Fails should be fails and not errors

[oxinabox]

Changing

error("new references can only be created in interactive mode ...)"

into

warn("new references can only be created in interactive mode ...)"
@test false

would do it.

But that makes testing the fail case harder.
One way to handle that is to test in a different process.
Which is what the code I was confused about in #4 did:

https://github.com/Evizero/ReferenceTests.jl/pull/4/files#diff-fce720c43af3c52c862fd7451c7374b8R26

[johnnychen94]

But that makes testing the fail case harder.

If we're providing better render and by functions to capture differences, I think this won't be an issue. For example, always throw warnings to explain why the test fails.

[oxinabox]

Just to be clear that line is about how to test this package. not for using this package to rest others.
Though perhaps you are saying we could use the tools for checking the logger?

[johnnychen94]

If you ]test Test, you'll find several all-broken testsets.

Test Summary:          | Broken  Total
@test_broken with fail |      5      5
Test Summary:            | Broken  Total
@test_broken with errors |      2      2
Test Summary: | Broken  Total
@test_skip    |      4      4

@testset "@test_broken with fail" begin
    @test_broken false
    @test_broken 1 == 2
    @test_broken 1 != 1
    @test_broken strip("\t  hi   \n") != "hi"
    @test_broken strip("\t  this should fail   \n") == "hi"
end

I assume this is accepted as long as it's described clearly.