1use crate::{
2 ExtensionLibraryKind, ExtensionManifest, GrammarManifestEntry, parse_wasm_extension_version,
3};
4use anyhow::{Context as _, Result, anyhow, bail};
5use async_compression::futures::bufread::GzipDecoder;
6use async_tar::Archive;
7use convert_case::{Case, Casing as _};
8use futures::AsyncReadExt;
9use futures::io::BufReader;
10use http_client::{self, AsyncBody, HttpClient};
11use serde::Deserialize;
12use std::{
13 env, fs, mem,
14 path::{Path, PathBuf},
15 process::Stdio,
16 sync::Arc,
17};
18use wasm_encoder::{ComponentSectionId, Encode as _, RawSection, Section as _};
19use wasmparser::Parser;
20use wit_component::ComponentEncoder;
21
22/// Currently, we compile with Rust's `wasm32-wasip1` target, which works with WASI `preview1`.
23/// But the WASM component model is based on WASI `preview2`. So we need an 'adapter' WASM
24/// module, which implements the `preview1` interface in terms of `preview2`.
25///
26/// Once Rust 1.78 is released, there will be a `wasm32-wasip2` target available, so we will
27/// not need the adapter anymore.
28const RUST_TARGET: &str = "wasm32-wasip1";
29const WASI_ADAPTER_URL: &str = "https://github.com/bytecodealliance/wasmtime/releases/download/v18.0.2/wasi_snapshot_preview1.reactor.wasm";
30
31/// Compiling Tree-sitter parsers from C to WASM requires Clang 17, and a WASM build of libc
32/// and clang's runtime library. The `wasi-sdk` provides these binaries.
33///
34/// Once Clang 17 and its wasm target are available via system package managers, we won't need
35/// to download this.
36const WASI_SDK_URL: &str = "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-21/";
37const WASI_SDK_ASSET_NAME: Option<&str> = if cfg!(target_os = "macos") {
38 Some("wasi-sdk-21.0-macos.tar.gz")
39} else if cfg!(any(target_os = "linux", target_os = "freebsd")) {
40 Some("wasi-sdk-21.0-linux.tar.gz")
41} else if cfg!(target_os = "windows") {
42 Some("wasi-sdk-21.0.m-mingw.tar.gz")
43} else {
44 None
45};
46
47pub struct ExtensionBuilder {
48 cache_dir: PathBuf,
49 pub http: Arc<dyn HttpClient>,
50}
51
52pub struct CompileExtensionOptions {
53 pub release: bool,
54}
55
56#[derive(Deserialize)]
57struct CargoToml {
58 package: CargoTomlPackage,
59}
60
61#[derive(Deserialize)]
62struct CargoTomlPackage {
63 name: String,
64}
65
66impl ExtensionBuilder {
67 pub fn new(http_client: Arc<dyn HttpClient>, cache_dir: PathBuf) -> Self {
68 Self {
69 cache_dir,
70 http: http_client,
71 }
72 }
73
74 pub async fn compile_extension(
75 &self,
76 extension_dir: &Path,
77 extension_manifest: &mut ExtensionManifest,
78 options: CompileExtensionOptions,
79 ) -> Result<()> {
80 populate_defaults(extension_manifest, extension_dir)?;
81
82 if extension_dir.is_relative() {
83 bail!(
84 "extension dir {} is not an absolute path",
85 extension_dir.display()
86 );
87 }
88
89 fs::create_dir_all(&self.cache_dir).context("failed to create cache dir")?;
90
91 if extension_manifest.lib.kind == Some(ExtensionLibraryKind::Rust) {
92 log::info!("compiling Rust extension {}", extension_dir.display());
93 self.compile_rust_extension(extension_dir, extension_manifest, options)
94 .await
95 .context("failed to compile Rust extension")?;
96 log::info!("compiled Rust extension {}", extension_dir.display());
97 }
98
99 for (grammar_name, grammar_metadata) in &extension_manifest.grammars {
100 let snake_cased_grammar_name = grammar_name.to_case(Case::Snake);
101 if grammar_name.as_ref() != snake_cased_grammar_name.as_str() {
102 bail!(
103 "grammar name '{grammar_name}' must be written in snake_case: {snake_cased_grammar_name}"
104 );
105 }
106
107 log::info!(
108 "compiling grammar {grammar_name} for extension {}",
109 extension_dir.display()
110 );
111 self.compile_grammar(extension_dir, grammar_name.as_ref(), grammar_metadata)
112 .await
113 .with_context(|| format!("failed to compile grammar '{grammar_name}'"))?;
114 log::info!(
115 "compiled grammar {grammar_name} for extension {}",
116 extension_dir.display()
117 );
118 }
119
120 log::info!("finished compiling extension {}", extension_dir.display());
121 Ok(())
122 }
123
124 async fn compile_rust_extension(
125 &self,
126 extension_dir: &Path,
127 manifest: &mut ExtensionManifest,
128 options: CompileExtensionOptions,
129 ) -> Result<(), anyhow::Error> {
130 self.install_rust_wasm_target_if_needed()?;
131 let adapter_bytes = self.install_wasi_preview1_adapter_if_needed().await?;
132
133 let cargo_toml_content = fs::read_to_string(extension_dir.join("Cargo.toml"))?;
134 let cargo_toml: CargoToml = toml::from_str(&cargo_toml_content)?;
135
136 log::info!(
137 "compiling Rust crate for extension {}",
138 extension_dir.display()
139 );
140 let output = util::command::new_std_command("cargo")
141 .args(["build", "--target", RUST_TARGET])
142 .args(options.release.then_some("--release"))
143 .arg("--target-dir")
144 .arg(extension_dir.join("target"))
145 // WASI builds do not work with sccache and just stuck, so disable it.
146 .env("RUSTC_WRAPPER", "")
147 .current_dir(extension_dir)
148 .output()
149 .context("failed to run `cargo`")?;
150 if !output.status.success() {
151 bail!(
152 "failed to build extension {}",
153 String::from_utf8_lossy(&output.stderr)
154 );
155 }
156
157 log::info!(
158 "compiled Rust crate for extension {}",
159 extension_dir.display()
160 );
161
162 let mut wasm_path = PathBuf::from(extension_dir);
163 wasm_path.extend([
164 "target",
165 RUST_TARGET,
166 if options.release { "release" } else { "debug" },
167 &cargo_toml
168 .package
169 .name
170 // The wasm32-wasip1 target normalizes `-` in package names to `_` in the resulting `.wasm` file.
171 .replace('-', "_"),
172 ]);
173 wasm_path.set_extension("wasm");
174
175 let wasm_bytes = fs::read(&wasm_path)
176 .with_context(|| format!("failed to read output module `{}`", wasm_path.display()))?;
177
178 let mut encoder = ComponentEncoder::default()
179 .module(&wasm_bytes)?
180 .adapter("wasi_snapshot_preview1", &adapter_bytes)
181 .context("failed to load adapter module")?
182 .validate(true);
183
184 log::info!(
185 "encoding wasm component for extension {}",
186 extension_dir.display()
187 );
188
189 let component_bytes = encoder
190 .encode()
191 .context("failed to encode wasm component")?;
192
193 let component_bytes = self
194 .strip_custom_sections(&component_bytes)
195 .context("failed to strip debug sections from wasm component")?;
196
197 let wasm_extension_api_version =
198 parse_wasm_extension_version(&manifest.id, &component_bytes)
199 .context("compiled wasm did not contain a valid zed extension api version")?;
200 manifest.lib.version = Some(wasm_extension_api_version);
201
202 let extension_file = extension_dir.join("extension.wasm");
203 fs::write(extension_file.clone(), &component_bytes)
204 .context("failed to write extension.wasm")?;
205
206 log::info!(
207 "extension {} written to {}",
208 extension_dir.display(),
209 extension_file.display()
210 );
211
212 Ok(())
213 }
214
215 async fn compile_grammar(
216 &self,
217 extension_dir: &Path,
218 grammar_name: &str,
219 grammar_metadata: &GrammarManifestEntry,
220 ) -> Result<()> {
221 let clang_path = self.install_wasi_sdk_if_needed().await?;
222
223 let mut grammar_repo_dir = extension_dir.to_path_buf();
224 grammar_repo_dir.extend(["grammars", grammar_name]);
225
226 let mut grammar_wasm_path = grammar_repo_dir.clone();
227 grammar_wasm_path.set_extension("wasm");
228
229 log::info!("checking out {grammar_name} parser");
230 self.checkout_repo(
231 &grammar_repo_dir,
232 &grammar_metadata.repository,
233 &grammar_metadata.rev,
234 )?;
235
236 let base_grammar_path = grammar_metadata
237 .path
238 .as_ref()
239 .map(|path| grammar_repo_dir.join(path))
240 .unwrap_or(grammar_repo_dir);
241
242 let src_path = base_grammar_path.join("src");
243 let parser_path = src_path.join("parser.c");
244 let scanner_path = src_path.join("scanner.c");
245
246 log::info!("compiling {grammar_name} parser");
247 let clang_output = util::command::new_std_command(&clang_path)
248 .args(["-fPIC", "-shared", "-Os"])
249 .arg(format!("-Wl,--export=tree_sitter_{grammar_name}"))
250 .arg("-o")
251 .arg(&grammar_wasm_path)
252 .arg("-I")
253 .arg(&src_path)
254 .arg(&parser_path)
255 .args(scanner_path.exists().then_some(scanner_path))
256 .output()
257 .context("failed to run clang")?;
258
259 if !clang_output.status.success() {
260 bail!(
261 "failed to compile {} parser with clang: {}",
262 grammar_name,
263 String::from_utf8_lossy(&clang_output.stderr),
264 );
265 }
266
267 Ok(())
268 }
269
270 fn checkout_repo(&self, directory: &Path, url: &str, rev: &str) -> Result<()> {
271 let git_dir = directory.join(".git");
272
273 if directory.exists() {
274 let remotes_output = util::command::new_std_command("git")
275 .arg("--git-dir")
276 .arg(&git_dir)
277 .args(["remote", "-v"])
278 .output()?;
279 let has_remote = remotes_output.status.success()
280 && String::from_utf8_lossy(&remotes_output.stdout)
281 .lines()
282 .any(|line| {
283 let mut parts = line.split(|c: char| c.is_whitespace());
284 parts.next() == Some("origin") && parts.any(|part| part == url)
285 });
286 if !has_remote {
287 bail!(
288 "grammar directory '{}' already exists, but is not a git clone of '{}'",
289 directory.display(),
290 url
291 );
292 }
293 } else {
294 fs::create_dir_all(directory).with_context(|| {
295 format!("failed to create grammar directory {}", directory.display(),)
296 })?;
297 let init_output = util::command::new_std_command("git")
298 .arg("init")
299 .current_dir(directory)
300 .output()?;
301 if !init_output.status.success() {
302 bail!(
303 "failed to run `git init` in directory '{}'",
304 directory.display()
305 );
306 }
307
308 let remote_add_output = util::command::new_std_command("git")
309 .arg("--git-dir")
310 .arg(&git_dir)
311 .args(["remote", "add", "origin", url])
312 .output()
313 .context("failed to execute `git remote add`")?;
314 if !remote_add_output.status.success() {
315 bail!(
316 "failed to add remote {url} for git repository {}",
317 git_dir.display()
318 );
319 }
320 }
321
322 let fetch_output = util::command::new_std_command("git")
323 .arg("--git-dir")
324 .arg(&git_dir)
325 .args(["fetch", "--depth", "1", "origin", rev])
326 .output()
327 .context("failed to execute `git fetch`")?;
328
329 let checkout_output = util::command::new_std_command("git")
330 .arg("--git-dir")
331 .arg(&git_dir)
332 .args(["checkout", rev])
333 .current_dir(directory)
334 .output()
335 .context("failed to execute `git checkout`")?;
336 if !checkout_output.status.success() {
337 if !fetch_output.status.success() {
338 bail!(
339 "failed to fetch revision {} in directory '{}'",
340 rev,
341 directory.display()
342 );
343 }
344 bail!(
345 "failed to checkout revision {} in directory '{}': {}",
346 rev,
347 directory.display(),
348 String::from_utf8_lossy(&checkout_output.stderr)
349 );
350 }
351
352 Ok(())
353 }
354
355 fn install_rust_wasm_target_if_needed(&self) -> Result<()> {
356 let rustc_output = util::command::new_std_command("rustc")
357 .arg("--print")
358 .arg("sysroot")
359 .output()
360 .context("failed to run rustc")?;
361 if !rustc_output.status.success() {
362 bail!(
363 "failed to retrieve rust sysroot: {}",
364 String::from_utf8_lossy(&rustc_output.stderr)
365 );
366 }
367
368 let sysroot = PathBuf::from(String::from_utf8(rustc_output.stdout)?.trim());
369 if sysroot.join("lib/rustlib").join(RUST_TARGET).exists() {
370 return Ok(());
371 }
372
373 let output = util::command::new_std_command("rustup")
374 .args(["target", "add", RUST_TARGET])
375 .stderr(Stdio::piped())
376 .stdout(Stdio::inherit())
377 .output()
378 .context("failed to run `rustup target add`")?;
379 if !output.status.success() {
380 bail!(
381 "failed to install the `{RUST_TARGET}` target: {}",
382 String::from_utf8_lossy(&rustc_output.stderr)
383 );
384 }
385
386 Ok(())
387 }
388
389 async fn install_wasi_preview1_adapter_if_needed(&self) -> Result<Vec<u8>> {
390 let cache_path = self.cache_dir.join("wasi_snapshot_preview1.reactor.wasm");
391 if let Ok(content) = fs::read(&cache_path) {
392 if Parser::is_core_wasm(&content) {
393 return Ok(content);
394 }
395 }
396
397 fs::remove_file(&cache_path).ok();
398
399 log::info!(
400 "downloading wasi adapter module to {}",
401 cache_path.display()
402 );
403 let mut response = self
404 .http
405 .get(WASI_ADAPTER_URL, AsyncBody::default(), true)
406 .await?;
407
408 let mut content = Vec::new();
409 let mut body = BufReader::new(response.body_mut());
410 body.read_to_end(&mut content).await?;
411
412 fs::write(&cache_path, &content)
413 .with_context(|| format!("failed to save file {}", cache_path.display()))?;
414
415 if !Parser::is_core_wasm(&content) {
416 bail!("downloaded wasi adapter is invalid");
417 }
418 Ok(content)
419 }
420
421 async fn install_wasi_sdk_if_needed(&self) -> Result<PathBuf> {
422 let url = if let Some(asset_name) = WASI_SDK_ASSET_NAME {
423 format!("{WASI_SDK_URL}/{asset_name}")
424 } else {
425 bail!("wasi-sdk is not available for platform {}", env::consts::OS);
426 };
427
428 let wasi_sdk_dir = self.cache_dir.join("wasi-sdk");
429 let mut clang_path = wasi_sdk_dir.clone();
430 clang_path.extend(["bin", &format!("clang{}", env::consts::EXE_SUFFIX)]);
431
432 if fs::metadata(&clang_path).map_or(false, |metadata| metadata.is_file()) {
433 return Ok(clang_path);
434 }
435
436 let mut tar_out_dir = wasi_sdk_dir.clone();
437 tar_out_dir.set_extension("archive");
438
439 fs::remove_dir_all(&wasi_sdk_dir).ok();
440 fs::remove_dir_all(&tar_out_dir).ok();
441
442 log::info!("downloading wasi-sdk to {}", wasi_sdk_dir.display());
443 let mut response = self.http.get(&url, AsyncBody::default(), true).await?;
444 let body = BufReader::new(response.body_mut());
445 let body = GzipDecoder::new(body);
446 let tar = Archive::new(body);
447
448 tar.unpack(&tar_out_dir)
449 .await
450 .context("failed to unpack wasi-sdk archive")?;
451
452 let inner_dir = fs::read_dir(&tar_out_dir)?
453 .next()
454 .ok_or_else(|| anyhow!("no content"))?
455 .context("failed to read contents of extracted wasi archive directory")?
456 .path();
457 fs::rename(&inner_dir, &wasi_sdk_dir).context("failed to move extracted wasi dir")?;
458 fs::remove_dir_all(&tar_out_dir).ok();
459
460 Ok(clang_path)
461 }
462
463 // This was adapted from:
464 // https://github.com/bytecodealliance/wasm-tools/blob/1791a8f139722e9f8679a2bd3d8e423e55132b22/src/bin/wasm-tools/strip.rs
465 fn strip_custom_sections(&self, input: &Vec<u8>) -> Result<Vec<u8>> {
466 use wasmparser::Payload::*;
467
468 let strip_custom_section = |name: &str| name.starts_with(".debug");
469
470 let mut output = Vec::new();
471 let mut stack = Vec::new();
472
473 for payload in Parser::new(0).parse_all(input) {
474 let payload = payload?;
475 let component_header = wasm_encoder::Component::HEADER;
476 let module_header = wasm_encoder::Module::HEADER;
477
478 // Track nesting depth, so that we don't mess with inner producer sections:
479 match payload {
480 Version { encoding, .. } => {
481 output.extend_from_slice(match encoding {
482 wasmparser::Encoding::Component => &component_header,
483 wasmparser::Encoding::Module => &module_header,
484 });
485 }
486 ModuleSection { .. } | ComponentSection { .. } => {
487 stack.push(mem::take(&mut output));
488 continue;
489 }
490 End { .. } => {
491 let mut parent = match stack.pop() {
492 Some(c) => c,
493 None => break,
494 };
495 if output.starts_with(&component_header) {
496 parent.push(ComponentSectionId::Component as u8);
497 output.encode(&mut parent);
498 } else {
499 parent.push(ComponentSectionId::CoreModule as u8);
500 output.encode(&mut parent);
501 }
502 output = parent;
503 }
504 _ => {}
505 }
506
507 if let CustomSection(c) = &payload {
508 if strip_custom_section(c.name()) {
509 continue;
510 }
511 }
512
513 if let Some((id, range)) = payload.as_section() {
514 RawSection {
515 id,
516 data: &input[range],
517 }
518 .append_to(&mut output);
519 }
520 }
521
522 Ok(output)
523 }
524}
525
526fn populate_defaults(manifest: &mut ExtensionManifest, extension_path: &Path) -> Result<()> {
527 // For legacy extensions on the v0 schema (aka, using `extension.json`), clear out any existing
528 // contents of the computed fields, since we don't care what the existing values are.
529 if manifest.schema_version.is_v0() {
530 manifest.languages.clear();
531 manifest.grammars.clear();
532 manifest.themes.clear();
533 }
534
535 let cargo_toml_path = extension_path.join("Cargo.toml");
536 if cargo_toml_path.exists() {
537 manifest.lib.kind = Some(ExtensionLibraryKind::Rust);
538 }
539
540 let languages_dir = extension_path.join("languages");
541 if languages_dir.exists() {
542 for entry in fs::read_dir(&languages_dir).context("failed to list languages dir")? {
543 let entry = entry?;
544 let language_dir = entry.path();
545 let config_path = language_dir.join("config.toml");
546 if config_path.exists() {
547 let relative_language_dir =
548 language_dir.strip_prefix(extension_path)?.to_path_buf();
549 if !manifest.languages.contains(&relative_language_dir) {
550 manifest.languages.push(relative_language_dir);
551 }
552 }
553 }
554 }
555
556 let themes_dir = extension_path.join("themes");
557 if themes_dir.exists() {
558 for entry in fs::read_dir(&themes_dir).context("failed to list themes dir")? {
559 let entry = entry?;
560 let theme_path = entry.path();
561 if theme_path.extension() == Some("json".as_ref()) {
562 let relative_theme_path = theme_path.strip_prefix(extension_path)?.to_path_buf();
563 if !manifest.themes.contains(&relative_theme_path) {
564 manifest.themes.push(relative_theme_path);
565 }
566 }
567 }
568 }
569
570 let icon_themes_dir = extension_path.join("icon_themes");
571 if icon_themes_dir.exists() {
572 for entry in fs::read_dir(&icon_themes_dir).context("failed to list icon themes dir")? {
573 let entry = entry?;
574 let icon_theme_path = entry.path();
575 if icon_theme_path.extension() == Some("json".as_ref()) {
576 let relative_icon_theme_path =
577 icon_theme_path.strip_prefix(extension_path)?.to_path_buf();
578 if !manifest.icon_themes.contains(&relative_icon_theme_path) {
579 manifest.icon_themes.push(relative_icon_theme_path);
580 }
581 }
582 }
583 }
584
585 let snippets_json_path = extension_path.join("snippets.json");
586 if snippets_json_path.exists() {
587 manifest.snippets = Some(snippets_json_path);
588 }
589
590 // For legacy extensions on the v0 schema (aka, using `extension.json`), we want to populate the grammars in
591 // the manifest using the contents of the `grammars` directory.
592 if manifest.schema_version.is_v0() {
593 let grammars_dir = extension_path.join("grammars");
594 if grammars_dir.exists() {
595 for entry in fs::read_dir(&grammars_dir).context("failed to list grammars dir")? {
596 let entry = entry?;
597 let grammar_path = entry.path();
598 if grammar_path.extension() == Some("toml".as_ref()) {
599 #[derive(Deserialize)]
600 struct GrammarConfigToml {
601 pub repository: String,
602 pub commit: String,
603 #[serde(default)]
604 pub path: Option<String>,
605 }
606
607 let grammar_config = fs::read_to_string(&grammar_path)?;
608 let grammar_config: GrammarConfigToml = toml::from_str(&grammar_config)?;
609
610 let grammar_name = grammar_path
611 .file_stem()
612 .and_then(|stem| stem.to_str())
613 .ok_or_else(|| anyhow!("no grammar name"))?;
614 if !manifest.grammars.contains_key(grammar_name) {
615 manifest.grammars.insert(
616 grammar_name.into(),
617 GrammarManifestEntry {
618 repository: grammar_config.repository,
619 rev: grammar_config.commit,
620 path: grammar_config.path,
621 },
622 );
623 }
624 }
625 }
626 }
627 }
628
629 Ok(())
630}