Keep the stderr drain thread off rspec doubles in connect specs

[ya-luotao]

What and why

CI on main failed once, at 07ad108 (between the merges of #158 and #159), on the Ruby 4.0 leg:

rspec ./spec/unit/subprocess_cli_transport_spec.rb:921
  expected block to not output to stderr, but output
  "#<Thread:0x… subprocess_cli_transport.rb:401 run> terminated with exception …"

The output came from a different example. Nine popen3 stubs (22 #connect calls over three helpers and
six inline examples in subprocess_cli_transport_spec.rb, plus the helper in query_run_end_spec.rb)
hand #connect an instance_double(IO) as stderr. #connect starts a thread that drains stderr, and these
examples never close the transport, so the thread outlives them:

• if it runs during its own example, #each_line is an unexpected message;
• if it runs between examples, rspec-mocks raises OutsideOfExampleError, a StandardError that the drain
  thread's rescue absorbs (silently);
• if it runs during a later example, the double has expired and rspec-mocks raises ExpiredTestDoubleError.

Both MockExpectationErrors are Exceptions, not StandardErrors. The thread dies with one and Ruby
prints the report on $stderr in whichever example is running at that moment. When that example is
inside an output.to_stderr matcher, it fails.

The leak predates this batch (the stubs are from 2026-02 to 2026-10-01). It shows up on Ruby 4.0, which CI
has run since #153: the green run at 1e56e56 printed the report 16 times in test (ubuntu-latest, Ruby 4.0)
and 10 times in deps (latest, Ruby 4.0), and 0 times on Ruby 3.x. Every report names the drain thread
(subprocess_cli_transport.rb:401). The v1.1.0 run on main printed none.

Change

Specs only.

• The stubs return StringIOs for stdout and stderr, as the #read_messages examples in the same file
  already do. The drain reads EOF and ends whenever it runs, and no rspec object is involved.
• They connect through a new connect_draining_stderr(transport) helper in spec/support/test_helpers.rb:
  #connect, then a bounded join of the drain thread. If a double comes back, its own example fails
  deterministically instead of an unrelated later one (on Ruby 3.4.5: with the join and the old doubles,
  the 21 examples behind those 22 calls fail with received unexpected message :each_line; with StringIOs, 0).
• spec/README.md: a convention that no background thread may read an rspec double, and the helper.

The drain thread's rescue StandardError in lib/ stays as it is: MockExpectationError < Exception is
a test-only fact.

Verification

• Red: with the join and the old doubles, 174 examples, 21 failures on Ruby 3.4.5. Green: with StringIOs,
  the two spec files pass (174 examples, 0 failures) and no thread reports anything.
• The suite on the locked deps, gemfiles/latest.gemfile and gemfiles/floor.gemfile; rake rbs:test;
  RuboCop: 3574 examples, 0 failures on each of the three; 3206, 0 failures under the RBS runtime checker; no offenses.
• This PR's CI (run 37108307257, all 12 jobs green): terminated with exception appears 0 times in all 12 job logs, leaked into another example 0 times. At 1e56e56 the same two Ruby 4.0 jobs printed it 16 and 10 times.

Checklist

• bundle exec rake passes (specs + RuboCop)
• Specs cover the change (the helper turns a regression into a failure of its own example)
• CHANGELOG entry: none, the change is test-only
• Docs updated: spec/README.md
• No file under lib/ changes

🤖 Generated with Claude Code

https://claude.ai/code/session_01CNgMt2RjeBLyRP3CWFWNy2

[ya-luotao]

@codex review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-03T08:03:54.862495Z	815149f	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Another round soon, please!

Reviewed commit: 815149f439

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".