mirror of
https://github.com/catchorg/Catch2.git
synced 2024-11-04 05:09:53 +01:00
265 lines
10 KiB
C++
265 lines
10 KiB
C++
/*
|
|
* Created by Phil on 13/5/2013.
|
|
* Copyright 2014 Two Blue Cubes Ltd. All rights reserved.
|
|
*
|
|
* Distributed under the Boost Software License, Version 1.0. (See accompanying
|
|
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
|
|
*/
|
|
|
|
#include "catch.hpp"
|
|
#include "internal/catch_test_spec_parser.h"
|
|
|
|
#ifdef __clang__
|
|
# pragma clang diagnostic ignored "-Wc++98-compat"
|
|
#endif
|
|
|
|
inline Catch::TestCase fakeTestCase( const char* name, const char* desc = "" ){ return Catch::makeTestCase( nullptr, "", name, desc, CATCH_INTERNAL_LINEINFO ); }
|
|
|
|
TEST_CASE( "Parse test names and tags" ) {
|
|
|
|
using Catch::parseTestSpec;
|
|
using Catch::TestSpec;
|
|
|
|
Catch::TestCase tcA = fakeTestCase( "a" );
|
|
Catch::TestCase tcB = fakeTestCase( "b", "[one][x]" );
|
|
Catch::TestCase tcC = fakeTestCase( "longer name with spaces", "[two][three][.][x]" );
|
|
Catch::TestCase tcD = fakeTestCase( "zlonger name with spacesz" );
|
|
|
|
SECTION( "Empty test spec should have no filters" ) {
|
|
TestSpec spec;
|
|
CHECK( spec.hasFilters() == false );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
}
|
|
|
|
SECTION( "Test spec from empty string should have no filters" ) {
|
|
TestSpec spec = parseTestSpec( "" );
|
|
CHECK( spec.hasFilters() == false );
|
|
CHECK( spec.matches(tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
}
|
|
|
|
SECTION( "Test spec from just a comma should have no filters" ) {
|
|
TestSpec spec = parseTestSpec( "," );
|
|
CHECK( spec.hasFilters() == false );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
}
|
|
|
|
SECTION( "Test spec from name should have one filter" ) {
|
|
TestSpec spec = parseTestSpec( "b" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == true );
|
|
}
|
|
|
|
SECTION( "Test spec from quoted name should have one filter" ) {
|
|
TestSpec spec = parseTestSpec( "\"b\"" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == true );
|
|
}
|
|
|
|
SECTION( "Test spec from name should have one filter" ) {
|
|
TestSpec spec = parseTestSpec( "b" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == true );
|
|
CHECK( spec.matches( tcC ) == false );
|
|
}
|
|
|
|
SECTION( "Wildcard at the start" ) {
|
|
TestSpec spec = parseTestSpec( "*spaces" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
CHECK( spec.matches( tcD ) == false );
|
|
CHECK( parseTestSpec( "*a" ).matches( tcA ) == true );
|
|
}
|
|
SECTION( "Wildcard at the end" ) {
|
|
TestSpec spec = parseTestSpec( "long*" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
CHECK( spec.matches( tcD ) == false );
|
|
CHECK( parseTestSpec( "a*" ).matches( tcA ) == true );
|
|
}
|
|
SECTION( "Wildcard at both ends" ) {
|
|
TestSpec spec = parseTestSpec( "*name*" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
CHECK( spec.matches( tcD ) == true );
|
|
CHECK( parseTestSpec( "*a*" ).matches( tcA ) == true );
|
|
}
|
|
SECTION( "Redundant wildcard at the start" ) {
|
|
TestSpec spec = parseTestSpec( "*a" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == true );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
}
|
|
SECTION( "Redundant wildcard at the end" ) {
|
|
TestSpec spec = parseTestSpec( "a*" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == true );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
}
|
|
SECTION( "Redundant wildcard at both ends" ) {
|
|
TestSpec spec = parseTestSpec( "*a*" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == true );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
}
|
|
SECTION( "Wildcard at both ends, redundant at start" ) {
|
|
TestSpec spec = parseTestSpec( "*longer*" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
CHECK( spec.matches( tcD ) == true );
|
|
}
|
|
SECTION( "Just wildcard" ) {
|
|
TestSpec spec = parseTestSpec( "*" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == true );
|
|
CHECK( spec.matches( tcB ) == true );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
CHECK( spec.matches( tcD ) == true );
|
|
}
|
|
|
|
SECTION( "Single tag" ) {
|
|
TestSpec spec = parseTestSpec( "[one]" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == true );
|
|
CHECK( spec.matches( tcC ) == false );
|
|
}
|
|
SECTION( "Single tag, two matches" ) {
|
|
TestSpec spec = parseTestSpec( "[x]" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == true );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
}
|
|
SECTION( "Two tags" ) {
|
|
TestSpec spec = parseTestSpec( "[two][x]" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
}
|
|
SECTION( "Two tags, spare separated" ) {
|
|
TestSpec spec = parseTestSpec( "[two] [x]" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
}
|
|
SECTION( "Wildcarded name and tag" ) {
|
|
TestSpec spec = parseTestSpec( "*name*[x]" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
CHECK( spec.matches( tcD ) == false );
|
|
}
|
|
SECTION( "Single tag exclusion" ) {
|
|
TestSpec spec = parseTestSpec( "~[one]" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == true );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
}
|
|
SECTION( "One tag exclusion and one tag inclusion" ) {
|
|
TestSpec spec = parseTestSpec( "~[two][x]" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == true );
|
|
CHECK( spec.matches( tcC ) == false );
|
|
}
|
|
SECTION( "One tag exclusion and one wldcarded name inclusion" ) {
|
|
TestSpec spec = parseTestSpec( "~[two]*name*" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == false );
|
|
CHECK( spec.matches( tcD ) == true );
|
|
}
|
|
SECTION( "One tag exclusion, using exclude:, and one wldcarded name inclusion" ) {
|
|
TestSpec spec = parseTestSpec( "exclude:[two]*name*" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == false );
|
|
CHECK( spec.matches( tcD ) == true );
|
|
}
|
|
SECTION( "name exclusion" ) {
|
|
TestSpec spec = parseTestSpec( "~b" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == true );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
CHECK( spec.matches( tcD ) == true );
|
|
}
|
|
SECTION( "wildcarded name exclusion" ) {
|
|
TestSpec spec = parseTestSpec( "~*name*" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == true );
|
|
CHECK( spec.matches( tcB ) == true );
|
|
CHECK( spec.matches( tcC ) == false );
|
|
CHECK( spec.matches( tcD ) == false );
|
|
}
|
|
SECTION( "wildcarded name exclusion with tag inclusion" ) {
|
|
TestSpec spec = parseTestSpec( "~*name*,[three]" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == true );
|
|
CHECK( spec.matches( tcB ) == true );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
CHECK( spec.matches( tcD ) == false );
|
|
}
|
|
SECTION( "wildcarded name exclusion, using exclude:, with tag inclusion" ) {
|
|
TestSpec spec = parseTestSpec( "exclude:*name*,[three]" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == true );
|
|
CHECK( spec.matches( tcB ) == true );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
CHECK( spec.matches( tcD ) == false );
|
|
}
|
|
SECTION( "two wildcarded names" ) {
|
|
TestSpec spec = parseTestSpec( "\"longer*\"\"*spaces\"" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == true );
|
|
CHECK( spec.matches( tcD ) == false );
|
|
}
|
|
SECTION( "empty tag" ) {
|
|
TestSpec spec = parseTestSpec( "[]" );
|
|
CHECK( spec.hasFilters() == false );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == false );
|
|
CHECK( spec.matches( tcD ) == false );
|
|
}
|
|
SECTION( "empty quoted name" ) {
|
|
TestSpec spec = parseTestSpec( "\"\"" );
|
|
CHECK( spec.hasFilters() == false );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == false );
|
|
CHECK( spec.matches( tcD ) == false );
|
|
}
|
|
SECTION( "quoted string followed by tag exclusion" ) {
|
|
TestSpec spec = parseTestSpec( "\"*name*\"~[.]" );
|
|
CHECK( spec.hasFilters() == true );
|
|
CHECK( spec.matches( tcA ) == false );
|
|
CHECK( spec.matches( tcB ) == false );
|
|
CHECK( spec.matches( tcC ) == false );
|
|
CHECK( spec.matches( tcD ) == true );
|
|
}
|
|
|
|
}
|