silence clang's `-Wcomma`

[timblechmann]

[codecov[bot]]

Codecov Report

Merging #2543 (289d254) into devel (728de35) will decrease coverage by 0.01%. The diff coverage is n/a.

@@            Coverage Diff             @@
##            devel    #2543      +/-   ##
==========================================
- Coverage   91.58%   91.57%   -0.01%     
==========================================
  Files         183      183              
  Lines        7564     7564              
==========================================
- Hits         6927     6926       -1     
- Misses        637      638       +1     

[horenmar]

Can you provide an example of when this warning triggers?

[timblechmann]

my code looks like this:

#include <catch2/catch_template_test_macros.hpp>

namespace a::b::c {

struct foo {};

using tests = std::tuple< foo >;

TEMPLATE_LIST_TEST_CASE( "test", "", tests )
{}

} // namespace a::b::c

(it seems to be triggered by putting the TEMPLATE_LIST_TEST_CASE macro into the namespace)

[horenmar]

I tried to quickly reproduce this on godbolt, but can't.

Given that this PR would silence -Wcomma for the entire test case code, including user's code, I do not want to merge it without a reproducer to see what the issue actually is, and that it is part of our code (as opposed to user code, whose warnings we should not touch.).

[timblechmann]

i've tried to reproduce it again. it seems to be triggered by using a clang icecc toolchain with local proprocessing. so probably the namespace was a red herring.

fwiw, the offending comma is this one

Catch::NameAndTags{ "test" " - " + std::string("tests") + " - " + std::to_string(index), "" } ), index++)... }; } }
                                                                                               ^

[horenmar]

Hi, sorry for not getting to this earlier. I thought the changes would silence -Wcomma in the test code as well, so I put it in my backlog to return to later.

Today I did, and turns out that suppression is already limited to the registration only.