Lambda capture seems to prevent source file name being obtained

[claremacrae]

Expected Behavior

That the filename obtained from a custom reporter should be valid, in this code - which can be run at https://godbolt.org/z/13Y7Wq1r7

#define APPROVALS_UT
#include "MinimalApprovalTests.h"

int main()
{
    using namespace boost::ut;
    using namespace ApprovalTests;

    "VerifyAFileWithAmpersand&"_test = [&]() { // <<< Changing [&] to [] makes the test pass
        ApprovalTestNamer namer;

        auto name = namer.getSourceFileName();
        expect(name == "ut_starter_test") << "value was:" << name; // <<< this test fails - the name is unknown

        auto test_case = namer.getTestName();
        expect(test_case == "VerifyAFileWithAmpersand&") << "value was:" << test_case;
    };
}

Actual Behavior

Two different users of Approval Tests have reported that when a lambda capture is used, the source name is reported as 'unknown':

• [boost/ut.hpp] Unknown source error when the topmost lambda captures approvals/ApprovalTests.cpp#177
  • [boost/ut.hpp] Unknown source error when the topmost lambda captures
• Ampersand Operator in ut test framework approvals/ApprovalTests.cpp#157
  • Ampersand Operator in ut test framework

Steps to Reproduce the Problem

I've created a massively cut-down version of the Approval Tests code, for a minimal example. It took some hours - I hope it makes it easier to track down the problem:

1. Run the code in https://godbolt.org/z/13Y7Wq1r7
2. The test on line 419 fails. It should pass

If downloading this code to run locally, you will need to:

1. Edit line 73 to change include your boost/ut.hpp instead of #include <https://raw.githubusercontent.com/boost-ext/ut/master/include/boost/ut.hpp>
2. Edit line 419 to change __stdin__ to be the base name (file without extension) of the source file you saved.

Specifications

• Version: ffe66dd
• Platform: Tested on Mac and Linux - the Godbolt example is using gcc trunk
• Subsystem: Unsure

[sam20908]

I have figured out a possible fix. This involves using the actual <source_location> when it's available, and it also requires making the signature in reflection::source_location::current from constexpr to consteval. This was tested to have success when I incorporated my own patch of ut.hpp with my own test compiled in MSVC.

Does this approach sound reasonable? I can open a PR if needed.

[sam20908]

It looks like this issue is solved by 76e2f56. This issue can be closed.

[claremacrae]

Hi, thanks for looking this - much appreciated!

Can someone please say why the failing test in the godbolt link above is still failing - it downloads the latest version of the header from here, so if this bug were fixed, I would have thought that the 2nd test in that code would be passing now....

Here's the link again:
https://godbolt.org/z/13Y7Wq1r7

[claremacrae]

@krzysztof-jusiak Hi Kris, in case you haven't seen it already, this issue isn't fixed - the godbolt link in the issue description and in the comment above still fails to compile... the error is still present...

Please could you re-open this ticket, or perhaps provide a workaround? Many thanks.

/CC @sam20908

[kris-jusiak]

oh, sorry, I didn't noice, thank you for pointing it out, repopened now