Extract #assert_rewrites_rbs helper

[amomchilov]

Background

As part of #787, I'm building a more comprehensive test suite that will ensure that the RBS-to-Sorbet translator preserves line numbering.

Achieving correct line numbers while still keeping the same meaning will require the translator to produce some pretty funky code, which is great for machines, but sucks for humans. So we'll want to support both:

1. a more human-readable format that limits line lengths, has reasonable new lines for clarity, etc.
   • this is what we already have
2. a line-matching format that "works" for machines, and preserves the correctness of line numbers in back traces, __LINE__, etc.
   • this is what I'm working on building

This PR

Refactors our existing tests into a new #assert_rewrites_rbs helper. Currently it just has one input and one expected output.

In a future PR, I'll be adding a second expected output for each test case. WIP, but you can get an idea here: 658304d

Very easy to review if you hide whitespace changes.

Alternatives considered

Make a dedicated test class for the line-matching rewriter. That way this test class' test cases have one input and one (human-readable) expected output, and the other class' test cases have one input and one (machine-readable) expected output.

I tried that out, but it became obvious that the two suites would be hard to keep in sync, and we would surely have new additions in the future that are only tested in one mode or the other, rather than both.

[KaanOzkan]

Looks like there are 7 occurrences of tests where the from and to_pretty_format_for_humans is equivalent. Wdyt about adding a default value so that if they're equal the callsites won't have to specify the kwargs:

def assert_rewrites_rbs(
  from = nil,
  to: from,
  max_line_length: nil,
  overloads_strategy: :translate_all
)

[amomchilov]

Spoke offline. Will following up with something similar, but as a separate method like assert_rbs_rewrite_is_no_op.

That way you don't have a call with a from: but not to:

[amomchilov]

Oh also, some things that are no-op today, might not be no-op in the new rewriter mode. I'll explain it in the PR when I open it.

[KaanOzkan]

Why is this wrapped in a begin?

[amomchilov]

Just for visual grouping

[alexcrocha]

Felt weird at first sight, but it quickly grew up on me 👍

[amomchilov]

For context, up the stack there will be two of these, running different sets of assertions/validations against the two different outputs:

spoom/test/spoom/sorbet/translate/rbs_comments_to_sorbet_sigs_test.rb

Lines 1035 to 1066 in 658304d

	begin # Validate the human-readable rewrite
	rewritten_output = rbs_comments_to_sorbet_sigs(
	source_with_rbs,
	max_line_length:,
	overloads_strategy:,
	)
	assert_equal(expected_pretty_format, rewritten_output)
	end
	begin # Validate the line-matched format
	case expected_line_matched_format
	when :same_as_pretty_output
	expected_line_matched_format = expected_pretty_format
	when Symbol
	raise ArgumentError, "Invalid symbol for expected_line_matched_format: #{expected_line_matched_format}"
	end
	# assert_equal(source_with_rbs.lines.count, expected_line_matched_format.lines.count, <<~MSG)
	# Precondition: the expected rewritten code should have the same line count as the RBS-containing input.
	# This is a mistake in the test case, not the rewriter.
	# MSG
	rewritten_output = rbs_comments_to_sorbet_sigs(
	source_with_rbs,
	max_line_length:,
	overloads_strategy:,
	)
	# TODO: run the validator to compare the two results
	assert_equal(expected_line_matched_format, rewritten_output)
	end