feat: ✨ Add WASM web demo and fix wasm conversion panics

[Sube-py]

No description provided.

[developer0hye]

Code Review — PR #133

Thanks for the contribution! The WASM web demo and panic-safety improvements are useful, but there are several issues that need to be addressed before merging. Detailed inline comments below.

---

CRITICAL

1. crate-type = ["cdylib", "rlib"] applied unconditionally

File: crates/office2pdf/Cargo.toml

Adding cdylib to the default crate-type affects every build target, not just WASM. On native builds, Cargo will now produce an additional shared library artifact (*.dylib / *.so / *.dll), increasing build time and disk usage for all users who never use WASM.

Cargo does not support conditional crate-type, so the options are:

• Remove cdylib from the lib crate and create a separate crates/office2pdf-wasm crate (thin wrapper that re-exports with crate-type = ["cdylib"]) — this is the standard pattern used by projects like typst, wasm-bindgen examples, etc.
• Or document that cdylib is only meaningful when targeting wasm32-unknown-unknown and accept the minor native overhead.

The first option is strongly preferred since it keeps the lib crate clean.

2. Indentation error in streaming path

File: crates/office2pdf/src/lib.rs (around the streaming single-chunk path)

            metrics: Some(ConvertMetrics {
                parse_duration,
            codegen_duration: Duration::ZERO,  // ← wrong indent (de-indented by 4 spaces)
            compile_duration: Duration::ZERO,  // ← wrong indent
                total_duration,

This is a formatting error — cargo fmt should have caught it. Per project rules, cargo fmt -- --check must pass before committing.

---

HIGH

3. Unnecessary Cargo.toml reformatting

File: crates/office2pdf/Cargo.toml

The diff converts every inline array (keywords = ["a", "b"]) to multiline format. This changes 40+ lines with zero functional impact, inflating the diff and making the actual changes harder to review.

Please revert all formatting-only changes and keep only the lines that add new dependencies or features (console_error_panic_hook, js-sys, crate-type).

4. New external dependency without justification

File: crates/office2pdf/Cargo.toml — console_error_panic_hook = { version = "0.1", optional = true }

Project rules (CLAUDE.md): "Always prefer Rust native implementations. Only use third-party crates when there is a clear, justified need."

console_error_panic_hook is a tiny crate (just calls std::panic::set_hook with web_sys::console::error), but the PR doesn't explain why a manual 5-line set_hook + web_sys::console::error_1 wouldn't suffice, or why web_sys (already transitively depended on via typst) can't be used directly.

Either justify the dependency or replace with an inline implementation.

---

MEDIUM

5. WASM timing always returns Duration::ZERO — metrics are silently broken

File: crates/office2pdf/src/lib.rs

#[cfg(target_arch = "wasm32")]
type TimingMark = ();

#[cfg(target_arch = "wasm32")]
fn start_timing() -> TimingMark {}

#[cfg(target_arch = "wasm32")]
fn elapsed_since(_start: TimingMark) -> Duration {
    Duration::ZERO
}

On WASM, ConvertMetrics will report parse_duration = 0, codegen_duration = 0, compile_duration = 0, total_duration = 0 for every conversion. This is misleading — users calling convert_bytes() on WASM and inspecting metrics get zeroes with no indication they're fake.

Options:

• Use js_sys::Date::now() (already added as a dependency) for actual millisecond-precision timing on WASM.
• Or make ConvertMetrics optional (metrics: None) on WASM so callers know timing is unavailable.

Zero durations that aren't actually zero is the worst option — it's silent data corruption.

6. catch_unwind added around render pipeline without explanation

File: crates/office2pdf/src/lib.rs (Stage 2+3 block)

The render pipeline (typst codegen + PDF compilation) was intentionally not wrapped in catch_unwind — only the parsing stage was, because upstream parsers (docx-rs, umya-spreadsheet) are known to panic on malformed input. Typst's compilation pipeline is not expected to panic.

Adding catch_unwind here:

• Masks potential bugs in our own codegen (panics should propagate during development so we fix them)
• Has a (small) performance cost
• Changes the error contract (previously: codegen panics crash the process; now: they're silently converted to ConvertError::Render)

If WASM requires this because wasm-bindgen can't handle unwinding, please explain that explicitly in a code comment. If it's just "defensive programming", it should be removed — we prefer to fix panics rather than catch them.

7. WASM always includes PDF timestamp, native does not

File: crates/office2pdf/src/render/pdf.rs

#[cfg(target_arch = "wasm32")]
let timestamp = Some(typst_pdf::Timestamp::new_utc(current_utc_datetime()));

#[cfg(not(target_arch = "wasm32"))]
let timestamp = {
    let needs_timestamp = pdf_standard.is_some() || pdf_ua;
    // ...
};

On native, timestamp is only included for PDF/A and PDF/UA. On WASM, it's always included. This behavioral divergence is not documented. Why should WASM PDFs always have timestamps? If Typst requires it on WASM for some reason, add a comment explaining why.

---

LOW

8. Demo WASM path is fragile

File: crates/office2pdf/examples/wasm-web/app.js

const resolvedWasmUrl = new URL("../../pkg/office2pdf_bg.wasm", import.meta.url);

This assumes wasm-pack build --target web was run from crates/office2pdf/ and output to crates/office2pdf/pkg/. The relative path ../../pkg/ will break if:

• The demo is served from a different directory
• wasm-pack output path changes
• Someone copies the demo elsewhere

Consider making the path configurable or at minimum add a comment explaining the expected directory layout.

9. README changes should be scoped to the example

The WASM demo instructions (wasm-pack build, python3 -m http.server) are useful but belong in crates/office2pdf/examples/wasm-web/README.md, not the project root README. The main README should link to the example directory, not contain step-by-step demo setup instructions.

---

Summary

#	Severity	Issue
1	CRITICAL	cdylib crate-type affects all native builds
2	CRITICAL	Indentation error in streaming metrics
3	HIGH	Unnecessary Cargo.toml reformatting (40+ noise lines)
4	HIGH	External dependency added without justification
5	MEDIUM	WASM metrics silently return zero
6	MEDIUM	catch_unwind on render pipeline without justification
7	MEDIUM	Behavioral divergence: WASM always includes timestamp
8	LOW	Fragile relative WASM path
9	LOW	Demo instructions in root README

Please address at least the CRITICAL and HIGH items. Happy to re-review once updated.