feat(multimodule): per-module split + tw link — the killer feature over genuinely separate binaries (#140)

crates/typed-wasm-codegen/README.md

@@ -58,4 +58,5 @@ cargo test -p typed-wasm-codegen
 
 - **#127** — codegen coverage across all 10 levels × all 6 examples (and the front-end → IR JSON seam).
 - **#130** — promote the round-trip tests into the ECHIDNA property corpus.
+- **Per-module split** — **done**: `module Name { … }` blocks parse into separate modules (`parse_modules`; importers seeded with the producer's actual schema), `tw build --split` emits one wasm per module, `tw link` certifies the graph (`tests/multimodule_split.rs`, `tests/fixtures/multimodule/game.twasm`).
 - **L13 positive-form / region-imports** — **done** (issue #140): `import region … from "…" { … }` parses into `Module::region_imports`, emits the `typedwasm.region-imports` carrier (proposal 0003 `[accepted]` / ADR-0007), and `verify_link_graph` certifies cross-module schema agreement (`tests/example02.rs`).

crates/typed-wasm-codegen/src/bin/tw.rs

@@ -4,10 +4,13 @@
 //! `tw` — the typed-wasm build CLI (codegen v0).
 //!
 //! Usage:
-//!   tw build <file.twasm> [-o <out.wasm>]
+//!   tw build <file.twasm> [-o <out>] [--emit wasm|wat|both] [--split]
+//!   tw link <a.wasm> <b.wasm> …
 //!
-//! v0 supports only the example-01 schema; general `.twasm` front-end →
-//! IR lowering is tracked in ADR-0004 and issue #127.
+//! `--split` emits one wasm per top-level `module Name { … }` block
+//! (`<out>.<module>.wasm`); `link` runs the L13 cross-module link-graph
+//! pass (ADR-0007) over already-built modules, naming each by its file
+//! stem, and reports certificates / violations.
 
 use std::path::Path;
 use std::process::ExitCode;
@@ -16,6 +19,7 @@ fn main() -> ExitCode {
     let args: Vec<String> = std::env::args().collect();
     match args.get(1).map(String::as_str) {
         Some("build") => build(&args[2..]),
+        Some("link") => link(&args[2..]),
         Some("--version") | Some("-V") => {
             println!("tw {}", env!("CARGO_PKG_VERSION"));
             ExitCode::SUCCESS
@@ -33,7 +37,8 @@ fn main() -> ExitCode {
 }
 
 fn usage() {
-    eprintln!("usage: tw build <file.twasm> [-o <out>] [--emit wasm|wat|both]");
+    eprintln!("usage: tw build <file.twasm> [-o <out>] [--emit wasm|wat|both] [--split]");
+    eprintln!("       tw link <a.wasm> <b.wasm> ...");
 }
 
 /// What `tw build` emits.
@@ -65,6 +70,7 @@ fn build(rest: &[String]) -> ExitCode {
     let mut input: Option<String> = None;
     let mut output: Option<String> = None;
     let mut emit = Emit::Wasm;
+    let mut split = false;
     let mut i = 0;
     while i < rest.len() {
         let arg = rest[i].as_str();
@@ -96,6 +102,7 @@ fn build(rest: &[String]) -> ExitCode {
                     return ExitCode::FAILURE;
                 }
             },
+            "--split" => split = true,
             s if !s.starts_with('-') => input = Some(s.to_string()),
             other => {
                 eprintln!("tw build: unknown option '{other}'");
@@ -119,8 +126,11 @@ fn build(rest: &[String]) -> ExitCode {
         }
     };
 
-    // Try to parse the .twasm file using the Rust parser (issue #127).
-    // This parser handles the paint-type schemas and example-01.
+    if split {
+        return build_split(&input, &src, output.as_deref(), emit);
+    }
+
+    // Parse via the canonical Rust front-end (ADR-0006).
     let bytes = match typed_wasm_codegen::parser::parse_module(&src) {
         Ok(module) => {
             let bytes = typed_wasm_codegen::emit(&module);
@@ -171,6 +181,114 @@ fn build(rest: &[String]) -> ExitCode {
     ExitCode::SUCCESS
 }
 
+/// `tw build --split`: one output per `module Name { … }` block.
+fn build_split(input: &str, src: &str, output: Option<&str>, emit: Emit) -> ExitCode {
+    let modules = match typed_wasm_codegen::parser::parse_modules(src) {
+        Ok(m) => m,
+        Err(e) => {
+            eprintln!("tw build: parse error in '{input}': {e}");
+            return ExitCode::FAILURE;
+        }
+    };
+    let base = output.unwrap_or(input);
+    let stem = Path::new(base)
+        .with_extension("")
+        .to_string_lossy()
+        .into_owned();
+    let mut wrote = Vec::new();
+    for (name, module) in &modules {
+        let bytes = typed_wasm_codegen::emit(module);
+        if let Err(diagnostics) = typed_wasm_codegen::self_verify(module) {
+            for msg in diagnostics {
+                eprintln!("tw build: self-verify warning [{name}]: {msg}");
+            }
+        }
+        if emit.wants_wasm() {
+            let path = format!("{stem}.{name}.wasm");
+            if let Err(e) = std::fs::write(&path, &bytes) {
+                eprintln!("tw build: cannot write '{path}': {e}");
+                return ExitCode::FAILURE;
+            }
+            wrote.push(format!("{path} ({} bytes)", bytes.len()));
+        }
+        if emit.wants_wat() {
+            let path = format!("{stem}.{name}.wat");
+            let text = typed_wasm_codegen::wat(&bytes);
+            if let Err(e) = std::fs::write(&path, text.as_bytes()) {
+                eprintln!("tw build: cannot write '{path}': {e}");
+                return ExitCode::FAILURE;
+            }
+            wrote.push(path);
+        }
+    }
+    eprintln!(
+        "tw build: split {} module(s): {}",
+        modules.len(),
+        wrote.join(" + ")
+    );
+    eprintln!("tw build: check cross-module schema agreement with `tw link <files…>`.");
+    ExitCode::SUCCESS
+}
+
+/// `tw link`: the L13 positive-form link-graph pass (ADR-0007) over
+/// built modules. Each module's wasm-level name is its file stem's last
+/// dot-segment (`game.physics.wasm` → `physics`).
+fn link(files: &[String]) -> ExitCode {
+    if files.is_empty() {
+        eprintln!("tw link: no modules given");
+        usage();
+        return ExitCode::FAILURE;
+    }
+    let mut named: Vec<(String, Vec<u8>)> = Vec::new();
+    for f in files {
+        let bytes = match std::fs::read(f) {
+            Ok(b) => b,
+            Err(e) => {
+                eprintln!("tw link: cannot read '{f}': {e}");
+                return ExitCode::FAILURE;
+            }
+        };
+        let stem = Path::new(f)
+            .with_extension("")
+            .file_name()
+            .map(|s| s.to_string_lossy().into_owned())
+            .unwrap_or_else(|| f.clone());
+        let name = stem.rsplit('.').next().unwrap_or(&stem).to_string();
+        named.push((name, bytes));
+    }
+    let graph: Vec<(&str, &[u8])> = named
+        .iter()
+        .map(|(n, b)| (n.as_str(), b.as_slice()))
+        .collect();
+    match typed_wasm_verify::verify_link_graph(&graph) {
+        Ok(report) => {
+            for c in &report.certificates {
+                println!(
+                    "CERTIFIED  {} imports {}.{}",
+                    c.consumer, c.producer, c.region_name
+                );
+            }
+            for e in &report.errors {
+                eprintln!("VIOLATION  {e}");
+            }
+            if report.errors.is_empty() {
+                eprintln!(
+                    "tw link: {} module(s), {} certificate(s), no violations",
+                    named.len(),
+                    report.certificates.len()
+                );
+                ExitCode::SUCCESS
+            } else {
+                ExitCode::FAILURE
+            }
+        }
+        Err(e) => {
+            eprintln!("tw link: wasm parse error: {e}");
+            ExitCode::FAILURE
+        }
+    }
+}
+
 /// `base` with its extension replaced by `ext`
 /// (e.g. `with_extension("foo.twasm", "wasm")` -> `"foo.wasm"`).
 fn with_extension(base: &str, ext: &str) -> String {

crates/typed-wasm-codegen/src/parser.rs

@@ -20,10 +20,127 @@
 use crate::{Field, FieldTy, Memory, Module, PtrKind, Region, Scalar, Wty};
 use std::collections::HashMap;
 
-/// Parse a .twasm source file into a Module IR.
+/// Parse a .twasm source file into a single merged Module IR.
+/// `module Name [isolated] { … }` blocks are flattened into one module
+/// (the historical single-module view the round-trip corpus uses);
+/// [`parse_modules`] gives the per-module split.
 pub fn parse_module(src: &str) -> Result<Module, String> {
-    let parser = Parser::new(src);
-    parser.parse_module()
+    let fragments = slice_modules(src);
+    if fragments.len() == 1 {
+        return Parser::new(src).parse_module();
+    }
+    let merged: String = fragments
+        .into_iter()
+        .map(|(_, frag)| frag)
+        .collect::<Vec<_>>()
+        .join("\n");
+    Parser::new(&merged).parse_module()
+}
+
+/// Parse a multi-module `.twasm` source into one Module per top-level
+/// `module Name [isolated] { … }` block (declarations outside any block
+/// form the `"main"` module, omitted when empty). Two passes: modules
+/// parse independently, then any module importing a region whose
+/// producer is in the same file is RE-parsed with the producer's actual
+/// region schemas seeded in — so cross-module accesses lower against
+/// the producer's real layout, not the expected subset's packing.
+pub fn parse_modules(src: &str) -> Result<Vec<(String, Module)>, String> {
+    let fragments = slice_modules(src);
+    let mut parsed: Vec<(String, Module, Vec<u64>, String)> = Vec::new();
+    for (name, frag) in fragments {
+        let (module, instances) = Parser::new(&frag).parse_with_instances()?;
+        if name == "main"
+            && module.regions.is_empty()
+            && module.funcs.is_empty()
+            && module.region_imports.is_empty()
+        {
+            continue; // no default-module content
+        }
+        parsed.push((name, module, instances, frag));
+    }
+
+    // Pass 2: seed importers with in-file producers' actual schemas.
+    let producer_snapshot: Vec<(String, Module, Vec<u64>)> = parsed
+        .iter()
+        .map(|(n, m, i, _)| (n.clone(), m.clone(), i.clone()))
+        .collect();
+    for (_, module, _, frag) in parsed.iter_mut() {
+        if module.region_imports.is_empty() {
+            continue;
+        }
+        let mut seeds: Vec<(Region, u64)> = Vec::new();
+        for imp in &module.region_imports {
+            if let Some((_, producer, instances)) = producer_snapshot
+                .iter()
+                .find(|(n, _, _)| *n == imp.producer_module)
+            {
+                if let Some(idx) =
+                    producer.regions.iter().position(|r| r.name == imp.region_name)
+                {
+                    seeds.push((
+                        producer.regions[idx].clone(),
+                        instances.get(idx).copied().unwrap_or(1),
+                    ));
+                }
+            }
+        }
+        if !seeds.is_empty() {
+            let mut p = Parser::new(frag);
+            for (region, instances) in seeds {
+                p.region_map.insert(region.name.clone(), p.regions.len());
+                p.regions.push(region);
+                p.region_instances.push(instances);
+            }
+            *module = p.parse_module()?;
+        }
+    }
+
+    Ok(parsed.into_iter().map(|(n, m, _, _)| (n, m)).collect())
+}
+
+/// Split source at top-level `module Name [isolated] { … }` blocks into
+/// `(name, fragment)` pairs; text outside any block accumulates into a
+/// `"main"` fragment (kept even when blank so `parse_module` can join).
+fn slice_modules(src: &str) -> Vec<(String, String)> {
+    let mut out: Vec<(String, String)> = Vec::new();
+    let mut default_frag = String::new();
+    let mut scanner = Parser::new(src);
+    let mut plain_start = 0usize;
+    loop {
+        scanner.skip_whitespace();
+        if scanner.pos >= src.len() {
+            break;
+        }
+        if scanner.peek_word("module") {
+            let block_start = scanner.pos;
+            scanner.pos += "module".len();
+            let name = scanner.parse_ident();
+            scanner.skip_whitespace();
+            let _ = scanner.try_keyword("isolated");
+            scanner.skip_whitespace();
+            if !name.is_empty() && scanner.try_char('{') {
+                default_frag.push_str(&src[plain_start..block_start]);
+                let body_start = scanner.pos;
+                scanner.skip_to_brace_close();
+                // skip_to_brace_close consumed through the closing `}`.
+                let body_end = scanner.pos.saturating_sub(1).max(body_start);
+                out.push((name, src[body_start..body_end].to_string()));
+                plain_start = scanner.pos;
+                continue;
+            }
+            // `module X;` or malformed: leave for the plain fragment.
+            scanner.pos = block_start + "module".len();
+        }
+        // Advance one declaration: consume through it so we never split
+        // inside a nested brace (fn bodies, regions, …).
+        scanner.skip_declaration();
+    }
+    default_frag.push_str(&src[plain_start..]);
+    if out.is_empty() {
+        return vec![("main".into(), default_frag)];
+    }
+    out.insert(0, ("main".into(), default_frag));
+    out
 }
 
 /// A simple hand-written parser for .twasm syntax.
@@ -63,7 +180,15 @@ impl<'a> Parser<'a> {
         }
     }
 
-    fn parse_module(mut self) -> Result<Module, String> {
+    fn parse_module(self) -> Result<Module, String> {
+        self.parse_with_instances().map(|(m, _)| m)
+    }
+
+    /// As [`Self::parse_module`], additionally returning the per-region
+    /// declared instance counts (needed to seed importers in
+    /// [`parse_modules`] so their scans iterate the producer's real
+    /// instance array).
+    fn parse_with_instances(mut self) -> Result<(Module, Vec<u64>), String> {
         while self.pos < self.src.len() {
             self.skip_whitespace();
             if self.pos >= self.src.len() {
@@ -88,14 +213,17 @@ impl<'a> Parser<'a> {
             }
         }
 
-        Ok(Module {
-            regions: self.regions,
-            memory: self.memory,
-            imports: self.imports,
-            funcs: self.funcs,
-            ownership: self.ownership,
-            region_imports: self.region_imports,
-        })
+        Ok((
+            Module {
+                regions: self.regions,
+                memory: self.memory,
+                imports: self.imports,
+                funcs: self.funcs,
+                ownership: self.ownership,
+                region_imports: self.region_imports,
+            },
+            self.region_instances,
+        ))
     }
 
     fn peek_word(&mut self, word: &str) -> bool {