Fix multiline interpolation and exact Rust output

[konard]

Problem

Multiline template interpolations could be treated as shell source by this
branch's special-case writer. Backticks, dollar signs, quotes, backslashes, and
newlines were therefore corrupted or executed when used in commands such as:

await $`echo "${complexContent}" > ${testFile}`;

The branch also diverged from the generic literal-interpolation design now on
main, and its fallback path broke multiline program arguments such as
node -e ${script}.

Fixes #37.

Solution

• merge current main and use its context-aware rule that every interpolated
  value is one literal argument, equivalent to a quoted shell variable
• remove the obsolete multiline/base64 virtual-writer changes during conflict
  resolution rather than maintain a second quoting path
• preserve normal shell behavior, including platform-specific echo option
  handling; use portable printf '%s' when no trailing newline is required
• drain Rust child stdout and stderr concurrently as byte chunks, avoiding pipe
  deadlocks and preserving whether either stream ended with a newline
• snapshot JavaScript shell settings when an external invocation starts so
  concurrent test cleanup cannot change an in-flight command's error behavior
• document printf '%s', stdin, and direct filesystem writes for exact text or
  binary output, with an executable JavaScript example
• retain the existing quoteContext, preQuotedPassthrough, and raw controls;
  no multiline-specific compatibility mode is added

The full issue payload now survives unchanged as data. Shell substitutions in
interpolated values do not execute.

The source-level capture change is intentionally Rust-only because the
JavaScript runner already preserves missing final newlines. The PR therefore
uses the repository's parity-exempt label; behavioral parity is asserted in
both language test suites.

Reproduction and regression coverage

New JavaScript and Rust suites cover:

• the complete issue payload with backticks, quotes, dollar signs, backslashes,
  and embedded newlines
• echo and echo -n shell parity, plus portable printf '%s' exact output
• redirection, append, paths containing spaces, and stdin writes
• the reported quoted echo redirection form
• multiline node -e / nested-shell programs remaining one argument
• command-substitution and command-separator injection attempts
• differential parity with /bin/sh using a quoted variable
• exact Rust stdout and stderr capture without invented final newlines, using
  portable Unix and Windows fixtures
• timeout-bounded concurrent capture with stderr output exceeding typical pipe
  capacity, guarding against stdout/stderr drain deadlocks

experiments/issue-37-multiline-competitors.mjs provides a reproducible
comparison with /bin/sh, command-stream, Bun Shell, zx, and execa. All five
matched for the tested multiline payload.

This is not a UI change, so screenshots do not apply.

Verification

• bun run --cwd js lint
• bun run --cwd js format:check
• bun run --cwd js check:duplication
• bun test js/tests/ --timeout 10000 — 1,421 passed, 6 skipped, 0 failed
• Node 20 ESM/CommonJS load checks and both Node artifact test files — passed
• cargo fmt --all -- --check
• RUSTFLAGS=-Dwarnings cargo clippy --all-targets --all-features -- -D warnings
• RUSTDOCFLAGS=-Dwarnings cargo doc --no-deps --all-features
• cargo test --all-features plus doc tests — passed
• release build, cargo package --allow-dirty, inline release-script tests,
  file-size check, and crate-size check — passed
• JavaScript changeset and Rust changelog-fragment checks — passed
• multiline example and competitor experiment — passed

Two fixed-wait SIGINT tests also surfaced as flaky under the complete suite;
both now wait for the child process's readiness message before signaling, and
failed fixtures clean up their child processes.

[konard]

We should double check more cases similar like this, make sure we have test coverage similar to all our competitors, and select behavior closer to how it would behave in sh scripts or with least surprise based on best practices from competitors. If there multiple options we should allow to configure, and use closer to sh behavior by default.

[konard]

🤖 AI Work Session Started

Starting automated work session at 2026-09-13T15:04:14.990Z

The PR has been converted to draft mode while work is in progress.

This comment marks the beginning of an AI work session. Please wait for the session to finish, and provide your feedback.

[konard]

Working session summary

Investigation is complete; no code has been changed yet. The branch remains clean.

Root causes:

• PR 111’s multiline special case writes files directly through _write_multiline_content, changing normal echo behavior.
• For every other multiline command, it inserts the value unquoted. This causes the recorded CI failure at yes-command-cleanup.test.mjs:171: /bin/sh executes the embedded JavaScript as shell syntax.
• The archived CI logs expired, but GitHub annotations preserved the failure. It reproduces locally: 6 pass, 1 fail.
• Current main already fixes the underlying problem generically through context-aware literal interpolation from PR 100 and PR 201. The failing test passes there, as does the exact issue reproduction.

Recommended design:

1. Merge current `main` and resolve conflicts by retaining its JS/Rust architecture and generic quoting implementation.
2. Remove the obsolete base64 virtual-writer approach from this PR.
3. Preserve shell-like defaults:
   - Interpolated multiline text remains one literal argument.
   - `$`, backticks, quotes, backslashes, and newlines are not re-evaluated.
   - `echo` adds its normal trailing newline.
   - Document `printf %s` or filesystem/stdin APIs when byte-exact writing is required.
4. Reuse existing compatibility controls (`quoteContext`, `preQuotedPassthrough`, and `raw`) rather than introduce another multiline-specific mode.
5. Add issue-focused JS and Rust regression coverage for:
   - The full reported special-character example.
   - `echo`, `echo -n`, `printf`, append redirection, and paths with spaces.
   - The previously failing `node -e ${script}` case.
   - Injection attempts and `/bin/sh` differential parity.
6. Add a reproducible competitor experiment covering the repository’s named competitors: Bun, zx, and execa. Their documented defaults likewise treat interpolated values as literal data; zx and Bun provide explicit escape hatches. [Bun Shell](https://bun.sh/docs/runtime/shell), [zx quoting](https://google.github.io/zx/quotes), [execa](https://github.com/sindresorhus/execa).
7. Update JS/Rust best-practice documentation and add the required release fragments.
8. Run all local JS/Rust CI-equivalent checks, commit atomically, push only `issue-37-e875842b`, update [PR 111](https://github.com/link-foundation/command-stream/pull/111), verify fresh CI timestamps/SHA, review the final diff, and mark it ready.

Alternative designs—porting the virtual writer or introducing a second multiline mode—would duplicate current quoting machinery and produce less shell-like behavior.

Please reply yes to approve this design and allow implementation.

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Model: GPT-5.6 Sol
• Provider: OpenAI
• Public pricing estimate: $3.403969

📊 Context and tokens usage:

• 151.7K / 200K (76%) input tokens, 15.9K / 128K (12%) output tokens

Total: (151.7K + 2.1M cached) input tokens, 15.9K output tokens, $3.403969 cost

🤖 Models used:

• Tool: OpenAI Codex
• Requested: gpt-5.6-sol
• Thinking level: xhigh (~31999 tokens)
• Model: GPT-5.6 Sol (gpt-5.6-sol)

📎 Log file uploaded as Gist (1631KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

🔄 Auto-restart 1/5

Reason: CI failures detected; Merge conflicts

Starting new session to address the issues.

---

Auto-restart-until-mergeable mode is active. This run will stop after 5 restart iterations in total.

[konard]

Working session summary

Resolved and pushed to PR #111.

• Merge conflicts resolved against latest main.
• Fixed exact Rust stdout/stderr capture and pipe deadlock handling.
• Fixed JavaScript in-flight shell-setting race.
• Added comprehensive multiline, portability, and regression tests.
• All 27 executable CI checks pass on commit 5fb8388; expected release/parity jobs were skipped.
• PR is mergeable, clean, and ready for review.
• Working tree is clean and matches the remote branch.

---

This summary was automatically extracted from the AI working session output.

[konard]

🔄 Auto-restart-until-mergeable Log 1/5

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Model: GPT-5.6 Sol
• Provider: OpenAI
• Public pricing estimate: $40.955098

📊 Context and tokens usage:

• 963.4K / 200K (482%) input tokens, 128.6K / 128K (100%) output tokens

Total: (963.4K + 36.7M cached) input tokens, 128.6K output tokens, $40.955098 cost

🤖 Models used:

• Tool: OpenAI Codex
• Requested: gpt-5.6-sol
• Model: GPT-5.6 Sol (gpt-5.6-sol)

📎 Log file uploaded as Gist (16960KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

🎉 Auto-merged

This pull request has been automatically merged by hive-mind.

• All CI checks have passed

---

Auto-merged by hive-mind with --auto-merge flag