stuff themselves. Now only `test_setup.sql` needs
  to be created from `test_setup.sql.in`, and the other `.sql` files
  depend on it, meaning that one no longer has to specify `TAPSCHEMA`
  for any `make` target other than the default.
* Eliminated all uses of `E''` in the tests, so that we don't have to
  process them for testing on 8.0.
* Fixed the spelling of `ON_ROLLBACK` in the test setup. Can't believe I
  had it with one L in all of the test files before! Thanks to Curtis
  "Ovid" Poe for the spot.
* Added a couple of variants of `todo()` and `skip()`, since I can never
  remember whether the numeric argument comes first or second. Thanks to
  PostgreSQL's functional polymorphism, I don't have to. Also, there are
  variants where the numeric value, if not passed, defaults to 1.
* Updated the link to the pgTAP home page in `pgtap.sql.in`.
* TODO tests can now nest.
* Added `todo_start()`, `todo_end()`, and `in_todo()`.
* Added variants of `throws_ok()` that test error messages as well as
  error codes.
* Converted some more tests to use `check_test()`.
* Added `can()` and `can_ok()`.
* Fixed a bug in `check_test()` where the leading white space for
  diagnostic messages could be off by 1 or more characters.
* Fixed the `installcheck` target so that it properly installs PL/pgSQL
  into the target database before the tests run.

0.10  2008-09-18T22:59:31Z
--------------------------
* Changed `pg_prove` to set `QUIET=1` when it runs, so as to prevent the
  output of extraneous stuff.
* Added some `GRANT` statements to `plan()` in order to make it easier
  to run tests using roles other than the one that called `plan()`.
  Patch from Rod Taylor.
* Replaced a call to `lastval()` with a call to `currval()` in order to
  improve compatibility with PostgreSQL 8.0. Reported by David Westbrook.
* Added support for TODO and SKIP tests.
* Removed the few uses of named parameters and added alias names instead.
  This improves compatibility for versions of PostgreSQL prior to 8.0.
  Reported by David Westbrook.
* Fixed the plural of "test" in the output from `finish()`. It was
  plural when there was one test and singular otherwise. So I reversed
  that.
* Moved assignment to declared variables from the `DECLARE` block to the
  body of the functions. Improves compatibility with versions of
  PostgreSQL prior to 8.0. Reported by David Westbrook.
* Eliminated passing the result of a function call to `RAISE` in order
  to better support older versions of PostgreSQL. Reported by David
  Westbrook.
* Switched from using `FOUND` to `GET DIAGNOSTICS ROW_COUNT` because the
  former does not seem to be set for `EXECUTE` SQL in older versions of
  PostgreSQL. Reported by David Westbrook.
* Added tests specifically targeting PostgreSQL 7.3. From David
  Westbrook.
* Ported all the way back to PostgreSQL 8.0, thanks to some Makefile
  hackery and a patch file. All tests pass on 8.0, 8.1, 8.2, and 8.3.
  The only exception is `throws_ok()` and `lives_ok()`, which are not
  supported on 8.0. Versions of PostgreSQL lower than 8.0 are not yet
  supported, even though we have made some changes to simplify getting
  things to work in earlier versions.
* Changed "got/expected" to "have/want", following Schwern's plans for
  Test::Builder 2. Also changed "caught/expected" to "caught/wanted" for
  nice parity when outputting diagnostics for exception testing.
* Added the `has_table()`, `has_view()`, `has_column()`, `has_pk()`,
  `has_fk()`, `fk_ok()`, `has_unique()` and `has_check()` test
  functions.
* Added the `col_not_null()` and `col_is_null()` test functions.
* Added the `col_type_is()`, `col_default_is()`, `col_is_pk()`,
  `col_is_fk()`, `col_is_unique()`, and `col_has_check()` test
  functions.
* Fixed `is()` and `isnt()` to better handle `NULL`s.
* Added `cmp_ok()`.
* Added the `--set ON_ERROR_STOP=1` option to the call to `psql` in
  `pg_prove`.
* Renamed `drop_pgtap.sql.in` to `uninstall_pgtap.sql.in`, which is more
  in line with typical PostgreSQL contrib modules.
* Removed verbose output of the `psql` command from `pg_prove`. Thanks
  to Andy Armstrong for the spot.
* Divided the tests up into many separate test script files, so that
  things are a bit better organized and easier to maintain.
* Added the `check_test()` function and started converting internal
  tests to use it instead of the hacked stuff they were doing before.
* As in Test::Builder 0.81_01, changed the message for extra tests run
  to show the number of tests run rather than the number extra to avoid
  the user having to do mental math.
* The regression test files are now processed by `make installcheck` and
  `make test` so that the schema can be properly set, if pgTAP is built
  with a schema.

0.02  2008-06-17T16:26:41Z
--------------------------
* Converted the documentation to Markdown.
* Added pg_prove, a Perl script that use TAP::Harness to run tests and
  report the results, just like the Perl program `prove`.
* Fixed `no_plan()` so that it no longer emits a plan, which apparently
  was wrong. Now `finish()` outputs it when there's no plan.
* Fixed the test script so that it now emits a proper plan.
* Removed all of the configuration settings from `pgtap.sql`, as they're
  now handled by `pg_prove`. I've mentioned them in the README for
  reference.
* Added `lives_ok()`.
* Moved the creation of temporary tables into `plan()`, so that
  everything is properly self-contained.
* Improved the handling of transactions. Test scripts are now assumed to
  be single transactions with a ROLLBACK at the end. This makes it so
  that test scripts don't have to include `drop_pgtap.sql`.
* Updated `pg_prove` to rollback on an error, rather than just stop.
  This allows all test functions to be properly rolled back, too, in a
  test script that includes them but then encounters an unhandled
  exception.
* Updated `pg_prove` to emit an appropriate error message if no test
  scripts are specified.
* Added a Makefile. It uses the typical PostgreSQL installation
  procedures to install pgTAP. The SQL files have now been turned into
  `.in` templates that are processed by `make`.
* Added support for schema qualification of test functions. Just set the
  `$TAPSCHEMA` environment variable when running `make`.
* Added support for standard PostgreSQL-type regression testing by just
  copying the test script, setting some variables inside it, and
  providing an `expected/` directory. The test now lives in the `sql/`
  directory.
* Changed all instances of `RETURN QUERY SELECT` to `RETURN NEXT`, which
  should allow pgtap to run on versions of PostgreSQL earlier than 8.3.
  Thanks to Neil Conway for the suggestion.

0.01  2008-06-07T05:24:27Z
--------------------------
* Initial public release. Announcement at
  https://justatheory.com/2008/06/introducing-pgtap/