Warning/error messages mismatching

[andrykonchin]

After recent update to the latest Prism we have two new failures in ruby/spec tests suite, related to a bit different warning/exception messages. So wondering if it's a known issue and 100% matching is supposed at all.

1

  it "warns if there are identical when clauses" do
    -> {
      eval <<~RUBY
        case 1
        when 2
          :foo
        when 2
          :bar
        end
      RUBY
    }.should complain(/warning: duplicated .when' clause with line \d+ is ignored/, verbose: true)
  end

Prism returns "'when' clause on line 4 duplicates 'when' clause on line 2 and is ignored"

2

  it "does not allow calculation or method calls in a pattern" do
    -> {
      eval <<~RUBY
        case 0
        in 1 + 1
          true
        end
      RUBY
    }.should raise_error(SyntaxError, /unexpected|expected a delimiter after the predicates of a `when` clause/)
  end

Prism returns two errors:

• "expected a delimiter after the patterns of aninclause"
• "unexpected '+', ignoring it"

CRuby returns the following error message:

syntax error, unexpected '+', expecting `then' or ';' or '\n'
in 1 + 1
     ^

[andrykonchin]

It seems pretty expected to have multiple syntax errors returned if the parser is used as a tool. But wondering how CRuby will (or already does) decide on which error message to use. And we are supposed to use the same approach in TruffleRuby to be compatible.

[kddnewton]

For the first one, yeah I suppose I'm returning the Ruby 3.4 error message and not the Ruby 3.3 error message, so I can fix that. For the second one, yeah I would concatenate them before you check, it's not always going to be the first error.

[andrykonchin]

Got it, thank you! So feel free to close this issue if it isn't needed to track any planned changing.

[eregon]

For the first one I think it's alright to use the clearer 3.4 error message even with version 3.3, we can just adapt specs to accept both, it doesn't seem worth adding a condition in Prism for this.

[andrykonchin]

Right, the ruby/spec's tests were changed to expect Prism-specific messages as well.

[kddnewton]

Perfect, thanks!