fix(parser): scanner drops every entity when HEADER FILE_NAME contains '#' (#654) (#657)

CATIA emits a FILE_NAME with a literal `#` inside the quoted filename
(`'…\X0\2#.ifc'`). EntityScanner's `memchr('#')` anchor latched onto
the in-string `#`, then find_entity_end started with the wrong
quote-parity assumption and skipped every `;` for the rest of the
file. The scanner returned 0 entities — IFCLite Viewer showed
properties but no geometry.

Parser fix: skip past `DATA;` on construction (quote-aware, handles
`''` escaped pairs), and require `#` to be followed by a digit before
treating it as an entity anchor. Matches the validation already in
`build_entity_index` and the TS StepTokenizer. Three regression tests.

Build pipeline: stop committing the wasm bundles to git. The original
symptom for #654 was stale `.wasm` shipping to production, and during
this PR we hit it twice — once when cargo's incremental cache reused
stale `ifc-lite-core`, once when a merge-conflict resolution picked
the wrong bytes. Pipeline now rebuilds wasm from Rust source on every
deploy: turbo wires the wasm packages into the workspace build graph;
Vercel install/build scripts bootstrap the Rust toolchain and
propagate the env through turbo subprocesses; the Desktop
Compatibility CI workflow installs Rust + wasm-pack too. Most deploys
are JS-only and hit turbo's content cache, paying ~0 s extra;
Rust-touching deploys add ~30-60 s warm.

Verified end-to-end against CatiaBuilding.ifc:
  0 → 83 375 entities scanned
  0 → 769/770 walls with mesh; all doors/slabs/windows with mesh

Closes #654.

Author: louistrue

.github/workflows/desktop-compat.yml

@@ -8,11 +8,20 @@ name: Desktop Compatibility
 on:
   pull_request:
     paths:
-      # Only run when shared code that the desktop app consumes changes
+      # Run when shared code the desktop app consumes changes, OR when the
+      # Rust sources that produce the WASM bundle change (we no longer
+      # commit the .wasm — see #654 — so the desktop frontend build has to
+      # rebuild it from source via turbo).
       - 'apps/viewer/src/**'
       - 'packages/*/src/**'
-      - 'packages/wasm/pkg/**'
       - 'apps/desktop/**'
+      - 'rust/core/src/**'
+      - 'rust/geometry/src/**'
+      - 'rust/wasm-bindings/src/**'
+      - 'rust/**/Cargo.toml'
+      - 'Cargo.lock'
+      - 'rust-toolchain.toml'
+      - 'scripts/build-wasm.sh'
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
@@ -42,6 +51,20 @@ jobs:
           node-version: 22
           cache: pnpm
 
+      # WASM bundles are no longer committed (see #654); turbo's wasm
+      # packages rebuild from source via scripts/build-wasm.sh, which
+      # needs a Rust toolchain matching rust-toolchain.toml + wasm-pack.
+      - name: Setup Rust (pinned by rust-toolchain.toml)
+        run: rustup show
+
+      - name: Cargo cache
+        uses: Swatinem/rust-cache@v2
+
+      - name: Install wasm-pack
+        uses: jetli/wasm-pack-action@v0.4.0
+        with:
+          version: latest
+
       - name: Install dependencies
         run: pnpm install --frozen-lockfile
 

.gitignore

@@ -45,6 +45,21 @@ test_*.sh
 # WASM files - use packages/wasm/pkg/ as canonical source
 apps/viewer/public/ifc-lite_bg.wasm
 
+# WASM build outputs — not committed; produced from rust/** by
+# scripts/build-wasm.sh on every fresh build. CI and Vercel install the
+# Rust toolchain and rebuild from source so a stale binary can never
+# ship (issue #654 root cause was a stale committed .wasm bundle).
+# The README.md and package.json in these dirs stay tracked.
+packages/wasm/pkg/ifc-lite_bg.wasm
+packages/wasm/pkg/ifc-lite_bg.wasm.d.ts
+packages/wasm/pkg/ifc-lite.js
+packages/wasm/pkg/ifc-lite.d.ts
+packages/wasm-threaded/pkg/ifc-lite_bg.wasm
+packages/wasm-threaded/pkg/ifc-lite_bg.wasm.d.ts
+packages/wasm-threaded/pkg/ifc-lite.js
+packages/wasm-threaded/pkg/ifc-lite.d.ts
+packages/wasm-threaded/pkg/snippets/
+
 # Tauri/Claude temp files
 tmpclaude*
 nul

README.md

@@ -235,9 +235,17 @@ Ready-to-run projects in [`examples/`](examples/):
 
 ## Contributing
 
-No Rust toolchain needed — WASM comes pre-built.
+The WASM bundle is built from `rust/` on every fresh build, so a Rust
+toolchain is required. `rust-toolchain.toml` pins the nightly channel
+and the `wasm32-unknown-unknown` target — `rustup show` (or the
+contributing setup guide) installs everything needed.
 
 ```bash
+# 1. Rust toolchain (one-time)
+curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
+cargo install wasm-pack   # or: curl https://rustwasm.github.io/wasm-pack/installer/init.sh -sSf | sh
+
+# 2. Clone and build
 git clone https://github.com/louistrue/ifc-lite.git
 cd ifc-lite
 pnpm install && pnpm build && pnpm dev   # opens viewer at localhost:5173

apps/viewer-embed/vercel.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://openapi.vercel.sh/vercel.json",
-  "buildCommand": "cd ../.. && pnpm install --frozen-lockfile && npx turbo build --filter=@ifc-lite/viewer-embed...",
-  "installCommand": "echo 'handled by buildCommand'",
+  "installCommand": "cd ../.. && bash scripts/vercel-install.sh",
+  "buildCommand": "cd ../.. && bash scripts/vercel-build.sh '@ifc-lite/viewer-embed...'",
   "outputDirectory": "dist",
   "framework": "vite",
   "headers": [

packages/wasm-threaded/package.json

@@ -8,6 +8,9 @@
   ],
   "description": "Thread-enabled WebAssembly bindings for IFC-Lite (uses wasm-bindgen-rayon for in-WASM parallelism). Requires cross-origin isolation (COOP/COEP). Generated by `THREADED=1 scripts/build-wasm.sh`.",
   "version": "0.1.0",
+  "scripts": {
+    "build": "THREADED=1 bash ../../scripts/build-wasm.sh"
+  },
   "license": "MPL-2.0",
   "repository": {
     "type": "git",