Warnings about REQUIRE when used in templates

[rr-]

Consider this:

template<class T> inline void test_reading_multiple_bytes(const TestType type)
{
    SECTION("...")
    {
        T reader("\x8F\x8F"_b); // 10001111
        if (type == TestType::Msb)
        {
            REQUIRE((reader.get(7) == (0x8F >> 1)));
            REQUIRE(reader.get(1));
            REQUIRE(reader.get(1));
            REQUIRE(!reader.get(1));
            REQUIRE((reader.get(4) == 3));
            REQUIRE((reader.get(2) == 3));
        }
        else
        {
            REQUIRE(reader.get(1));
            REQUIRE((reader.get(7) == (0x8F >> 1)));
            REQUIRE(reader.get(1));
            REQUIRE(reader.get(1));
            REQUIRE(reader.get(2) == 3);
            REQUIRE(reader.get(4) == 8);
        }
    }
}

TEST_CASE("...")
{
    test_reading_multiple_bytes<MsbBitReader>(TestType::Msb);
    test_reading_multiple_bytes<LsbBitReader>(TestType::Lsb);
}

This produces warnings with gcc 5.3 on GNU/Linux:

/home/rr-/src/arc_unpacker/tests/io/bit_reader_test.cc: In instantiation of ‘void test_reading_multiple_bytes({anonymous}::TestType) [with T = au::io::OldBitReader]’:
/home/rr-/src/arc_unpacker/tests/io/bit_reader_test.cc:277:64:   required from here
/home/rr-/src/arc_unpacker/tests/io/bit_reader_test.cc:89:39: warning: suggest parentheses around comparison in operand of ‘==’ [-Wparentheses]
             REQUIRE(reader.get(2) == 3);
                                   ^
/home/rr-/src/arc_unpacker/tests/io/bit_reader_test.cc:90:39: warning: suggest parentheses around comparison in operand of ‘==’ [-Wparentheses]
             REQUIRE(reader.get(4) == 8);

When I wrap with ( ) like gcc asks me to, it stops complaining, although on top of code becoming ugly, I lose information on the reason behind failed tests (i.e. rather than being shown useful 3 == 8, I am shown false only - not useful much).

[rr-]

...and in case you wonder - the reason behind templated tests is basically rolling my own solution to #357.

[horenmar]

Since we recently improved supression of -Wparentheses I though I could just close this, but as it turns out, g++ will not warn on the same code outside of templates, and will not accept _Pragma based supression inside of templates.

[philsquared]

The current workaround for this issue is to suppress the warning for your whole test file (using the appropriate pragma for your compiler)

[bdb]

Just another vote for fixing this. I have a few templated tests that worked fine in 1.7 and 1.6 but are now broken in 1.8.2 (with GCC 5.4).

As a workaround I just wrapped the template definition with the CATCH_INTERNAL... macros instead of setting the pragma for the whole file.

[horenmar]

I am going to close this, as I moved Catch back to force disabling -Wparentheses everywhere using #pragma in 8f85d08 and this sample code compiles cleanly.

#define CATCH_CONFIG_MAIN
#include "catch.hpp"

template <typename T>
void foo(T) {
    int a, b;
    a = 1; b = 2;
    REQUIRE(a == b);
}

TEST_CASE("Wparentheses") {
    foo(true);
}

This means that we disable the warning for GCC users even outside our assertions, but the spotty _Pragma support meant that the status quo was untenable.