feat(jupyter,quarto-core): cell-options facility + error policy (bd-ohvl879u)

[cscheid]

Summary

Follow-up to #360 (bd-gthycd33). Fixes bd-ohvl879u: q2's jupyter engine unconditionally embedded a failing cell's error as .cell-output-error output and reported success, while knitr fails the whole render (Q1's default execute: error: false policy). The fix rides on a new shared cell-options facility — the design center of this PR.

The facility: quarto-core/src/cell_options/

q2's single implementation of identifying and extracting the #|-style YAML options block at the head of an executable code cell (Q1 grew four divergent copies; q2 had two partial ad-hoc ones — the crossref shorthand's string matcher and the LSP's highlight-only directive_tokens):

• Language-aware registry ported from Q1's kLangCommentChars: #| for python/R, --| for lua/sql, //| for js/rust, %| matlab, block-comment forms (/*| … */ for C, (*| … *) OCaml, *| … ; SAS); unknown → #.
• Partition mirrors Q1's partition-cell-options.ts: leading-run only, column-0 anchored, prefix + ws + | + one optional space, suffix required+elided for block-comment languages.
• Exact source attribution: the reassembled YAML's provenance is SourceInfo::concat of per-line substrings; for prefix-only languages the ranges run through each line's newline, so every byte of the reassembled YAML is a real source byte — map_offset on any parsed node resolves exactly. Parsed with quarto_yaml::parse_with_parent.
• Scoped resolution (new vs Q1): cell options convert to ConfigValue (pampa::yaml_to_config_value) and merge over the document's already-merged execute metadata (MergedConfig, cell wins) — the start of scoped document-metadata resolution in cell options, a longstanding Q1 limitation.

The engine consumer

text_execute.rs now partitions each cell and:

• strips option lines from kernel input and echoed .cell-code (Q1/knitr parity — verified empirically that knitr strips them; cell magics would otherwise break);
• resolves error: via the scoped merge: per-cell #| error: true ⇒ embed error output; document execute: error: true ⇒ default-allow; cell error: false overrides a document allow; absent everywhere ⇒ the render fails (knitr/Q1 parity), halting before later cells run;
• malformed option YAML is a hard, located error.

Diagnostics resolve to original-file coordinates through ExecutionContext.source_info (no new plumbing needed), e.g.:

Error: Execution failed in jupyter: code cell at …/fail.qmd:9:2 raised Exception: boom-e2e
Use `#| error: true` on the cell (or `execute: error: true` in the document metadata) to show the error in the output instead.

Tests

• 28 facility unit tests (registry / partition / source-mapping / scoped-resolution matrix), TDD reds confirmed before implementation.
• engine_error_policy.rs: 7 kernel-gated integration tests covering the full policy matrix.
• engine_output_parity.rs gains a behavioral parity case (both engines must fail identically on un-annotated errors); all bd-gthycd33 fences unchanged and green.
• Workspace suite 9924/9924 on the rebased tree; full cargo xtask verify passed; clippy/lint/fmt clean.
• E2E through the real binary: all three policy paths inspected (q2 render), plus a q2 preview browser spot-check (directive-bearing healthy cell: stripped echo, spliced output) after the full WASM chain rebuild.

Follow-ups filed (discovered-from bd-ohvl879u): bd-eizgnxlx (migrate crossref shorthand to the facility), bd-1gty7f7o (LSP reuse), bd-2xkpy5ra (body-only CodeBlock SourceInfo), bd-moef1ec4 (eval option), bd-2lc8qu6e (guessChunkOptionsFormat).

Plan with full evidence: claude-notes/plans/2026-07-02-bd-ohvl879u-cell-options-error-policy.md.

🤖 Generated with Claude Code

[posit-snyk-bot]

✅ Snyk checks have passed. No issues have been found so far.

Status	Scan Engine	Critical	High	Medium	Low	Total (0)
✅	Open Source Security	0	0	0	0	0 issues
✅	Licenses	0	0	0	0	0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.