CAPTURE of structured binding element no longer works

[u3shit]

Description

Take a look at the example below where I capture a variable from a structured binding. With 2.3.3 and before it worked fine, with 2.3.4 I receive an error like this:

foo.cpp:8:11: error: reference to local binding 'a' declared in enclosing function '_DOCTEST_ANON_FUNC_6'
  CAPTURE(a);
          ^
./doctest/doctest.h:2480:17: note: expanded from macro 'CAPTURE'
#define CAPTURE DOCTEST_CAPTURE
                ^
foo.cpp:7:9: note: 'a' declared here
  auto [a,b] = x;
        ^
1 error generated.

It looks like it was broken in 5e493d3

Steps to reproduce

#define DOCTEST_CONFIG_IMPLEMENT_WITH_MAIN
#include <doctest/doctest.h>

TEST_CASE("x")
{
  int x[2] = {1,2};
  auto [a,b] = x;
  CAPTURE(a);
}

Compile it with clang++ -std=c++17. It also fails if I use an std::pair/tuple instead of an array.

Extra information

• doctest version: v2.3.4
• Operating System: Gentoo Linux amd64
• Compiler+version: clang 8.0.1 (both with libc++ and gcc's libstdc++)

[onqtam]

INFO() relies on a [&] lambda capture underneath and the problem you've hit is this: https://stackoverflow.com/questions/46114214

........ This works with GCC but fails with clang (haven't tested MSVC).

You could work it around by making a separate auto& name = a; for now and CAPTURE(name); and hopefully the C++ 20/23 standard figures out what to do with structured bindings and lambda captures. Thanks for reporting!

We won't revert to the old implementation of INFO() because it had a horrible limitation - it worked only with lvalues and was a bunch of code which had to be maintained - the new lambda approach is far superior (thanks to @DaanDeMeyer ).

Btw it's impressive that you've tracked down the commit which caused this - in this case it didn't matter much but I wish more bug reports were as thorough as yours!

[u3shit]

The funny thing is, I already had a macro for that workaround you suggested because of the rvalue limitation... looks like I no longer need it for rvalues, but now I need structured bindings. Am I the only one who feels c++ standards got buggier recently?
I haven't looked at the implementation, but wouldn't the initializer capture workaround work here? Like [a=&a]() { return *a; }

[onqtam]

it won't work in the following case: INFO("aaa" << var1 << var2);

I don't know what the user will be capturing so the capture has to be [&]... I can't generate a proper capture at compile time with macros from whatever the user has passed. Hopefully the bug/hole in the standard will get fixed nicely in the future. Btw I really dislike C++ since I explored Nim and will even be giving a talk how it should replace it at code::dive 2019 :)
https://codedive.pl/index/speaker/name/viktor-kirilov

[onqtam]

releasing 2.3.5

[gitsbi]

This is not a bug in a compiler (nor in the standard), but a deliberate decision. The standard says structured bindings are no variables, and thus cannot be captured. (See the discussion here with relevant links.)