[nz]

Errgonomic 0.9.1 makes Option#to_s and Result#to_s raise Errgonomic::SerializeError again, as 0.8.3 did, with a message that names the value and says what to call instead. Closes #80 and closes #75. The reviewer is asked to confirm one thing: #80 asks for an Errgonomic.to_s_behavior flag, and this change declines it in favor of an unconditional raise, because the one case rendering serves, a log line inside a rescue, is already served by inspect.

What changes

A string built from an Option or a Result fails where it is built. String interpolation, Array#join, format, String(), a bare ERB <%= %> and a Hash key on its way to JSON all reach the value through to_s, and each raises rather than writing Some("...") or None into a hostname, a UNIQUE column or a page. The message is one line:

Some("hot") refuses to_s; use inspect for a log line, or unwrap_or / expect! for the value

inspect is unchanged. A rescue Errgonomic::SerializeError written against 0.8.x still matches. The changelog carries an upgrading note for code written against 0.9.0's rendering, and the 0.9.0 section stays as history.

How it works

• Option::Any#to_s and Result::Any#to_s raise. Each builds its message from a private bounded_inspect, the same 60-character bound the to_json refusal on Option already used, so a record or a long payload does not bury the message.
• The doctests on both to_s methods pin the exact message for interpolation, join, format, String() and the bounded form.
• Three tests in test/rails_test.rb carry the reproductions from to_s rendering (0.9.0) turns the loudest failure class into silent data corruption — make it opt-in #80 and as_json/to_json refusal misses Option hash KEYS — wrapper ships to payloads silently (0.9.0 regression) #75: test_a_wrapper_refuses_to_become_a_string, test_a_bare_erb_tag_refuses_a_wrapper through ActionView, and test_an_option_in_a_hash_key_refuses_to_serialize through the json gem and ActiveSupport's as_json.
• The README says to_s refuses and what to call instead.

Evidence

Every line from #80 and #75, run on 2026-09-10 against this branch with Ruby 3.4.9, ActiveSupport 8.1.3.1 and json 2.21.2. 0.9.0 is main at 43bbd0a.

Line	0.9.0	This branch
[Some("org"), Some("metrics")].join("/")	Some("org")/Some("metrics")	raises
"#{None()}.us-east-1.example"	None.us-east-1.example	raises
"data_#{Some("hot")}"	data_Some("hot")	raises
format("%s", Some(1)), String(Some(1))	Some(1)	raises
ERB <%= Some("visible") %>	Some(&quot;visible&quot;)	raises, as the cause of ActionView::Template::Error
{Some(1) => 2}.as_json	{"Some(1)" => 2}	raises
JSON.generate({Some(1) => 2})	{"Some(1)":2}	raises
group_by { Some/None }.to_json, json gem only	{"None":[1,3],"Some(:even)":[2]}	raises
Some(1).inspect	Some(1)	Some(1)

The #75 key position closes through to_s alone: the json gem's generator and ActiveSupport's Hash#as_json both stringify a key with key.to_s. No to_str is added. One line in #75 did not reproduce as filed: with ActiveSupport loaded, {Some(1) => 2}.to_json already raised on 0.9.0, because ActiveSupport 8.1 runs a key through as_json before the generator sees it. The same call raises on this branch either way.

Gates, all run locally in the branch's devshell on 2026-09-10:

• bundle exec rubocop: 6 offenses, all present on main at 43bbd0a and none on a changed line. See open items.
• bundle exec yard doctest: 174 runs, 436 assertions, 0 failures, 0 errors.
• bundle exec rake test: 132 runs, 464 assertions, 0 failures, 0 errors.
• bundle exec rake test:strict: 132 runs, 464 assertions, 0 failures, 0 errors.
• bundle exec rake: the three suites above, all passing.
• nix build .#errgonomic: built errgonomic-0.9.1; the check phase ran the doctests.
• nix flake check --all-systems: all checks passed.

CI on this branch ran yard doctest, rake test and rake test:strict on Ruby 3.4 and passed on 2026-09-10 (run 34495870058).

What is left open

• A None in a bare ERB <%= %> renders as nothing under the Rails integration, where a Some raises. None#nil? answers true there, and ActionView's output buffer skips a nil before it asks for to_s. The test records that as it is. Recommendation: accept here. It is the nil? compromise the README names, not to_s, and a raise there means revisiting nil?.
• rubocop reports six offenses on main that this branch does not touch: Metrics on Option#method_missing, Result#method_missing and optional_dig, and Style/Documentation on Errgonomic::Rails. CI does not run rubocop. Recommendation: fix in their own commit outside this hotfix.
• Result#to_json and Result#as_json still say cannot serialize an unwrapped Result without naming the value, where Option's refusal names it. Recommendation: fix outside this hotfix.
• After the merge, the 0.9.1 gem is cut from main with rake release.

After review

Review on 2026-09-10 found three places where a raising to_s reads badly and two test gaps. Following the driver's ruling that such seams are documented rather than patched, this branch now carries:

• Result#unwrap_err! on an Ok builds its message from the value's bounded inspect, so Ok(Some(1)).unwrap_err! prints Some(1) where it printed a bare class name and its message raised. It was the only place in the gem that passed a value as an exception message. Changelog bullet under 0.9.1.
• The README says what a multi-branch case/in that matches nothing on a wrapper looks like, a bare NoMatchingPatternError and a Minitest reporter crash with no test name, and that the fix is the missing in branch or an else. It also says logger.info(opt) raises under Rails' TaggedLogging once a tag such as request_id is set, so it can pass in tests and raise in production, and to write logger.info(opt.inspect). The upgrading note names both, and a test pins each behavior the text describes.
• The two hash-key to_json assertions that already raised on 0.9.0 with ActiveSupport loaded are replaced by a test in a child process that loads only the json gem, where 0.9.0 rendered {"Some(1)":2}. The as_json/to_json refusal misses Option hash KEYS — wrapper ships to payloads silently (0.9.0 regression) #75 lines [Some(1), None()].join(",") and a converted model's group_by(&:author).to_json are tests as the issue writes them.

Gates after the fixes, 2026-09-10: rubocop 6 offenses, the same six; yard doctest 174 runs, 438 assertions; rake test and rake test:strict 136 runs, 480 assertions each; nix build .#errgonomic and nix flake check --all-systems pass. CI passed on 3c8ff50.