This commit also forbids composing lvalues of composed matchers, as per previous deprecation notice. I do not expect this to be contentious in practice, because there was a bug in that usage for years, and nobody complained.
49 KiB
Release notes
Contents
2.11.1
2.11.0
2.10.2
2.10.1
2.10.0
2.9.2
2.9.1
2.9.0
2.8.0
2.7.2
2.7.1
2.7.0
2.6.1
2.6.0
2.5.0
2.4.2
2.4.1
2.4.0
2.3.0
2.2.3
2.2.2
2.2.1
2.2.0
2.1.2
2.1.1
2.1.0
2.0.1
Older versions
Even Older versions
3.0.0 (in progress)
(Potentially) Breaking changes
ANON_TEST_CASE
has been removed, useTEST_CASE
with no arguments instead (#1220)--list*
commands no longer have non-zero return code (#1410)--list-test-names-only
has been removed (#1190)- You should use verbosity-modifiers for
--list-tests
instead
- You should use verbosity-modifiers for
--list*
commands are now piped through the reporters- The top-level reporter interface provides default implementation that works just as the old one
- XmlReporter outputs a machine-parseable XML
TEST_CASE
description support has been removed- If the second argument has text outside tags, the text will be ignored.
- Hidden test cases are no longer included just because they don't match an exclusion tag
- Previously, a
TEST_CASE("A", "[.foo]")
would be included by asking for~[bar]
.
- Previously, a
PredicateMatcher
is no longer type erased.- This means that the type of the provided predicate is part of the
PredicateMatcher
's type
- This means that the type of the provided predicate is part of the
SectionInfo
no longer contains section description as a member (#1319)- You can still write
SECTION("ShortName", "Long and wordy description")
, but the description is thrown away - The description type now must be a
const char*
or implicitly convertible to it
- You can still write
- The
[!hide]
tag has been removed.- Use
[.]
or[.foo]
instead.
- Use
- Lvalues of composed matchers cannot be composed further
Fixes
- The
INFO
macro no longer contains superfluous semicolon (#1456) - The
--list*
family of command line flags now return 0 on success (#1410, #1146)
Other changes
CATCH_CONFIG_DISABLE_MATCHERS
no longer exists.- If you do not want to use Matchers in a TU, do not include their header.
CATCH_CONFIG_ENABLE_CHRONO_STRINGMAKER
no longer exists.StringMaker
specializations for are always provided
2.11.1
Improvements
- Breaking into debugger is supported on iOS (#1817)
google-build-using-namespace
clang-tidy warning is suppressed (#1799)
Fixes
- Clang on Windows is no longer assumed to implement MSVC's traditional preprocessor (#1806)
ObjectStorage
now behaves properly inconst
contexts (#1820)GENERATE_COPY(a, b)
now compiles properly (#1809, #1815)- Some more cleanups in the benchmarking support
2.11.0
Improvements
- JUnit reporter output now contains more details in case of failure (#1347, #1719)
- Added SonarQube Test Data reporter (#1738)
- It is in a separate header, just like the TAP, Automake, and TeamCity reporters
range
generator now allows floating point numbers (#1776)- Reworked part of internals to increase throughput
Fixes
- The single header version should contain full benchmarking support (#1800)
[.foo]
is now properly parsed as[.][foo]
when used on the command line (#1798)- Fixed compilation of benchmarking on platforms where
steady_clock::period
is notstd::nano
(#1794)
2.10.2
Improvements
- Catch2 will now compile on platform where
INFINITY
is double (#1782)
Fixes
- Warning suppressed during listener registration will no longer leak
2.10.1
Improvements
- Catch2 now guards itself against
min
andmax
macros fromwindows.h
(#1772) - Templated tests will now compile with ICC (#1748)
WithinULP
matcher now uses scientific notation for stringification (#1760)
Fixes
- Templated tests no longer trigger
-Wunused-templates
(#1762) - Suppressed clang-analyzer false positive in context getter (#1230, #1735)
Miscellaneous
- CMake no longer prohibits in-tree build when Catch2 is used as a subproject (#1773, #1774)
2.10.0
Fixes
TEMPLATE_LIST_TEST_CASE
now properly handles non-copyable and non-movable types (#1729)- Fixed compilation error on Solaris caused by a system header defining macro
TT
(#1722, #1723) REGISTER_ENUM
will now fail at compilation time if the registered enum is too large- Removed use of
std::is_same_v
in C++17 mode (#1757) - Fixed parsing of escaped special characters when reading test specs from a file (#1767, #1769)
Improvements
- Trailing and leading whitespace in test/section specs are now ignored.
- Writing to Android debug log now uses
__android_log_write
instead of__android_log_print
- Android logging support can now be turned on/off at compile time (#1743)
- The toggle is
CATCH_CONFIG_ANDROID_LOGWRITE
- The toggle is
- Added a generator that returns elements of a range
- Use via
from_range(from, to)
orfrom_range(container)
- Use via
- Added support for CRTs that do not provide
std::nextafter
(#1739)- They must still provide global
nextafter{f,l,}
- Enabled via
CATCH_CONFIG_GLOBAL_NEXTAFTER
- They must still provide global
- Special cased
Approx(inf)
not to match non-infinite values- Very strictly speaking this might be a breaking change, but it should match user expectations better
- The output of benchmarking through the Console reporter when
--benchmark-no-analysis
is set is now much simpler (#1768) - Added a matcher that can be used for checking an exceptions message (#1649, #1728)
- The matcher helper function is called
Message
- The exception must publicly derive from
std::exception
- The matching is done exactly, including case and whitespace
- The matcher helper function is called
- Added a matcher that can be used for checking relative equality of floating point numbers (#1746)
- Unlike
Approx
, it considers both sides when determining the allowed margin - Special cases
NaN
andINFINITY
to match user expectations - The matcher helper function is called
WithinRel
- Unlike
- The ULP matcher now allows for any possible distance between the two numbers
- The random number generators now use Catch-global instance of RNG (#1734, #1736)
- This means that nested random number generators actually generate different numbers
Miscellaneous
- In-repo PNGs have been optimized to lower overhead of using Catch2 via git clone
- Catch2 now uses its own implementation of the URBG concept
- In the future we also plan to use our own implementation of the distributions from
<random>
to provide cross-platform repeatability of random results
- In the future we also plan to use our own implementation of the distributions from
2.9.2
Fixes
ChunkGenerator
can now be used with chunks of size 0 (#1671)- Nested subsections are now run properly when specific section is run via the
-c
argument (#1670, #1673) - Catch2 now consistently uses
_WIN32
to detect Windows platform (#1676) TEMPLATE_LIST_TEST_CASE
now support non-default constructible type lists (#1697)- Fixed a crash in the XMLReporter when a benchmark throws exception during warmup (#1706)
- Fixed a possible infinite loop in CompactReporter (#1715)
- Fixed
-w NoTests
returning 0 even when no tests were matched (#1449, #1683, #1684) - Fixed matcher compilation under Obj-C++ (#1661)
Improvements
RepeatGenerator
andFixedValuesGenerator
now fail to compile when used withbool
(#1692)- Previously they would fail at runtime.
- Catch2 now supports Android's debug logging for its debug output (#1710)
- Catch2 now detects and configures itself for the RTX platform (#1693)
- You still need to pass
--benchmark-no-analysis
if you are using benchmarking under RTX
- You still need to pass
- Removed a "storage class is not first" warning when compiling Catch2 with PGI compiler (#1717)
Miscellaneous
- Documentation now contains indication when a specific feature was introduced (#1695)
- These start with Catch2 v2.3.0, (a bit over a year ago).
docs/contributing.md
has been updated to provide contributors guidance on how to add these to newly written documentation
- Various other documentation improvements
- ToC fixes
- Documented
--order
and--rng-seed
command line options - Benchmarking documentation now clearly states that it requires opt-in
- Documented
CATCH_CONFIG_CPP17_OPTIONAL
andCATCH_CONFIG_CPP17_BYTE
macros - Properly documented built-in vector matchers
- Improved
*_THROWS_MATCHES
documentation a bit
- CMake config file is now arch-independent even if
CMAKE_SIZEOF_VOID_P
is in CMake cache (#1660) CatchAddTests
now properly escapes[
and]
in test names (#1634, #1698)- Reverted
CatchAddTests
adding tags as CTest labels (#1658)- The script broke when test names were too long
- Overwriting
LABELS
caused trouble for users who set them manually - CMake does not let users append to
LABELS
if the test name has spaces
2.9.1
Fixes
- Fix benchmarking compilation failure in files without
CATCH_CONFIG_EXTERNAL_INTERFACES
(or implementation)
2.9.0
Improvements
- The experimental benchmarking support has been replaced by integrating Nonius code (#1616)
- This provides a much more featurefull micro-benchmarking support.
- Due to the compilation cost, it is disabled by default. See the documentation for details.
- As far as backwards compatibility is concerned, this feature is still considered experimental in that we might change the interface based on user feedback.
WithinULP
matcher now shows the acceptable range (#1581)- Template test cases now support type lists (#1627)
2.8.0
Improvements
- Templated test cases no longer check whether the provided types are unique (#1628)
- This allows you to e.g. test over
uint32_t
,uint64_t
, andsize_t
without compilation failing
- This allows you to e.g. test over
- The precision of floating point stringification can be modified by user (#1612, #1614)
- We now provide
REGISTER_ENUM
convenience macro for generatingStringMaker
specializations for enums- See the "String conversion" documentation for details
- Added new set of macros for template test cases that enables the use of NTTPs (#1531, #1609)
- See "Test cases and sections" documentation for details
Fixes
UNSCOPED_INFO
macro now has a prefixed/disabled/prefixed+disabled versions (#1611)- Reporting errors at startup should no longer cause a segfault under certain circumstances (#1626)
Miscellaneous
- CMake will now prevent you from attempting in-tree build (#1636, #1638)
- Previously it would break with an obscure error message during the build step
2.7.2
Improvements
- Added an approximate vector matcher (#1499)
Fixes
- Filters will no longer be shown if there were none
- Fixed compilation error when using Homebrew GCC on OS X (#1588, #1589)
- Fixed the console reporter not showing messages that start with a newline (#1455, #1470)
- Modified JUnit reporter's output so that rng seed and filters are reported according to the JUnit schema (#1598)
- Fixed some obscure warnings and static analysis passes
Miscellaneous
- Various improvements to
ParseAndAddCatchTests
(#1559, #1601)- When a target is parsed, it receives
ParseAndAddCatchTests_TESTS
property which summarizes found tests - Fixed problem with tests not being found if the
OptionalCatchTestLauncher
variables is used - Including the script will no longer forcefully modify
CMAKE_MINIMUM_REQUIRED_VERSION
- CMake object libraries are ignored when parsing to avoid needless warnings
- When a target is parsed, it receives
CatchAddTests
now adds test's tags to their CTest labels (#1600)- Added basic CPack support to our build
2.7.1
Improvements
- Reporters now print out the filters applied to test cases (#1550, #1585)
- Added
GENERATE_COPY
andGENERATE_REF
macros that can use variables inside the generator expression- Because of the significant danger of lifetime issues, the default
GENERATE
macro still does not allow variables
- Because of the significant danger of lifetime issues, the default
- The
map
generator helper now deduces the mapped return type (#1576)
Fixes
- Fixed ObjC++ compilation (#1571)
- Fixed test tag parsing so that
[.foo]
is now parsed as[.][foo]
. - Suppressed warning caused by the Windows headers defining SE codes in different manners (#1575)
2.7.0
Improvements
TEMPLATE_PRODUCT_TEST_CASE
now uses the resulting type in the name, instead of the serial number (#1544)- Catch2's single header is now strictly ASCII (#1542)
- Added generator for random integral/floating point types
- The types are inferred within the
random
helper
- The types are inferred within the
- Added back RangeGenerator (#1526)
- RangeGenerator returns elements within a certain range
- Added ChunkGenerator generic transform (#1538)
- A ChunkGenerator returns the elements from different generator in chunks of n elements
- Added
UNSCOPED_INFO
(#415, #983, #1522)- This is a variant of
INFO
that lives until next assertion/end of the test case.
- This is a variant of
Fixes
- All calls to C stdlib functions are now
std::
qualified (#1541)- Code brought in from Clara was also updated.
- Running tests will no longer open the specified output file twice (#1545)
- This would cause trouble when the file was not a file, but rather a named pipe
- Fixes the CLion/Resharper integration with Catch
- Fixed
-Wunreachable-code
occurring with (old) ccache+cmake+clang combination (#1540) - Fixed
-Wdefaulted-function-deleted
warning with Clang 8 (#1537) - Catch2's type traits and helpers are now properly namespaced inside
Catch::
(#1548) - Fixed std{out,err} redirection for failing test (#1514, #1525)
- Somehow, this bug has been present for well over a year before it was reported
Contrib
ParseAndAddCatchTests
now properly escapes commas in the test name
2.6.1
Improvements
- The JUnit reporter now also reports random seed (#1520, #1521)
Fixes
- The TAP reporter now formats comments with test name properly (#1529)
CATCH_REQUIRE_THROWS
's internals were unified withREQUIRE_THROWS
(#1536)- This fixes a potential
-Wunused-value
warning when used
- This fixes a potential
- Fixed a potential segfault when using any of the
--list-*
options (#1533, #1534)
2.6.0
With this release the data generator feature is now fully supported.
Improvements
- Added
TEMPLATE_PRODUCT_TEST_CASE
(#1454, #1468)- This allows you to easily test various type combinations, see documentation for details
- The error message for
&&
and||
inside assertions has been improved (#1273, #1480) - The error message for chained comparisons inside assertions has been improved (#1481)
- Added
StringMaker
specialization forstd::optional
(#1510) - The generator interface has been redone once again (#1516)
- It is no longer considered experimental and is fully supported
- The new interface supports "Input" generators
- The generator documentation has been fully updated
- We also added 2 generator examples
Fixes
- Fixed
-Wredundant-move
on newer Clang (#1474) - Removed unreachable mentions
std::current_exception
,std::rethrow_exception
in no-exceptions mode (#1462)- This should fix compilation with IAR
- Fixed missing
<type_traits>
include (#1494) - Fixed various static analysis warnings
- Unrestored stream state in
XmlWriter
(#1489) - Potential division by zero in
estimateClockResolution
(#1490) - Uninitialized member in
RunContext
(#1491) SourceLineInfo
move ops are now markednoexcept
CATCH_BREAK_INTO_DEBUGGER
is now always a function
- Unrestored stream state in
- Fix double run of a test case if user asks for a specific section (#1394, #1492)
- ANSI colour code output now respects
-o
flag and writes to the file as well (#1502) - Fixed detection of
std::variant
support for compilers other than Clang (#1511)
Contrib
ParseAndAddCatchTests
has learned how to useDISABLED
CTest property (#1452)ParseAndAddCatchTests
now works when there is a whitspace before the test name (#1493)
Miscellaneous
- We added new issue templates for reporting issues on GitHub
contributing.md
has been updated to reflect the current test status (#1484)
2.5.0
Improvements
- Added support for templated tests via
TEMPLATE_TEST_CASE
(#1437)
Fixes
- Fixed compilation of
PredicateMatcher<const char*>
by removing partial specialization ofMatcherMethod<T*>
- Listeners now implicitly support any verbosity (#1426)
- Fixed compilation with Embarcadero builder by introducing
Catch::isnan
polyfill (#1438) - Fixed
CAPTURE
asserting for non-trivial captures (#1436, #1448)
Miscellaneous
- We should now be providing first party Conan support via https://bintray.com/catchorg/Catch2 (#1443)
- Added new section "deprecations and planned changes" to the documentation
- It contains summary of what is deprecated and might change with next major version
- From this release forward, the released headers should be pgp signed (#430)
- KeyID
E29C 46F3 B8A7 5028 6079 3B7D ECC9 C20E 314B 2360
- or https://codingnest.com/files/horenmar-publickey.asc
- KeyID
2.4.2
Improvements
- XmlReporter now also outputs the RNG seed that was used in a run (#1404)
Catch::Session::applyCommandLine
now also acceptswchar_t
arguments.- However, Catch2 still does not support unicode.
- Added
STATIC_REQUIRE
macro (#1356, #1362) - Catch2's singleton's are now cleaned up even if tests are run (#1411)
- This is mostly useful as a FP prevention for users who define their own main.
- Specifying an invalid reporter via
-r
is now reported sooner (#1351, #1422)
Fixes
- Stringification no longer assumes that
char
is signed (#1399, #1407)- This caused a
Wtautological-compare
warning.
- This caused a
- SFINAE for
operator<<
no longer sees different overload set than the actual insertion (#1403)
Contrib
catch_discover_tests
correctly adds tests with comma in name (#1327, #1409)- Added a new customization point in how the tests are launched to
catch_discover_tests
2.4.1
Improvements
- Added a StringMaker for
std::(w)string_view
(#1375, #1376) - Added a StringMaker for
std::variant
(#1380)- This one is disabled by default to avoid increased compile-time drag
- Added detection for cygwin environment without
std::to_string
(#1396, #1397)
Fixes
UnorderedEqualsMatcher
will no longer accept erroneously accept vectors that share suffix, but are not permutation of the desired vector- Abort after (
-x N
) can no longer be overshot by nestedREQUIRES
and subsequently ignored (#1391, #1392)
2.4.0
This release brings two new experimental features, generator support
and a -fno-exceptions
support. Being experimental means that they
will not be subject to the usual stability guarantees provided by semver.
Improvements
- Various small runtime performance improvements
CAPTURE
macro is now variadic- Added
AND_GIVEN
macro (#1360) - Added experimental support for data generators
- See their documentation for details
- Added support for compiling and running Catch without exceptions
- Doing so limits the functionality somewhat
- Look into the documentation for details
Fixes
- Suppressed
-Wnon-virtual-dtor
warnings in Matchers (#1357) - Suppressed
-Wunreachable-code
warnings in floating point matchers (#1350)
CMake
- It is now possible to override which Python is used to run Catch's tests (#1365)
- Catch now provides infrastructure for adding tests that check compile-time configuration
- Catch no longer tries to install itself when used as a subproject (#1373)
- Catch2ConfigVersion.cmake is now generated as arch-independent (#1368)
- This means that installing Catch from 32-bit machine and copying it to 64-bit one works
- This fixes conan installation of Catch
2.3.0
This release changes the include paths provided by our CMake and
pkg-config integration. The proper include path for the single-header
when using one of the above is now <catch2/catch.hpp>
. This change
also necessitated changes to paths inside the repository, so that the
single-header version is now at single_include/catch2/catch.hpp
, rather
than single_include/catch.hpp
.
Fixes
- Fixed Objective-C++ build
-Wunused-variable
suppression no longer leaks from Catch's header under Clang- Implementation of the experimental new output capture can now be disabled (#1335)
- This allows building Catch2 on platforms that do not provide things like
dup
ortmpfile
.
- This allows building Catch2 on platforms that do not provide things like
- The JUnit and XML reporters will no longer skip over successful tests when running without
-s
(#1264, #1267, #1310)- See improvements for more details
Improvements
- pkg-config and CMake integration has been rewritten
- If you use them, the new include path is
#include <catch2/catch.hpp>
- CMake installation now also installs scripts from
contrib/
- For details see the new documentation
- If you use them, the new include path is
- Reporters now have a new customization point,
ReporterPreferences::shouldReportAllAssertions
- When this is set to
false
and the tests are run without-s
, passing assertions are not sent to the reporter. - Defaults to
false
.
- When this is set to
- Added
DYNAMIC_SECTION
, a section variant that constructs its name using stream- This means that you can do
DYNAMIC_SECTION("For X := " << x)
.
- This means that you can do
2.2.3
To fix some of the bugs, some behavior had to change in potentially breaking manner. This means that even though this is a patch release, it might not be a drop-in replacement.
Fixes
- Listeners are now called before reporter
- This was always documented to be the case, now it actually works that way
- Catch's commandline will no longer accept multiple reporters
- This was done because multiple reporters never worked properly and broke things in non-obvious ways
- This has potential to be a breaking change
- MinGW is now detected as Windows platform w/o SEH support (#1257)
- This means that Catch2 no longer tries to use POSIX signal handling when compiled with MinGW
- Fixed potential UB in parsing tags using non-ASCII characters (#1266)
- Note that Catch2 still supports only ASCII test names/tags/etc
TEST_CASE_METHOD
can now be used on classnames containing commas (#1245)- You have to enclose the classname in extra set of parentheses
- Fixed insufficient alt stack size for POSIX signal handling (#1225)
- Fixed compilation error on Android due to missing
std::to_string
in C++11 mode (#1280) - Fixed the order of user-provided
FALLBACK_STRINGIFIER
in stringification machinery (#1024)- It was intended to be replacement for built-in fallbacks, but it was used after them.
- This has potential to be a breaking change
- Fixed compilation error when a type has an
operator<<
with templated lhs (#1285, #1306)
Improvements
- Added a new, experimental, output capture (#1243)
- This capture can also redirect output written via C apis, e.g.
printf
- To opt-in, define
CATCH_CONFIG_EXPERIMENTAL_REDIRECT
in the implementation file
- This capture can also redirect output written via C apis, e.g.
- Added a new fallback stringifier for classes derived from
std::exception
- Both
StringMaker
specialization andoperator<<
overload are given priority
- Both
Miscellaneous
contrib/
now contains dbg scripts that skip over Catch's internals (#904, #1283)gdbinit
for gdblldbinit
for lldb
CatchAddTests.cmake
no longer strips whitespace from tests (#1265, #1281)- Online documentation now describes
--use-colour
option (#1263)
2.2.2
Fixes
- Fixed bug in
WithinAbs::match()
failing spuriously (#1228) - Fixed clang-tidy diagnostic about virtual call in destructor (#1226)
- Reduced the number of GCC warnings suppression leaking out of the header (#1090, #1091)
- Only
-Wparentheses
should be leaking now
- Only
- Added upper bound on the time benchmark timer calibration is allowed to take (#1237)
- On platforms where
std::chrono::high_resolution_clock
's resolution is low, the calibration would appear stuck
- On platforms where
- Fixed compilation error when stringifying static arrays of
unsigned char
s (#1238)
Improvements
- XML encoder now hex-encodes invalid UTF-8 sequences (#1207)
- This affects xml and junit reporters
- Some invalid UTF-8 parts are left as is, e.g. surrogate pairs. This is because certain extensions of UTF-8 allow them, such as WTF-8.
- CLR objects (
T^
) can now be stringified (#1216)- This affects code compiled as C++/CLI
- Added
PredicateMatcher
, a matcher that takes an arbitrary predicate function (#1236)
Others
- Modified CMake-installed pkg-config to allow
#include <catch.hpp>
(#1239)- The plans to standardize on
#include <catch2/catch.hpp>
are still in effect
- The plans to standardize on
2.2.1
Fixes
- Fixed compilation error when compiling Catch2 with
std=c++17
against libc++ (#1214)- Clara (Catch2's CLI parsing library) used
std::optional
without including it explicitly
- Clara (Catch2's CLI parsing library) used
- Fixed Catch2 return code always being 0 (#1215)
- In the words of STL, "We feel superbad about letting this in"
2.2.0
Fixes
- Hidden tests are not listed by default when listing tests (#1175)
- This makes
catch_discover_tests
CMake script work better
- This makes
- Fixed regression that meant
<windows.h>
could potentially not be included properly (#1197) - Fixed installing
Catch2ConfigVersion.cmake
when Catch2 is a subproject.
Improvements
- Added an option to warn (+ exit with error) when no tests were ran (#1158)
- Use as
-w NoTests
- Use as
- Added provisional support for Emscripten (#1114)
- Added a way to override the fallback stringifier (#1024)
- This allows project's own stringification machinery to be easily reused for Catch
Catch::Session::run()
now acceptschar const * const *
, allowing it to accept array of string literals (#1031, #1178)- The embedded version of Clara was bumped to v1.1.3
- Various minor performance improvements
- Added support for DJGPP DOS crosscompiler (#1206)
2.1.2
Fixes
- Fixed compilation error with
-fno-rtti
(#1165) - Fixed NoAssertion warnings
operator<<
is used before range-based stringification (#1172)- Fixed
-Wpedantic
warnings (extra semicolons and binary literals) (#1173)
Improvements
- Added
CATCH_VERSION_{MAJOR,MINOR,PATCH}
macros (#1131) - Added
BrightYellow
colour for use in reporters (#979)- It is also used by ConsoleReporter for reconstructed expressions
Other changes
- Catch is now exported as a CMake package and linkable target (#1170)
2.1.1
Improvements
- Static arrays are now properly stringified like ranges across MSVC/GCC/Clang
- Embedded newer version of Clara -- v1.1.1
- This should fix some warnings dragged in from Clara
- MSVC's CLR exceptions are supported
Fixes
- Fixed compilation when comparison operators do not return bool (#1147)
- Fixed CLR exceptions blowing up the executable during translation (#1138)
Other changes
- Many CMake changes
NO_SELFTEST
option is deprecated, useBUILD_TESTING
instead.- Catch specific CMake options were prefixed with
CATCH_
for namespacing purposes - Other changes to simplify Catch2's packaging
2.1.0
Improvements
- Various performance improvements
- On top of the performance regression fixes
- Experimental support for PCH was added (#1061)
CATCH_CONFIG_EXTERNAL_INTERFACES
now brings in declarations of Console, Compact, XML and JUnit reportersMatcherBase
no longer has a pointless second template argument- Reduced the number of warning suppressions that leak into user's code
- Bugs in g++ 4.x and 5.x mean that some of them have to be left in
Fixes
- Fixed performance regression from Catch classic
- One of the performance improvement patches for Catch classic was not applied to Catch2
- Fixed platform detection for iOS (#1084)
- Fixed compilation when
g++
is used together withlibc++
(#1110) - Fixed TeamCity reporter compilation with the single header version
- To fix the underlying issue we will be versioning reporters in single_include folder per release
- The XML reporter will now report
WARN
messages even when not used with-s
- Fixed compilation when
VectorContains
matcher was combined using&&
(#1092) - Fixed test duration overflowing after 10 seconds (#1125, #1129)
- Fixed
std::uncaught_exception
deprecation warning (#1124)
New features
- New Matchers
- Regex matcher for strings,
Matches
. - Set-equal matcher for vectors,
UnorderedEquals
- Floating point matchers,
WithinAbs
andWithinULP
.
- Regex matcher for strings,
- Stringification now attempts to decompose all containers (#606)
- Containers are objects that respond to ADL
begin(T)
andend(T)
.
- Containers are objects that respond to ADL
Other changes
- Reporters will now be versioned in the
single_include
folder to ensure their compatibility with the last released version
2.0.1
Breaking changes
- Removed C++98 support
- Removed legacy reporter support
- Removed legacy generator support
- Generator support will come back later, reworked
- Removed
Catch::toString
support- The new stringification machinery uses
Catch::StringMaker
specializations first andoperator<<
overloads second.
- The new stringification machinery uses
- Removed legacy
SCOPED_MSG
andSCOPED_INFO
macros - Removed
INTERNAL_CATCH_REGISTER_REPORTER
CATCH_REGISTER_REPORTER
should be used to register reporters
- Removed legacy
[hide]
tag[.]
,[.foo]
and[!hide]
are still supported
- Output into debugger is now colourized
*_THROWS_AS(expr, exception_type)
now unconditionally appendsconst&
to the exception type.CATCH_CONFIG_FAST_COMPILE
now affects theCHECK_
family of assertions as well asREQUIRE_
family of assertions- This is most noticeable in
CHECK(throws())
, which would previously report failure, properly stringify the exception and continue. Now it will report failure and stop executing current section.
- This is most noticeable in
- Removed deprecated matcher utility functions
Not
,AllOf
andAnyOf
.- They are superseded by operators
!
,&&
and||
, which are natural and do not have limited arity
- They are superseded by operators
- Removed support for non-const comparison operators
- Non-const comparison operators are an abomination that should not exist
- They were breaking support for comparing function to function pointer
std::pair
andstd::tuple
are no longer stringified by default- This is done to avoid dragging in
<tuple>
and<utility>
headers in common path - Their stringification can be enabled per-file via new configuration macros
- This is done to avoid dragging in
Approx
is subtly different and hopefully behaves more as users would expectApprox::scale
defaults to0.0
Approx::epsilon
no longer applies to the larger of the two compared values, but only to theApprox
's valueINFINITY == Approx(INFINITY)
returns true
Improvements
- Reporters and Listeners can be defined in files different from the main file
- The file has to define
CATCH_CONFIG_EXTERNAL_INTERFACES
before including catch.hpp.
- The file has to define
- Errors that happen during set up before main are now caught and properly reported once main is entered
- If you are providing your own main, you can access and use these as well.
- New assertion macros, *_THROWS_MATCHES(expr, exception_type, matcher) are provided
- As the arguments suggest, these allow you to assert that an expression throws desired type of exception and pass the exception to a matcher.
- JUnit reporter no longer has significantly different output for test cases with and without sections
- Most assertions now support expressions containing commas (ie
REQUIRE(foo() == std::vector<int>{1, 2, 3});
) - Catch now contains experimental micro benchmarking support
- See
projects/SelfTest/Benchmark.tests.cpp
for examples - The support being experiment means that it can be changed without prior notice
- See
- Catch uses new CLI parsing library (Clara)
- Users can now easily add new command line options to the final executable
- This also leads to some changes in
Catch::Session
interface
- All parts of matchers can be removed from a TU by defining
CATCH_CONFIG_DISABLE_MATCHERS
- This can be used to somewhat speed up compilation times
- An experimental implementation of
CATCH_CONFIG_DISABLE
has been added- Inspired by Doctest's
DOCTEST_CONFIG_DISABLE
- Useful for implementing tests in source files
- ie for functions in anonymous namespaces
- Removes all assertions
- Prevents
TEST_CASE
registrations - Exception translators are not registered
- Reporters are not registered
- Listeners are not registered
- Inspired by Doctest's
- Reporters/Listeners are now notified of fatal errors
- This means specific signals or structured exceptions
- The Reporter/Listener interface provides default, empty, implementation to preserve backward compatibility
- Stringification of
std::chrono::duration
andstd::chrono::time_point
is now supported- Needs to be enabled by a per-file compile time configuration option
- Add
pkg-config
support to CMake install command
Fixes
- Don't use console colour if running in XCode
- Explicit constructor in reporter base class
- Swept out
-Wweak-vtables
,-Wexit-time-destructors
,-Wglobal-constructors
warnings - Compilation for Universal Windows Platform (UWP) is supported
- SEH handling and colorized output are disabled when compiling for UWP
- Implemented a workaround for
std::uncaught_exception
issues in libcxxrt- These issues caused incorrect section traversals
- The workaround is only partial, user's test can still trigger the issue by using
throw;
to rethrow an exception
- Suppressed C4061 warning under MSVC
Internal changes
- The development version now uses .cpp files instead of header files containing implementation.
- This makes partial rebuilds much faster during development
- The expression decomposition layer has been rewritten
- The evaluation layer has been rewritten
- New library (TextFlow) is used for formatting text to output
Older versions
1.12.x
1.12.2
Fixes
- Fixed missing include
1.12.1
Fixes
- Fixed deprecation warning in
ScopedMessage::~ScopedMessage
- All uses of
min
ormax
identifiers are now wrapped in parentheses- This avoids problems when Windows headers define
min
andmax
macros
- This avoids problems when Windows headers define
1.12.0
Fixes
- Fixed compilation for strict C++98 mode (ie not gnu++98) and older compilers (#1103)
INFO
messages are included in thexml
reporter output even without-s
specified.
1.11.x
1.11.0
Fixes
- The original expression in
REQUIRE_FALSE( expr )
is now reporter properly as!( expr )
(#1051)- Previously the parentheses were missing and
x != y
would be expanded as!x != x
- Previously the parentheses were missing and
Approx::Margin
is now inclusive (#952)- Previously it was meant and documented as inclusive, but the check itself wasn't
- This means that
REQUIRE( 0.25f == Approx( 0.0f ).margin( 0.25f ) )
passes, instead of fails
RandomNumberGenerator::result_type
is now unsigned (#1050)
Improvements
__JETBRAINS_IDE__
macro handling is now CLion version specific (#1017)- When CLion 2017.3 or newer is detected,
__COUNTER__
is used instead of
- When CLion 2017.3 or newer is detected,
- TeamCity reporter now explicitly flushes output stream after each report (#1057)
- On some platforms, output from redirected streams would show up only after the tests finished running
ParseAndAddCatchTests
now can add test files as dependency to CMake configuration- This means you do not have to manually rerun CMake configuration step to detect new tests
1.10.x
1.10.0
Fixes
- Evaluation layer has been rewritten (backported from Catch 2)
- The new layer is much simpler and fixes some issues (#981)
- Implemented workaround for VS 2017 raw string literal stringification bug (#995)
- Fixed interaction between
[!shouldfail]
and[!mayfail]
tags and sections- Previously sections with failing assertions would be marked as failed, not failed-but-ok
Improvements
- Added libidentify support
- Added "wait-for-keypress" option
1.9.x
1.9.6
Improvements
- Catch's runtime overhead has been significantly decreased (#937, #939)
- Added
--list-extra-info
cli option (#934).- It lists all tests together with extra information, ie filename, line number and description.
1.9.5
Fixes
- Truthy expressions are now reconstructed properly, not as booleans (#914)
- Various warnings are no longer erroneously suppressed in test files (files that include
catch.hpp
, but do not defineCATCH_CONFIG_MAIN
orCATCH_CONFIG_RUNNER
) (#871) - Catch no longer fails to link when main is compiled as C++, but linked against Objective-C (#855)
- Fixed incorrect gcc version detection when deciding to use
__COUNTER__
(#928)- Previously any GCC with minor version less than 3 would be incorrectly classified as not supporting
__COUNTER__
.
- Previously any GCC with minor version less than 3 would be incorrectly classified as not supporting
- Suppressed C4996 warning caused by upcoming updated to MSVC 2017, marking
std::uncaught_exception
as deprecated. (#927)
Improvements
- CMake integration script now incorporates debug messages and registers tests in an improved way (#911)
- Various documentation improvements
1.9.4
Fixes
CATCH_FAIL
macro no longer causes compilation error without variadic macro supportINFO
messages are no longer cleared after being reported once
Improvements and minor changes
- Catch now uses
wmain
when compiled under Windows andUNICODE
is defined.- Note that Catch still officially supports only ASCII
1.9.3
Fixes
- Completed the fix for (lack of) uint64_t in earlier Visual Studios
1.9.2
Improvements and minor changes
- All of
Approx
's member functions now accept strong typedefs in C++11 mode (#888)- Previously
Approx::scale
,Approx::epsilon
,Approx::margin
andApprox::operator()
didn't.
- Previously
Fixes
- POSIX signals are now disabled by default under QNX (#889)
- QNX does not support current enough (2001) POSIX specification
- JUnit no longer counts exceptions as failures if given test case is marked as ok to fail.
Catch::Option
should now have its storage properly aligned.- Catch no longer attempts to define
uint64_t
on windows (#862)- This was causing trouble when compiled under Cygwin
Other
- Catch is now compiled under MSVC 2017 using
std:c++latest
(C++17 mode) in CI - We now provide cmake script that autoregisters Catch tests into ctest.
- See
contrib
folder.
- See
1.9.1
Fixes
- Unexpected exceptions are no longer ignored by default (#885, #887)
1.9.0
Improvements and minor changes
- Catch no longer attempts to ensure the exception type passed by user in
REQUIRE_THROWS_AS
is a constant reference.- It was causing trouble when
REQUIRE_THROWS_AS
was used inside templated functions - This actually reverts changes made in v1.7.2
- It was causing trouble when
- Catch's
Version
struct should no longer be double freed when multiple instances of Catch tests are loaded into single program (#858)- It is now a static variable in an inline function instead of being an
extern
ed struct.
- It is now a static variable in an inline function instead of being an
- Attempt to register invalid tag or tag alias now throws instead of calling
exit()
.- Because this happen before entering main, it still aborts execution
- Further improvements to this are coming
CATCH_CONFIG_FAST_COMPILE
now speeds-up compilation ofREQUIRE*
assertions by further ~15%.- The trade-off is disabling translation of unexpected exceptions into text.
- When Catch is compiled using C++11,
Approx
is now constructible with anything that can be explicitly converted todouble
. - Captured messages are now printed on unexpected exceptions
Fixes:
- Clang's
-Wexit-time-destructors
should be suppressed for Catch's internals - GCC's
-Wparentheses
is now suppressed for all TU's that includecatch.hpp
.- This is functionally a revert of changes made in 1.8.0, where we tried using
_Pragma
based suppression. This should have kept the suppression local to Catch's assertions, but bugs in GCC's handling of_Pragma
s in C++ mode meant that it did not always work.
- This is functionally a revert of changes made in 1.8.0, where we tried using
- You can now tell Catch to use C++11-based check when checking whether a type can be streamed to output.
- This fixes cases when an unstreamable type has streamable private base (#877)
- Details can be found in documentation
Other notes:
- We have added VS 2017 to our CI
- Work on Catch 2 should start soon
1.8.x
1.8.2
Improvements and minor changes
- TAP reporter now behaves as if
-s
was always set- This should be more consistent with the protocol desired behaviour.
- Compact reporter now obeys
-d yes
argument (#780)- The format is "XXX.123 s: " (3 decimal places are always present).
- Before it did not report the durations at all.
- XML reporter now behaves the same way as Console reporter in regards to
INFO
- This means it reports
INFO
messages on success, if output on success (-s
) is enabled. - Previously it only reported
INFO
messages on failure.
- This means it reports
CAPTURE(expr)
now stringifiesexpr
in the same way assertion macros do (#639)- Listeners are now finally documented.
- Listeners provide a way to hook into events generated by running your tests, including start and end of run, every test case, every section and every assertion.
Fixes:
- Catch no longer attempts to reconstruct expression that led to a fatal error (#810)
- This fixes possible signal/SEH loop when processing expressions, where the signal was triggered by expression decomposition.
- Fixed (C4265) missing virtual destructor warning in Matchers (#844)
std::string
s are now taken byconst&
everywhere (#842).- Previously some places were taking them by-value.
- Catch should no longer change errno (#835).
- This was caused by libstdc++ bug that we now work around.
- Catch now provides
FAIL_CHECK( ... )
macro (#765).- Same as
FAIL( ... )
, but does not abort the test.
- Same as
- Functions like
fabs
,tolower
,memset
,isalnum
are now used withstd::
qualification (#543). - Clara no longer assumes first argument (binary name) is always present (#729)
- If it is missing, empty string is used as default.
- Clara no longer reads 1 character past argument string (#830)
- Regression in Objective-C bindings (Matchers) fixed (#854)
Other notes:
- We have added VS 2013 and 2015 to our CI
- Catch Classic (1.x.x) now contains its own, forked, version of Clara (the argument parser).
1.8.1
Fixes
Cygwin issue with gettimeofday
- #define
was not early enough
1.8.0
New features/ minor changes
- Matchers have new, simpler (and documented) interface.
- Catch provides string and vector matchers.
- For details see Matchers documentation.
- Changed console reporter test duration reporting format (#322)
- Old format:
Some simple comparisons between doubles completed in 0.000123s
- New format:
xxx.123s: Some simple comparisons between doubles
(There will always be exactly 3 decimal places)
- Old format:
- Added opt-in leak detection under MSVC + Windows (#439)
- Enable it by compiling Catch's main with
CATCH_CONFIG_WINDOWS_CRTDBG
- Enable it by compiling Catch's main with
- Introduced new compile-time flag,
CATCH_CONFIG_FAST_COMPILE
, trading features for compilation speed.- Moves debug breaks out of tests and into implementation, speeding up test compilation time (~10% on linux).
- More changes are coming
- Added TAP (Test Anything Protocol) and Automake reporters.
- These are not present in the default single-include header and need to be downloaded from GitHub separately.
- For details see documentation about integrating with build systems.
- XML reporter now reports filename as part of the
Section
andTestCase
tags. Approx
now supports an optional margin of absolute error- It has also received new documentation.
Fixes
- Silenced C4312 ("conversion from int to 'ClassName *") warnings in the evaluate layer.
- Fixed C4512 ("assignment operator could not be generated") warnings under VS2013.
- Cygwin compatibility fixes
- Signal handling is no longer compiled by default.
- Usage of
gettimeofday
inside Catch should no longer cause compilation errors.
- Improved
-Wparentheses
suppression for gcc (#674)- When compiled with gcc 4.8 or newer, the suppression is localized to assertions only
- Otherwise it is suppressed for the whole TU
- Fixed test spec parser issue (with escapes in multiple names)
Other
- Various documentation fixes and improvements
1.7.x
1.7.2
Fixes and minor improvements
Xml:
(technically the first two are breaking changes but are also fixes and arguably break few if any people)
- C-escape control characters instead of XML encoding them (which requires XML 1.1)
- Revert XML output to XML 1.0
- Can provide stylesheet references by extending the XML reporter
- Added description and tags attributes to XML Reporter
- Tags are closed and the stream flushed more eagerly to avoid stdout interpolation
Other:
REQUIRE_THROWS_AS
now catches exception byconst&
and reports expected type- In
SECTION
s the file/ line is now of theSECTION
. not theTEST_CASE
- Added std:: qualification to some functions from C stdlib
- Removed use of RTTI (
dynamic_cast
) that had crept back in - Silenced a few more warnings in different circumstances
- Travis improvements
1.7.1
Fixes:
- Fixed inconsistency in defining
NOMINMAX
andWIN32_LEAN_AND_MEAN
insidecatch.hpp
. - Fixed SEH-related compilation error under older MinGW compilers, by making Windows SEH handling opt-in for compilers other than MSVC.
- For specifics, look into the documentation.
- Fixed compilation error under MinGW caused by improper compiler detection.
- Fixed XML reporter sometimes leaving an empty output file when a test ends with signal/structured exception.
- Fixed XML reporter not reporting captured stdout/stderr.
- Fixed possible infinite recursion in Windows SEH.
- Fixed possible compilation error caused by Catch's operator overloads being ambiguous in regards to user-defined templated operators.
1.7.0
Features/ Changes:
- Catch now runs significantly faster for passing tests
- Microbenchmark focused on Catch's overhead went from ~3.4s to ~0.7s.
- Real world test using JSON for Modern C++'s test suite went from ~6m 25s to ~4m 14s.
- Catch can now run specific sections within test cases.
- For now the support is only basic (no wildcards or tags), for details see the documentation.
- Catch now supports SEH on Windows as well as signals on Linux.
- After receiving a signal, Catch reports failing assertion and then passes the signal onto the previous handler.
- Approx can be used to compare values against strong typedefs (available in C++11 mode only).
- Strong typedefs mean types that are explicitly convertible to double.
- CHECK macro no longer stops executing section if an exception happens.
- Certain characters (space, tab, etc) are now pretty printed.
- This means that a
char c = ' '; REQUIRE(c == '\t');
would be printed as' ' == '\t'
, instead of== 9
.
- This means that a
Fixes:
- Text formatting no longer attempts to access out-of-bounds characters under certain conditions.
- THROW family of assertions no longer trigger
-Wunused-value
on expressions containing explicit cast. - Breaking into debugger under OS X works again and no longer required
DEBUG
to be defined. - Compilation no longer breaks under certain compiler if a lambda is used inside assertion macro.
Other:
- Catch's CMakeLists now defines install command.
- Catch's CMakeLists now generates projects with warnings enabled.
1.6.x
1.6.1
Features/ Changes:
- Catch now supports breaking into debugger on Linux
Fixes:
- Generators no longer leak memory (generators are still unsupported in general)
- JUnit reporter now reports UTC timestamps, instead of "tbd"
CHECK_THAT
macro is now properly defined asCATCH_CHECK_THAT
when usingCATCH_
prefixed macros
Other:
- Types with overloaded
&&
operator are no longer evaluated twice when used in an assertion macro. - The use of
__COUNTER__
is suppressed when Catch is parsed by CLion- This change is not active when compiling a binary
- Approval tests can now be run on Windows
- CMake will now warn if a file is present in the
include
folder but not is not enumerated as part of the project - Catch now defines
NOMINMAX
andWIN32_LEAN_AND_MEAN
before includingwindows.h
- This can be disabled if needed, see documentation for details.
1.6.0
Cmake/ projects:
- Moved CMakeLists.txt to root, made it friendlier for CLion and generating XCode and VS projects, and removed the manually maintained XCode and VS projects.
Features/ Changes:
- Approx now supports
>=
and<=
- Can now use
\
to escape chars in test names on command line - Standardize C++11 feature toggles
Fixes:
- Blue shell colour
- Missing argument to
CATCH_CHECK_THROWS
- Don't encode extended ASCII in XML
- use
std::shuffle
on more compilers (fixes deprecation warning/error) - Use
__COUNTER__
more consistently (where available)
Other:
- Tweaks and changes to scripts - particularly for Approval test - to make them more portable
Even Older versions
Release notes were not maintained prior to v1.6.0, but you should be able to work them out from the Git history