1use crate::ExtensionManifest;
2use crate::{extension_manifest::ExtensionLibraryKind, GrammarManifestEntry};
3use anyhow::{anyhow, bail, Context as _, Result};
4use async_compression::futures::bufread::GzipDecoder;
5use async_tar::Archive;
6use futures::io::BufReader;
7use futures::AsyncReadExt;
8use serde::Deserialize;
9use std::{
10 env, fs, mem,
11 path::{Path, PathBuf},
12 process::{Command, Stdio},
13 sync::Arc,
14};
15use util::http::{self, AsyncBody, HttpClient};
16use wasm_encoder::{ComponentSectionId, Encode as _, RawSection, Section as _};
17use wasmparser::Parser;
18use wit_component::ComponentEncoder;
19
20/// Currently, we compile with Rust's `wasm32-wasi` target, which works with WASI `preview1`.
21/// But the WASM component model is based on WASI `preview2`. So we need an 'adapter' WASM
22/// module, which implements the `preview1` interface in terms of `preview2`.
23///
24/// Once Rust 1.78 is released, there will be a `wasm32-wasip2` target available, so we will
25/// not need the adapter anymore.
26const RUST_TARGET: &str = "wasm32-wasi";
27const WASI_ADAPTER_URL: &str =
28 "https://github.com/bytecodealliance/wasmtime/releases/download/v18.0.2/wasi_snapshot_preview1.reactor.wasm";
29
30/// Compiling Tree-sitter parsers from C to WASM requires Clang 17, and a WASM build of libc
31/// and clang's runtime library. The `wasi-sdk` provides these binaries.
32///
33/// Once Clang 17 and its wasm target are available via system package managers, we won't need
34/// to download this.
35const WASI_SDK_URL: &str = "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-21/";
36const WASI_SDK_ASSET_NAME: Option<&str> = if cfg!(target_os = "macos") {
37 Some("wasi-sdk-21.0-macos.tar.gz")
38} else if cfg!(target_os = "linux") {
39 Some("wasi-sdk-21.0-linux.tar.gz")
40} else {
41 None
42};
43
44pub struct ExtensionBuilder {
45 cache_dir: PathBuf,
46 pub http: Arc<dyn HttpClient>,
47}
48
49pub struct CompileExtensionOptions {
50 pub release: bool,
51}
52
53#[derive(Deserialize)]
54struct CargoToml {
55 package: CargoTomlPackage,
56}
57
58#[derive(Deserialize)]
59struct CargoTomlPackage {
60 name: String,
61}
62
63impl ExtensionBuilder {
64 pub fn new(cache_dir: PathBuf) -> Self {
65 Self {
66 cache_dir,
67 http: http::client(),
68 }
69 }
70
71 pub async fn compile_extension(
72 &self,
73 extension_dir: &Path,
74 extension_manifest: &ExtensionManifest,
75 options: CompileExtensionOptions,
76 ) -> Result<()> {
77 if extension_dir.is_relative() {
78 bail!(
79 "extension dir {} is not an absolute path",
80 extension_dir.display()
81 );
82 }
83
84 fs::create_dir_all(&self.cache_dir).context("failed to create cache dir")?;
85
86 let cargo_toml_path = extension_dir.join("Cargo.toml");
87 if extension_manifest.lib.kind == Some(ExtensionLibraryKind::Rust)
88 || fs::metadata(&cargo_toml_path).map_or(false, |stat| stat.is_file())
89 {
90 log::info!("compiling Rust extension {}", extension_dir.display());
91 self.compile_rust_extension(extension_dir, options)
92 .await
93 .context("failed to compile Rust extension")?;
94 }
95
96 for (grammar_name, grammar_metadata) in &extension_manifest.grammars {
97 self.compile_grammar(extension_dir, grammar_name.as_ref(), grammar_metadata)
98 .await
99 .with_context(|| format!("failed to compile grammar '{grammar_name}'"))?;
100 }
101
102 log::info!("finished compiling extension {}", extension_dir.display());
103 Ok(())
104 }
105
106 async fn compile_rust_extension(
107 &self,
108 extension_dir: &Path,
109 options: CompileExtensionOptions,
110 ) -> Result<(), anyhow::Error> {
111 self.install_rust_wasm_target_if_needed()?;
112 let adapter_bytes = self.install_wasi_preview1_adapter_if_needed().await?;
113
114 let cargo_toml_content = fs::read_to_string(&extension_dir.join("Cargo.toml"))?;
115 let cargo_toml: CargoToml = toml::from_str(&cargo_toml_content)?;
116
117 log::info!("compiling rust extension {}", extension_dir.display());
118 let output = Command::new("cargo")
119 .args(["build", "--target", RUST_TARGET])
120 .args(options.release.then_some("--release"))
121 .arg("--target-dir")
122 .arg(extension_dir.join("target"))
123 .current_dir(&extension_dir)
124 .output()
125 .context("failed to run `cargo`")?;
126 if !output.status.success() {
127 bail!(
128 "failed to build extension {}",
129 String::from_utf8_lossy(&output.stderr)
130 );
131 }
132
133 let mut wasm_path = PathBuf::from(extension_dir);
134 wasm_path.extend([
135 "target",
136 RUST_TARGET,
137 if options.release { "release" } else { "debug" },
138 &cargo_toml
139 .package
140 .name
141 // The wasm32-wasi target normalizes `-` in package names to `_` in the resulting `.wasm` file.
142 .replace('-', "_"),
143 ]);
144 wasm_path.set_extension("wasm");
145
146 let wasm_bytes = fs::read(&wasm_path)
147 .with_context(|| format!("failed to read output module `{}`", wasm_path.display()))?;
148
149 let encoder = ComponentEncoder::default()
150 .module(&wasm_bytes)?
151 .adapter("wasi_snapshot_preview1", &adapter_bytes)
152 .context("failed to load adapter module")?
153 .validate(true);
154
155 let component_bytes = encoder
156 .encode()
157 .context("failed to encode wasm component")?;
158
159 let component_bytes = self
160 .strip_custom_sections(&component_bytes)
161 .context("failed to strip debug sections from wasm component")?;
162
163 fs::write(extension_dir.join("extension.wasm"), &component_bytes)
164 .context("failed to write extension.wasm")?;
165
166 Ok(())
167 }
168
169 async fn compile_grammar(
170 &self,
171 extension_dir: &Path,
172 grammar_name: &str,
173 grammar_metadata: &GrammarManifestEntry,
174 ) -> Result<()> {
175 let clang_path = self.install_wasi_sdk_if_needed().await?;
176
177 let mut grammar_repo_dir = extension_dir.to_path_buf();
178 grammar_repo_dir.extend(["grammars", grammar_name]);
179
180 let mut grammar_wasm_path = grammar_repo_dir.clone();
181 grammar_wasm_path.set_extension("wasm");
182
183 log::info!("checking out {grammar_name} parser");
184 self.checkout_repo(
185 &grammar_repo_dir,
186 &grammar_metadata.repository,
187 &grammar_metadata.rev,
188 )?;
189
190 let src_path = grammar_repo_dir.join("src");
191 let parser_path = src_path.join("parser.c");
192 let scanner_path = src_path.join("scanner.c");
193
194 log::info!("compiling {grammar_name} parser");
195 let clang_output = Command::new(&clang_path)
196 .args(["-fPIC", "-shared", "-Os"])
197 .arg(format!("-Wl,--export=tree_sitter_{grammar_name}"))
198 .arg("-o")
199 .arg(&grammar_wasm_path)
200 .arg("-I")
201 .arg(&src_path)
202 .arg(&parser_path)
203 .args(scanner_path.exists().then_some(scanner_path))
204 .output()
205 .context("failed to run clang")?;
206 if !clang_output.status.success() {
207 bail!(
208 "failed to compile {} parser with clang: {}",
209 grammar_name,
210 String::from_utf8_lossy(&clang_output.stderr),
211 );
212 }
213
214 Ok(())
215 }
216
217 fn checkout_repo(&self, directory: &Path, url: &str, rev: &str) -> Result<()> {
218 let git_dir = directory.join(".git");
219
220 if directory.exists() {
221 let remotes_output = Command::new("git")
222 .arg("--git-dir")
223 .arg(&git_dir)
224 .args(["remote", "-v"])
225 .output()?;
226 let has_remote = remotes_output.status.success()
227 && String::from_utf8_lossy(&remotes_output.stdout)
228 .lines()
229 .any(|line| {
230 let mut parts = line.split(|c: char| c.is_whitespace());
231 parts.next() == Some("origin") && parts.any(|part| part == url)
232 });
233 if !has_remote {
234 bail!(
235 "grammar directory '{}' already exists, but is not a git clone of '{}'",
236 directory.display(),
237 url
238 );
239 }
240 } else {
241 fs::create_dir_all(&directory).with_context(|| {
242 format!("failed to create grammar directory {}", directory.display(),)
243 })?;
244 let init_output = Command::new("git")
245 .arg("init")
246 .current_dir(&directory)
247 .output()?;
248 if !init_output.status.success() {
249 bail!(
250 "failed to run `git init` in directory '{}'",
251 directory.display()
252 );
253 }
254
255 let remote_add_output = Command::new("git")
256 .arg("--git-dir")
257 .arg(&git_dir)
258 .args(["remote", "add", "origin", url])
259 .output()
260 .context("failed to execute `git remote add`")?;
261 if !remote_add_output.status.success() {
262 bail!(
263 "failed to add remote {url} for git repository {}",
264 git_dir.display()
265 );
266 }
267 }
268
269 let fetch_output = Command::new("git")
270 .arg("--git-dir")
271 .arg(&git_dir)
272 .args(["fetch", "--depth", "1", "origin", &rev])
273 .output()
274 .context("failed to execute `git fetch`")?;
275
276 let checkout_output = Command::new("git")
277 .arg("--git-dir")
278 .arg(&git_dir)
279 .args(["checkout", &rev])
280 .current_dir(&directory)
281 .output()
282 .context("failed to execute `git checkout`")?;
283 if !checkout_output.status.success() {
284 if !fetch_output.status.success() {
285 bail!(
286 "failed to fetch revision {} in directory '{}'",
287 rev,
288 directory.display()
289 );
290 }
291 bail!(
292 "failed to checkout revision {} in directory '{}': {}",
293 rev,
294 directory.display(),
295 String::from_utf8_lossy(&checkout_output.stderr)
296 );
297 }
298
299 Ok(())
300 }
301
302 fn install_rust_wasm_target_if_needed(&self) -> Result<()> {
303 let rustc_output = Command::new("rustc")
304 .arg("--print")
305 .arg("sysroot")
306 .output()
307 .context("failed to run rustc")?;
308 if !rustc_output.status.success() {
309 bail!(
310 "failed to retrieve rust sysroot: {}",
311 String::from_utf8_lossy(&rustc_output.stderr)
312 );
313 }
314
315 let sysroot = PathBuf::from(String::from_utf8(rustc_output.stdout)?.trim());
316 if sysroot.join("lib/rustlib").join(RUST_TARGET).exists() {
317 return Ok(());
318 }
319
320 let output = Command::new("rustup")
321 .args(["target", "add", RUST_TARGET])
322 .stderr(Stdio::inherit())
323 .stdout(Stdio::inherit())
324 .output()
325 .context("failed to run `rustup target add`")?;
326 if !output.status.success() {
327 bail!("failed to install the `{RUST_TARGET}` target");
328 }
329
330 Ok(())
331 }
332
333 async fn install_wasi_preview1_adapter_if_needed(&self) -> Result<Vec<u8>> {
334 let cache_path = self.cache_dir.join("wasi_snapshot_preview1.reactor.wasm");
335 if let Ok(content) = fs::read(&cache_path) {
336 if Parser::is_core_wasm(&content) {
337 return Ok(content);
338 }
339 }
340
341 fs::remove_file(&cache_path).ok();
342
343 log::info!(
344 "downloading wasi adapter module to {}",
345 cache_path.display()
346 );
347 let mut response = self
348 .http
349 .get(WASI_ADAPTER_URL, AsyncBody::default(), true)
350 .await?;
351
352 let mut content = Vec::new();
353 let mut body = BufReader::new(response.body_mut());
354 body.read_to_end(&mut content).await?;
355
356 fs::write(&cache_path, &content)
357 .with_context(|| format!("failed to save file {}", cache_path.display()))?;
358
359 if !Parser::is_core_wasm(&content) {
360 bail!("downloaded wasi adapter is invalid");
361 }
362 Ok(content)
363 }
364
365 async fn install_wasi_sdk_if_needed(&self) -> Result<PathBuf> {
366 let url = if let Some(asset_name) = WASI_SDK_ASSET_NAME {
367 format!("{WASI_SDK_URL}/{asset_name}")
368 } else {
369 bail!("wasi-sdk is not available for platform {}", env::consts::OS);
370 };
371
372 let wasi_sdk_dir = self.cache_dir.join("wasi-sdk");
373 let mut clang_path = wasi_sdk_dir.clone();
374 clang_path.extend(["bin", "clang-17"]);
375
376 if fs::metadata(&clang_path).map_or(false, |metadata| metadata.is_file()) {
377 return Ok(clang_path);
378 }
379
380 let mut tar_out_dir = wasi_sdk_dir.clone();
381 tar_out_dir.set_extension("archive");
382
383 fs::remove_dir_all(&wasi_sdk_dir).ok();
384 fs::remove_dir_all(&tar_out_dir).ok();
385
386 log::info!("downloading wasi-sdk to {}", wasi_sdk_dir.display());
387 let mut response = self.http.get(&url, AsyncBody::default(), true).await?;
388 let body = BufReader::new(response.body_mut());
389 let body = GzipDecoder::new(body);
390 let tar = Archive::new(body);
391 tar.unpack(&tar_out_dir)
392 .await
393 .context("failed to unpack wasi-sdk archive")?;
394
395 let inner_dir = fs::read_dir(&tar_out_dir)?
396 .next()
397 .ok_or_else(|| anyhow!("no content"))?
398 .context("failed to read contents of extracted wasi archive directory")?
399 .path();
400 fs::rename(&inner_dir, &wasi_sdk_dir).context("failed to move extracted wasi dir")?;
401 fs::remove_dir_all(&tar_out_dir).ok();
402
403 Ok(clang_path)
404 }
405
406 // This was adapted from:
407 // https://github.com/bytecodealliance/wasm-tools/1791a8f139722e9f8679a2bd3d8e423e55132b22/src/bin/wasm-tools/strip.rs
408 fn strip_custom_sections(&self, input: &Vec<u8>) -> Result<Vec<u8>> {
409 use wasmparser::Payload::*;
410
411 let strip_custom_section = |name: &str| name.starts_with(".debug");
412
413 let mut output = Vec::new();
414 let mut stack = Vec::new();
415
416 for payload in Parser::new(0).parse_all(input) {
417 let payload = payload?;
418
419 // Track nesting depth, so that we don't mess with inner producer sections:
420 match payload {
421 Version { encoding, .. } => {
422 output.extend_from_slice(match encoding {
423 wasmparser::Encoding::Component => &wasm_encoder::Component::HEADER,
424 wasmparser::Encoding::Module => &wasm_encoder::Module::HEADER,
425 });
426 }
427 ModuleSection { .. } | ComponentSection { .. } => {
428 stack.push(mem::take(&mut output));
429 continue;
430 }
431 End { .. } => {
432 let mut parent = match stack.pop() {
433 Some(c) => c,
434 None => break,
435 };
436 if output.starts_with(&wasm_encoder::Component::HEADER) {
437 parent.push(ComponentSectionId::Component as u8);
438 output.encode(&mut parent);
439 } else {
440 parent.push(ComponentSectionId::CoreModule as u8);
441 output.encode(&mut parent);
442 }
443 output = parent;
444 }
445 _ => {}
446 }
447
448 match &payload {
449 CustomSection(c) => {
450 if strip_custom_section(c.name()) {
451 continue;
452 }
453 }
454
455 _ => {}
456 }
457
458 if let Some((id, range)) = payload.as_section() {
459 RawSection {
460 id,
461 data: &input[range],
462 }
463 .append_to(&mut output);
464 }
465 }
466
467 Ok(output)
468 }
469}