[cscheid]

Closes #939, closes #940, closes #941, closes #942.

Four Lua API gaps reported by @atheriel while porting the Posit Connect docs from Quarto 1.9.32 to q2 0.36.0. All four are things a Quarto 1 filter does on its first few lines, and all four reproduced on main at 3f7d4b0. Plan with findings and design: claude-notes/plans/2026-10-27-lua-api-q1-parity-gh-939-942.md (epic bd-pqj37chw; one --no-ff merge per sub-task).

What changes

#942 — quarto.version is a real pandoc.types.Version (bd-5qepzwst)
quarto.version was the literal table {0, 1, 0}. It is now built from quarto_util::cli_version() (nightly override included, trimmed to the leading dotted-numeric part as Q1 does) as a new LuaVersion userdata behind pandoc.types.Version: comparable with strings on either side (quarto.version >= "1.4"), tostring, must_be_at_least, 1-based component indexing, and table.concat(quarto.version, ".") keeps working (Lua 5.4's table.concat accepts a userdata with __index + __len; pinned by test). PANDOC_VERSION / PANDOC_API_VERSION are Versions too; pandoc.utils.type reports Version.

#939 — quarto.project and QUARTO_PROFILE (bd-ip1lrgra)
New HostContext on pampa's LuaOptions (input/output file, project directory / output directory / offset, active profiles, environment snapshot), filled by quarto-core's new lmd::lua_options_for_document for both the user-filter stage and the shortcode transform. quarto.project has Q1's shape: directory / offset / output_directory nil outside a project, profile always a pandoc.List. offset is Q1's projectOffset (relative path back to the root, forward slashes, "." at the root).

os.getenv now reads from an environment snapshot the host takes once in StageContext::new, on the thread that builds the context, before any script runs: the runtime's environment, then _environment file values for keys the real environment lacks, then QUARTO_PROFILE forced to the active list. Rationale: filters run on tokio worker threads under block_in_place, and libc getenv there races any setenv in the process; a snapshot also gives every script in a render one consistent environment. Without a host (pampa tests, conformance suite) the stdlib os.getenv is left alone. NativeRuntime::env_all switched to vars_os so a non-UTF-8 entry no longer panics.

#940 — quarto.doc.input_file / quarto.doc.output_file and a minimal PANDOC_STATE (bd-diu7rtes)
Absolute paths, as in Q1, in both the filter and shortcode environments. PANDOC_STATE = { input_files = pandoc.List{input}, output_file }; the other pandoc fields stay nil on purpose so a filter that needs one fails at its own line rather than reading a plausible wrong default.

#941 — os.exit fails the document, not the process (bd-9u10sb4b)
os.exit on the os table (native stdlib or WASM synthetic) now raises an external LuaExit { code } error with Lua's code mapping; apply_lua_filter reports it as LuaFilterError::Exit: Lua filter <path> called os.exit(<code>); the document was not rendered. The rest of the project renders, later errors are still reported, the summary prints, and the run exits non-zero. os.exit(0) also fails the document (Q1 parity: pandoc exiting without output was a failed render). pcall(os.exit, 1) catches it, like any error. error() is unchanged.

Docs: three new sections in docs/guides/authoring/lua-filters.qmd (where the filter is running; checking versions; failing a render).

Verification

• cargo xtask verify: all steps green, including the hub-client tests against a WASM rebuilt from this branch (one smoke fixture, extensions/builtin-version-shortcode, hard-coded 0.1.0 and was updated to expect the real version; it is checked by both the native and the WASM smoke runs).
• WASM Lua tests (cargo test -p pampa --test wasm_lua --target wasm32-unknown-unknown ...): 10/10, including the extended synthetic-os test (os.getenv, os.exit).
• New tests: pampa unit tests (version.rs, host_context.rs, os_exit.rs), quarto-core tests for the snapshot layers and the offset, CLI e2e lua_host_context_cli.rs (two-page website under --profile with an _environment file; single-file render) and lua_os_exit_cli.rs (the issue's a/b/c project; os.exit(0)).
• The four issue repros re-run by hand against the built binary; each now prints what the issue's "Expected" section asks for.
• Not verified: Windows. cargo clippy --target x86_64-pc-windows-msvc cannot run from macOS for pampa, quarto-core or quarto (aws-lc-sys, wasmtime, tree-sitter compile C for the target). The new code uses no platform APIs; path assertions in the e2e tests are built with Path::join. A Windows run of cargo clippy -p pampa -p quarto-core -p quarto --all-targets -- -D warnings and the two new e2e tests would be welcome.

Follow-ups (not filed)

• The rest of PANDOC_STATE (resource_path, verbosity, log, ...).
• EnvShortcodeHandler could read the same snapshot instead of its lookup-time merge.
• Siblings pampa: plural possessive apostrophes (Dogs' toys) are reported as Q-2-10 and fail the render #943–project-level filters: paths are resolved relative to each document directory, so pages in subdirectories fail #945 are separate issues with separate causes; note that project-level filters: paths are resolved relative to each document directory, so pages in subdirectories fail #945 (filter path resolution) is why the new e2e fixture names its filter by absolute path.

🤖 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.