Fixup issue with multiline diffs in custom matchers.

[JonRowe]

Fix #420 by not forcing expected to be an array in custom matchers.

[JonRowe]

I''d like a review on this to make sure its a sensible strategy for solving this problem. /cc @myronmarston

[myronmarston]

Yep, planning to review this tonight. I'm out most of the afternoon.

[myronmarston]

What is @no_array meant to be used for? I see it set here but never used again.

[JonRowe]

Was for something that later proved unnecessary...

[myronmarston]

This looks like a reasonable fix, although it doesn't demonstrate adequately that it fixes the diff issue since there's no spec here demonstrating that the diff is the same for a custom matcher as for a matcher like eq. What do you think of bringing over the spec I wrote up in #420? That would help demonstrate this fixes the bug and also guard against future regressions since the diffing behavior of custom matchers is an important behavior to maintain in future releases.

[JonRowe]

Added a spec, it's not quite as simple as proposed in #420 cause ASCII control codes...

[myronmarston]

LGTM.

[JonRowe]

I'd like to hold off till I get a chance to look at the behaviour of expected/actual a bit more actually, I got a weird diff with true/false when playing around with the sample custom matcher.

[myronmarston]

What's an example of where this produces a weird diff?

[JonRowe]

See: https://gist.github.com/JonRowe/2c1200452672ffec0322

[JonRowe]

Did you get a chance to look at the weird diff?

[myronmarston]

Not yet, sorry, thanks for the reminder.

[JonRowe]

Ping?

[myronmarston]

So, looking at your example, I think the diff from before is weird, and the new behavior with this change (e.g. no diff) is correct. What do you think is weird about it?

[JonRowe]

I'm ok with the lack of diff on the true/false but note that the strings aren't being diffed either...

[myronmarston]

I'm ok with the lack of diff on the true/false but note that the strings aren't being diffed either...

We only provide diffs when there are multiline strings involved:

rspec-expectations/lib/rspec/expectations/fail_with.rb

Lines 24 to 26 in 9003d2d

	if any_multiline_strings?(actual, expected)
	message << "\nDiff:" << differ.diff_as_string(coerce_to_string(actual), coerce_to_string(expected))
	end

Your example did not provide multiline strings so I would not expect a diff.

[JonRowe]

Fair enough

[JonRowe]

Merging because the rbx build issue is separate.

[myronmarston]

You still planning to add a deprecation warning to 2.99 for this?

[JonRowe]

Yeah on my list