Neuter exception error for ActiveRecord::RecordInvalid

[Reinier de Lange]

For some reason, an AR object I have is unmarshallable (rails 7.0.3.1). When this record is passed in ActiveRecord::RecordInvalid and raised in a test, the following error occurs since 5.16.0:

Minitest::UnexpectedError:         NoMethodError: undefined method `errors' for "Validation failed: Receipt is invalid":String
            activerecord (7.0.3.1) lib/active_record/validations.rb:21:in `initialize'
            minitest (5.16.2) lib/minitest/test.rb:223:in `new'
            minitest (5.16.2) lib/minitest/test.rb:223:in `new_exception'
            minitest (5.16.2) lib/minitest/test.rb:215:in `neuter_exception'
            minitest (5.16.2) lib/minitest/test.rb:208:in `rescue in sanitize_exception'
            minitest (5.16.2) lib/minitest/test.rb:204:in `sanitize_exception'
            minitest (5.16.2) lib/minitest/test.rb:201:in `rescue in capture_exceptions'
            minitest (5.16.2) lib/minitest/test.rb:194:in `capture_exceptions'
            minitest (5.16.2) lib/minitest/test.rb:95:in `block (2 levels) in run'
            minitest-hooks (1.5.0) lib/minitest/hooks/test.rb:38:in `block (2 levels) in time_it'
            minitest-hooks (1.5.0) lib/minitest/hooks/test.rb:31:in `around'
            minitest-hooks (1.5.0) lib/minitest/hooks/test.rb:37:in `block in time_it'
            minitest (5.16.2) lib/minitest.rb:296:in `time_it'
            minitest-hooks (1.5.0) lib/minitest/hooks/test.rb:36:in `time_it'
            minitest (5.16.2) lib/minitest/test.rb:94:in `block in run'
            minitest (5.16.2) lib/minitest.rb:391:in `on_signal'
            minitest (5.16.2) lib/minitest/test.rb:243:in `with_info_handler'
            minitest (5.16.2) lib/minitest/test.rb:93:in `run'
            minitest-reporters (1.5.0) lib/minitest/reporters.rb:48:in `run_with_hooks'
            minitest (5.16.2) lib/minitest.rb:1059:in `run_one_method'
            minitest (5.16.2) lib/minitest.rb:365:in `run_one_method'
            minitest (5.16.2) lib/minitest.rb:352:in `block (2 levels) in run'
            minitest (5.16.2) lib/minitest.rb:351:in `each'
            minitest (5.16.2) lib/minitest.rb:351:in `block in run'
            minitest (5.16.2) lib/minitest.rb:391:in `on_signal'
            minitest (5.16.2) lib/minitest.rb:378:in `with_info_handler'
            minitest-hooks (1.5.0) lib/minitest/hooks/test.rb:81:in `block in with_info_handler'
            minitest-hooks (1.5.0) lib/minitest/hooks/test.rb:26:in `around_all'
            minitest-hooks (1.5.0) lib/minitest/hooks/test.rb:70:in `with_info_handler'
            minitest (5.16.2) lib/minitest.rb:350:in `run'
            railties (7.0.3.1) lib/rails/test_unit/line_filtering.rb:10:in `run'
            minitest (5.16.2) lib/minitest.rb:182:in `block in __run'
            minitest (5.16.2) lib/minitest.rb:182:in `map'
            minitest (5.16.2) lib/minitest.rb:182:in `__run'
            minitest (5.16.2) lib/minitest.rb:159:in `run'
            minitest (5.16.2) lib/minitest.rb:83:in `block in autorun'
            activesupport (7.0.3.1) lib/active_support/fork_tracker.rb:18:in `fork'
            activesupport (7.0.3.1) lib/active_support/fork_tracker.rb:18:in `fork'
            /Users/be/.rbenv/versions/2.7.5/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:83:in `require'
            /Users/be/.rbenv/versions/2.7.5/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:83:in `require'
            -e:1:in `<main>'

This is because the same error class is instantiated with a String message, while the constructor expects an AR object: https://github.com/rails/rails/blob/main/activerecord/lib/active_record/validations.rb#L18. I hope I can find out why the object cannot be marshalled (maybe caused by pry?), but I think the neutering of the exception should be improved anyway to properly handle this case.

[Ryan Davis]

C'mon people... Ya gotta give me more information than that. This is unworkable so far.

[Reinier de Lange]

Oh, I thought my explanation was pretty clear. In short, when an exception is unmarshallable, the fallback approach is to instantiate the same exception class with a message string (https://github.com/minitest/minitest/blob/master/lib/minitest/test.rb#L215). However, custom exceptions like ActiveRecord::RecordInvalid expect the first argument to be an AR object (https://github.com/rails/rails/blob/main/activerecord/lib/active_record/validations.rb#L18).

I don't think you can really deduce from a custom exception what the first argument in its constructor is supposed to be. So maybe the final approach in https://github.com/minitest/minitest/blob/master/lib/minitest/test.rb#L219 should always be done instead?

I can write a repro if you like. It doesn't need to be an ActiveRecord::RecordInvalid exception of course; it can be any exception with a custom constructor..

[Reinier de Lange]

Please see moiristo@9cebd67. This fails with

  1) Error:
TestMinitestTest#test_spec_marshal_with_custom_constructor_exception:
NoMethodError: undefined method `text' for "this is bad!":String

[Reinier de Lange]

Proposed fix is to never reuse the original exception class: moiristo@47d2a01

[Victor Maslov]

I wonder if the https://ruby-doc.org/core-3.1.2/Exception.html#method-i-exception could be used on ActiveRecord object and help here https://github.com/moiristo/minitest/blob/47d2a013e03d308bb92bd90a82aea215d7c1d1fe/lib/minitest/test.rb#L222

Though the #exception method is sometimes "custom" even in ruby stdlib too. Just few days ago discovered it to be defined in Timeout https://ruby-doc.org/stdlib-3.1.2/libdoc/timeout/rdoc/Timeout/Error.html#method-i-exception and have no clue what it is doing but it was ruining my passed args.

[Reinier de Lange]

In what way? It doesn't make the exception marshallable I think. Today, we found out that this bug also breaks our CI build and custom reporting when an error occurs, so for now we've monkeypatched sanitize_exception, as we are not really interested in marshallability anyway:

module Minitest
  class Test
    def sanitize_exception e
      e
    end
  end
end

[Ryan Davis]

Would removing the TypeError restriction be sufficient?