Version 4.0 - V.I. Warshawski
What's Changed
Breaking Changes
- tts::reverse_ramp is a ramp read backward: for a size N, its values are start+(N-1)*step down to start, so the last one is start and none of them falls below it. It used to count down from start, which put reverse_ramp{0} entirely on negative values and made it unusable for anything indexed from zero.
- TTS_ABSOLUTE_EQUAL, TTS_RELATIVE_EQUAL and TTS_ULP_EQUAL now require both operands to have the same type. They used to compare them after promoting both to their common type, which measured the distance in the unit of the promoted type rather than the one under test: comparing a float against its own value as a double reported 9.84e+07 ULP instead of 0. A mismatch is now a compile-time error naming both types; convert the expected value at the call site.
- TTS_RELATIVE_EQUAL and TTS_ALL_RELATIVE_EQUAL report a ratio where they used to report a percentage, and the unit they print is rel rather than %. abs(L-R) / max(abs(L), abs(R), 1) is compared to the tolerance directly, so every existing tolerance has to be divided by a hundred. The factor was the odd one out: ulp and absolute already reported the quantity itself, and a consumer overloading the hook naturally returns the quotient; one that did was testing a hundred times looser than its tolerances read.
- A relative tolerance of one or more is reported as a failure rather than compared: it accepts any error up to the value itself, and is a percentage left over from the previous unit. The message says to divide it by a hundred. This applies to the native floating point path only; a tts::precision specialisation keeps whatever unit it returns.
- Equality and ordering checks are unchanged. TTS_EQUAL and its typed counterparts forward to operator== or to a compare_equal overload, and introduce no conversion of their own.
- TTS_BIT_EQUAL and TTS_BIT_NOT_EQUAL go through the bit_equal member of tts::comparison<L, R> rather than calling memcmp themselves. The default is that same memcmp, so an unspecialized type behaves as before; a type whose storage holds padding can now say which bits carry a value.
- Every customization point is a trait, and the free functions they replace are gone. tts::precision<T>, tts::display<T>, tts::comparison<L, R>, tts::generation<T> and tts::conversion<T, V> stand where ulp_distance, relative_distance, absolute_distance, ieee_equal, to_text, compare_equal, compare_less, produce and convert_as used to be reached by name. A specialization writes the members it changes and leaves the rest out: every member has a default, and a member the specialization does not carry keeps it. A member that is there but does not accept the operands is reported rather than ignored.
- operator<< is not consulted when a value is rendered, and has not been since 3.0, which dropped that path along with <sstream> without saying so. Up to 2.2, as_text tried an ostream insertion before falling back on the element dump for a range and on the byte dump for anything else. A type that used to print through its own operator<< now prints as a byte dump, and one that also exposes begin and end prints element by element, which fails to compile when its iterator is a proxy. Specialise tts::display<T> for such a type. Bringing the path back costs 13% of median compile memory across the suite, measured, which is what 3.0 bought by removing it.
- An overload of one of the seven customization names left behind is reported where it would have been used, rather than ignored in silence. produce and convert_as are the exception: they are the dispatchers rather than customization points, and TTS reaches them qualified, so an overload of either name in another namespace is never found.
- The built-in sinks come with #include <tts/tts.hpp> only in the translation unit that defines TTS_MAIN. Any other translation unit that builds one includes <tts/sinks/sinks.hpp>, and no longer parses the driver it does not run.
- <tts/tts.hpp> no longer brings <array> along. A test that uses std::array includes it.
New Features
- tts::is_randoms and its tts::is_randoms_v alias name the range generator by its type. A tts::generation specialization keys on the type being built and cannot see the generator in an overload set, so a range that has to be drawn apart is recognized this way.
Version 3.0 - Beatrice Adela Bradley
What's Changed
A large body of work (https://github.com/jfalcou/tts/pull/105) has been done to make TTS compile-time far better. This affects the user API a tiny bit, hence the bump to TTS v3.0.
Infrastructure
Bug Fixes
New Features
New Contributors
Detective who?
Beatrice Adela Bradley.
Version 2.2 - Aubrey Bradimore
What's Changed
Infrastructure
- We are now under the Boost Software Licence 1.0
- Enable Android tests and support
- Clarify some output when failures occur
New Features
- Implement ALL_EQUAL
- Test can now use configurable, automatic data generator (https://github.com/jfalcou/tts/pull/53)
- TTS_EXPECT_COMPILES/COMPILE_NOT check if a fragment of code can be compiled or not
- Reimplement TTS_WHEN and TTS_AND_THEN to support sub-tests
- Add typed tests that checks both value and exact type
Bug Fixes
- Fix Visual Studio interaction with constexpr tests
- Improve random generators and seed management
Detective who?
Aubrey Bradimore, this is a Christmas Release afterall
Version 2.1 - Cordelia Gray
What's Changed
- Fix #42 - Proper FetchContent support
- Better compatibility
- Enable EMSCRIPTEN tests and supports
- Add tests over sequences
Detective who?
Cordelia Gray, a very suitable detective
Version 2.0 - Kay Scarpetta
TTS version 2.0 is taking a major break from an earlier version. Compile-time was creeping upward and some features were not that useful. So the code base got streamlined and simplified.
This version is a Major Release With API Changes release:
Breaking API Changes:
- Removal of support for colors
- Removal of support for sub-scenario
- Silent mode for tests is the default. The only information displayed concerns failures
Features:
- Improved compile-times
- Output streamlined to fit machine processing in CI context or scripts
- Better handling of template test cases: less generated symbol, better flexibility in inputs (list of types or types list generator)
Detective who?
Kay Scarpetta, the private forensic consultant.
Version 1.0 - Phoebe Daring
This version is a Major Release With API Changes release:
Breaking API Changes:
To support some of the new feature and simplify some macros, TTS API is now based on a lambda system. If this doesn't change the way tests are performed, it now requires to end all tests cases by a ;.
Bug Fixes:
- Fix #3 - TTS_TPL_CASES now use meta-function based system for types generation
- Fix #32 - Provide a way to log arbitrary data when failures occur
- Fix #33 - Provide the REQUIRED option on tests to halt tests at first failure
- Fix #34 - Provide a command-line option to fix precision when displaying floating points
- Fix #37 - Stop displaying char const* as a string.
Detective who?
Phoebe Daring, the first half of the Daring Twins since 1912.
Version 0.2 - Kate Fansler
This version is a fix+features release:
- Fix #20 - Make string and pointers display in a more intuitive way
- Fix #23 - Add proper parens in EXPECT macro
- Add support for constexpr expectation and relational tests
- TTS_CASE_TPL now generates one scenario per type
- Fix #21 - Add runtime filtering for tests based on description string
Detective who?
Kate Fansler, solving academic crimes since 1964.
Version 0.1 - Cadfael
First public release.
TTS first complete release enables numerically oriented TDD including:
- precision testing
- checks over the data set
- easy to customize the use of user-defined types.
Detective who?
Cadfael is probably the eldest of all amateur detectives in history.
