1use crate::{
2 ExtensionLibraryKind, ExtensionManifest, GrammarManifestEntry, build_debug_adapter_schema_path,
3 parse_wasm_extension_version,
4};
5use anyhow::{Context as _, Result, bail};
6use async_compression::futures::bufread::GzipDecoder;
7use async_tar::Archive;
8use futures::io::BufReader;
9use heck::ToSnakeCase;
10use http_client::{self, AsyncBody, HttpClient};
11use serde::Deserialize;
12use std::{
13 env, fs, mem,
14 path::{Path, PathBuf},
15 process::Stdio,
16 str::FromStr,
17 sync::Arc,
18};
19use wasm_encoder::{ComponentSectionId, Encode as _, RawSection, Section as _};
20use wasmparser::Parser;
21
22/// Currently, we compile with Rust's `wasm32-wasip2` target, which works with WASI `preview2` and the component model.
23const RUST_TARGET: &str = "wasm32-wasip2";
24
25/// Compiling Tree-sitter parsers from C to WASM requires Clang 17, and a WASM build of libc
26/// and clang's runtime library. The `wasi-sdk` provides these binaries.
27///
28/// Once Clang 17 and its wasm target are available via system package managers, we won't need
29/// to download this.
30const WASI_SDK_URL: &str = "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-25/";
31const WASI_SDK_ASSET_NAME: Option<&str> = if cfg!(all(target_os = "macos", target_arch = "x86_64"))
32{
33 Some("wasi-sdk-25.0-x86_64-macos.tar.gz")
34} else if cfg!(all(target_os = "macos", target_arch = "aarch64")) {
35 Some("wasi-sdk-25.0-arm64-macos.tar.gz")
36} else if cfg!(all(target_os = "linux", target_arch = "x86_64")) {
37 Some("wasi-sdk-25.0-x86_64-linux.tar.gz")
38} else if cfg!(all(target_os = "linux", target_arch = "aarch64")) {
39 Some("wasi-sdk-25.0-arm64-linux.tar.gz")
40} else if cfg!(all(target_os = "freebsd", target_arch = "x86_64")) {
41 Some("wasi-sdk-25.0-x86_64-linux.tar.gz")
42} else if cfg!(all(target_os = "freebsd", target_arch = "aarch64")) {
43 Some("wasi-sdk-25.0-arm64-linux.tar.gz")
44} else if cfg!(all(target_os = "windows", target_arch = "x86_64")) {
45 Some("wasi-sdk-25.0-x86_64-windows.tar.gz")
46} else {
47 None
48};
49
50pub struct ExtensionBuilder {
51 cache_dir: PathBuf,
52 pub http: Arc<dyn HttpClient>,
53}
54
55pub struct CompileExtensionOptions {
56 pub release: bool,
57}
58
59#[derive(Deserialize)]
60struct CargoToml {
61 package: CargoTomlPackage,
62}
63
64#[derive(Deserialize)]
65struct CargoTomlPackage {
66 name: String,
67}
68
69impl ExtensionBuilder {
70 pub fn new(http_client: Arc<dyn HttpClient>, cache_dir: PathBuf) -> Self {
71 Self {
72 cache_dir,
73 http: http_client,
74 }
75 }
76
77 pub async fn compile_extension(
78 &self,
79 extension_dir: &Path,
80 extension_manifest: &mut ExtensionManifest,
81 options: CompileExtensionOptions,
82 ) -> Result<()> {
83 populate_defaults(extension_manifest, extension_dir)?;
84
85 if extension_dir.is_relative() {
86 bail!(
87 "extension dir {} is not an absolute path",
88 extension_dir.display()
89 );
90 }
91
92 fs::create_dir_all(&self.cache_dir).context("failed to create cache dir")?;
93
94 if extension_manifest.lib.kind == Some(ExtensionLibraryKind::Rust) {
95 log::info!("compiling Rust extension {}", extension_dir.display());
96 self.compile_rust_extension(extension_dir, extension_manifest, options)
97 .await
98 .context("failed to compile Rust extension")?;
99 log::info!("compiled Rust extension {}", extension_dir.display());
100 }
101
102 for (debug_adapter_name, meta) in &mut extension_manifest.debug_adapters {
103 let debug_adapter_schema_path =
104 extension_dir.join(build_debug_adapter_schema_path(debug_adapter_name, meta));
105
106 let debug_adapter_schema = fs::read_to_string(&debug_adapter_schema_path)
107 .with_context(|| {
108 format!("failed to read debug adapter schema for `{debug_adapter_name}` from `{debug_adapter_schema_path:?}`")
109 })?;
110 _ = serde_json::Value::from_str(&debug_adapter_schema).with_context(|| {
111 format!("Debug adapter schema for `{debug_adapter_name}` (path: `{debug_adapter_schema_path:?}`) is not a valid JSON")
112 })?;
113 }
114 for (grammar_name, grammar_metadata) in &extension_manifest.grammars {
115 let snake_cased_grammar_name = grammar_name.to_snake_case();
116 if grammar_name.as_ref() != snake_cased_grammar_name.as_str() {
117 bail!(
118 "grammar name '{grammar_name}' must be written in snake_case: {snake_cased_grammar_name}"
119 );
120 }
121
122 log::info!(
123 "compiling grammar {grammar_name} for extension {}",
124 extension_dir.display()
125 );
126 self.compile_grammar(extension_dir, grammar_name.as_ref(), grammar_metadata)
127 .await
128 .with_context(|| format!("failed to compile grammar '{grammar_name}'"))?;
129 log::info!(
130 "compiled grammar {grammar_name} for extension {}",
131 extension_dir.display()
132 );
133 }
134
135 log::info!("finished compiling extension {}", extension_dir.display());
136 Ok(())
137 }
138
139 async fn compile_rust_extension(
140 &self,
141 extension_dir: &Path,
142 manifest: &mut ExtensionManifest,
143 options: CompileExtensionOptions,
144 ) -> anyhow::Result<()> {
145 self.install_rust_wasm_target_if_needed()?;
146
147 let cargo_toml_content = fs::read_to_string(extension_dir.join("Cargo.toml"))?;
148 let cargo_toml: CargoToml = toml::from_str(&cargo_toml_content)?;
149
150 log::info!(
151 "compiling Rust crate for extension {}",
152 extension_dir.display()
153 );
154 let status = util::command::new_std_command("cargo")
155 .args(["build", "--target", RUST_TARGET])
156 .args(options.release.then_some("--release"))
157 .arg("--target-dir")
158 .arg(extension_dir.join("target"))
159 // WASI builds do not work with sccache and just stuck, so disable it.
160 .env("RUSTC_WRAPPER", "")
161 .current_dir(extension_dir)
162 .status()
163 .context("failed to run `cargo`")?;
164 if !status.success() {
165 bail!("failed to build extension",);
166 }
167
168 log::info!(
169 "compiled Rust crate for extension {}",
170 extension_dir.display()
171 );
172
173 let mut wasm_path = PathBuf::from(extension_dir);
174 wasm_path.extend([
175 "target",
176 RUST_TARGET,
177 if options.release { "release" } else { "debug" },
178 &cargo_toml
179 .package
180 .name
181 // The wasm32-wasip2 target normalizes `-` in package names to `_` in the resulting `.wasm` file.
182 .replace('-', "_"),
183 ]);
184 wasm_path.set_extension("wasm");
185
186 log::info!(
187 "encoding wasm component for extension {}",
188 extension_dir.display()
189 );
190
191 let component_bytes = fs::read(&wasm_path)
192 .with_context(|| format!("failed to read output module `{}`", wasm_path.display()))?;
193
194 let component_bytes = self
195 .strip_custom_sections(&component_bytes)
196 .context("failed to strip debug sections from wasm component")?;
197
198 let wasm_extension_api_version =
199 parse_wasm_extension_version(&manifest.id, &component_bytes)
200 .context("compiled wasm did not contain a valid zed extension api version")?;
201 manifest.lib.version = Some(wasm_extension_api_version);
202
203 let extension_file = extension_dir.join("extension.wasm");
204 fs::write(extension_file.clone(), &component_bytes)
205 .context("failed to write extension.wasm")?;
206
207 log::info!(
208 "extension {} written to {}",
209 extension_dir.display(),
210 extension_file.display()
211 );
212
213 Ok(())
214 }
215
216 async fn compile_grammar(
217 &self,
218 extension_dir: &Path,
219 grammar_name: &str,
220 grammar_metadata: &GrammarManifestEntry,
221 ) -> Result<()> {
222 let clang_path = self.install_wasi_sdk_if_needed().await?;
223
224 let mut grammar_repo_dir = extension_dir.to_path_buf();
225 grammar_repo_dir.extend(["grammars", grammar_name]);
226
227 let mut grammar_wasm_path = grammar_repo_dir.clone();
228 grammar_wasm_path.set_extension("wasm");
229
230 log::info!("checking out {grammar_name} parser");
231 self.checkout_repo(
232 &grammar_repo_dir,
233 &grammar_metadata.repository,
234 &grammar_metadata.rev,
235 )?;
236
237 let base_grammar_path = grammar_metadata
238 .path
239 .as_ref()
240 .map(|path| grammar_repo_dir.join(path))
241 .unwrap_or(grammar_repo_dir);
242
243 let src_path = base_grammar_path.join("src");
244 let parser_path = src_path.join("parser.c");
245 let scanner_path = src_path.join("scanner.c");
246
247 log::info!("compiling {grammar_name} parser");
248 let clang_status = util::command::new_std_command(&clang_path)
249 .args(["-fPIC", "-shared", "-Os"])
250 .arg(format!("-Wl,--export=tree_sitter_{grammar_name}"))
251 .arg("-o")
252 .arg(&grammar_wasm_path)
253 .arg("-I")
254 .arg(&src_path)
255 .arg(&parser_path)
256 .args(scanner_path.exists().then_some(scanner_path))
257 .status()
258 .context("failed to run clang")?;
259
260 if !clang_status.success() {
261 bail!("failed to compile {} parser with clang", grammar_name,);
262 }
263
264 Ok(())
265 }
266
267 fn checkout_repo(&self, directory: &Path, url: &str, rev: &str) -> Result<()> {
268 let git_dir = directory.join(".git");
269
270 if directory.exists() {
271 let remotes_output = util::command::new_std_command("git")
272 .arg("--git-dir")
273 .arg(&git_dir)
274 .args(["remote", "-v"])
275 .output()?;
276 let has_remote = remotes_output.status.success()
277 && String::from_utf8_lossy(&remotes_output.stdout)
278 .lines()
279 .any(|line| {
280 let mut parts = line.split(|c: char| c.is_whitespace());
281 parts.next() == Some("origin") && parts.any(|part| part == url)
282 });
283 if !has_remote {
284 bail!(
285 "grammar directory '{}' already exists, but is not a git clone of '{}'",
286 directory.display(),
287 url
288 );
289 }
290 } else {
291 fs::create_dir_all(directory).with_context(|| {
292 format!("failed to create grammar directory {}", directory.display(),)
293 })?;
294 let init_output = util::command::new_std_command("git")
295 .arg("init")
296 .current_dir(directory)
297 .output()?;
298 if !init_output.status.success() {
299 bail!(
300 "failed to run `git init` in directory '{}'",
301 directory.display()
302 );
303 }
304
305 let remote_add_output = util::command::new_std_command("git")
306 .arg("--git-dir")
307 .arg(&git_dir)
308 .args(["remote", "add", "origin", url])
309 .output()
310 .context("failed to execute `git remote add`")?;
311 if !remote_add_output.status.success() {
312 bail!(
313 "failed to add remote {url} for git repository {}",
314 git_dir.display()
315 );
316 }
317 }
318
319 let fetch_output = util::command::new_std_command("git")
320 .arg("--git-dir")
321 .arg(&git_dir)
322 .args(["fetch", "--depth", "1", "origin", rev])
323 .output()
324 .context("failed to execute `git fetch`")?;
325
326 let checkout_output = util::command::new_std_command("git")
327 .arg("--git-dir")
328 .arg(&git_dir)
329 .args(["checkout", rev])
330 .current_dir(directory)
331 .output()
332 .context("failed to execute `git checkout`")?;
333 if !checkout_output.status.success() {
334 if !fetch_output.status.success() {
335 bail!(
336 "failed to fetch revision {} in directory '{}'",
337 rev,
338 directory.display()
339 );
340 }
341 bail!(
342 "failed to checkout revision {} in directory '{}': {}",
343 rev,
344 directory.display(),
345 String::from_utf8_lossy(&checkout_output.stderr)
346 );
347 }
348
349 Ok(())
350 }
351
352 fn install_rust_wasm_target_if_needed(&self) -> Result<()> {
353 let rustc_output = util::command::new_std_command("rustc")
354 .arg("--print")
355 .arg("sysroot")
356 .output()
357 .context("failed to run rustc")?;
358 if !rustc_output.status.success() {
359 bail!(
360 "failed to retrieve rust sysroot: {}",
361 String::from_utf8_lossy(&rustc_output.stderr)
362 );
363 }
364
365 let sysroot = PathBuf::from(String::from_utf8(rustc_output.stdout)?.trim());
366 if sysroot.join("lib/rustlib").join(RUST_TARGET).exists() {
367 return Ok(());
368 }
369
370 let output = util::command::new_std_command("rustup")
371 .args(["target", "add", RUST_TARGET])
372 .stderr(Stdio::piped())
373 .stdout(Stdio::inherit())
374 .output()
375 .context("failed to run `rustup target add`")?;
376 if !output.status.success() {
377 bail!(
378 "failed to install the `{RUST_TARGET}` target: {}",
379 String::from_utf8_lossy(&rustc_output.stderr)
380 );
381 }
382
383 Ok(())
384 }
385
386 async fn install_wasi_sdk_if_needed(&self) -> Result<PathBuf> {
387 let url = if let Some(asset_name) = WASI_SDK_ASSET_NAME {
388 format!("{WASI_SDK_URL}{asset_name}")
389 } else {
390 bail!("wasi-sdk is not available for platform {}", env::consts::OS);
391 };
392
393 let wasi_sdk_dir = self.cache_dir.join("wasi-sdk");
394 let mut clang_path = wasi_sdk_dir.clone();
395 clang_path.extend(["bin", &format!("clang{}", env::consts::EXE_SUFFIX)]);
396
397 if fs::metadata(&clang_path).is_ok_and(|metadata| metadata.is_file()) {
398 return Ok(clang_path);
399 }
400
401 let mut tar_out_dir = wasi_sdk_dir.clone();
402 tar_out_dir.set_extension("archive");
403
404 fs::remove_dir_all(&wasi_sdk_dir).ok();
405 fs::remove_dir_all(&tar_out_dir).ok();
406
407 log::info!("downloading wasi-sdk to {}", wasi_sdk_dir.display());
408 let mut response = self.http.get(&url, AsyncBody::default(), true).await?;
409 let body = BufReader::new(response.body_mut());
410 let body = GzipDecoder::new(body);
411 let tar = Archive::new(body);
412
413 tar.unpack(&tar_out_dir)
414 .await
415 .context("failed to unpack wasi-sdk archive")?;
416
417 let inner_dir = fs::read_dir(&tar_out_dir)?
418 .next()
419 .context("no content")?
420 .context("failed to read contents of extracted wasi archive directory")?
421 .path();
422 fs::rename(&inner_dir, &wasi_sdk_dir).context("failed to move extracted wasi dir")?;
423 fs::remove_dir_all(&tar_out_dir).ok();
424
425 Ok(clang_path)
426 }
427
428 // This was adapted from:
429 // https://github.com/bytecodealliance/wasm-tools/blob/e8809bb17fcf69aa8c85cd5e6db7cff5cf36b1de/src/bin/wasm-tools/strip.rs
430 fn strip_custom_sections(&self, input: &Vec<u8>) -> Result<Vec<u8>> {
431 use wasmparser::Payload::*;
432
433 let strip_custom_section = |name: &str| {
434 // Default strip everything but:
435 // * the `name` section
436 // * any `component-type` sections
437 // * the `dylink.0` section
438 // * our custom version section
439 name != "name"
440 && !name.starts_with("component-type:")
441 && name != "dylink.0"
442 && name != "zed:api-version"
443 };
444
445 let mut output = Vec::new();
446 let mut stack = Vec::new();
447
448 for payload in Parser::new(0).parse_all(input) {
449 let payload = payload?;
450
451 // Track nesting depth, so that we don't mess with inner producer sections:
452 match payload {
453 Version { encoding, .. } => {
454 output.extend_from_slice(match encoding {
455 wasmparser::Encoding::Component => &wasm_encoder::Component::HEADER,
456 wasmparser::Encoding::Module => &wasm_encoder::Module::HEADER,
457 });
458 }
459 ModuleSection { .. } | ComponentSection { .. } => {
460 stack.push(mem::take(&mut output));
461 continue;
462 }
463 End { .. } => {
464 let mut parent = match stack.pop() {
465 Some(c) => c,
466 None => break,
467 };
468 if output.starts_with(&wasm_encoder::Component::HEADER) {
469 parent.push(ComponentSectionId::Component as u8);
470 output.encode(&mut parent);
471 } else {
472 parent.push(ComponentSectionId::CoreModule as u8);
473 output.encode(&mut parent);
474 }
475 output = parent;
476 }
477 _ => {}
478 }
479
480 if let CustomSection(c) = &payload
481 && strip_custom_section(c.name())
482 {
483 continue;
484 }
485 if let Some((id, range)) = payload.as_section() {
486 RawSection {
487 id,
488 data: &input[range],
489 }
490 .append_to(&mut output);
491 }
492 }
493
494 Ok(output)
495 }
496}
497
498fn populate_defaults(manifest: &mut ExtensionManifest, extension_path: &Path) -> Result<()> {
499 // For legacy extensions on the v0 schema (aka, using `extension.json`), clear out any existing
500 // contents of the computed fields, since we don't care what the existing values are.
501 if manifest.schema_version.is_v0() {
502 manifest.languages.clear();
503 manifest.grammars.clear();
504 manifest.themes.clear();
505 }
506
507 let cargo_toml_path = extension_path.join("Cargo.toml");
508 if cargo_toml_path.exists() {
509 manifest.lib.kind = Some(ExtensionLibraryKind::Rust);
510 }
511
512 let languages_dir = extension_path.join("languages");
513 if languages_dir.exists() {
514 for entry in fs::read_dir(&languages_dir).context("failed to list languages dir")? {
515 let entry = entry?;
516 let language_dir = entry.path();
517 let config_path = language_dir.join("config.toml");
518 if config_path.exists() {
519 let relative_language_dir =
520 language_dir.strip_prefix(extension_path)?.to_path_buf();
521 if !manifest.languages.contains(&relative_language_dir) {
522 manifest.languages.push(relative_language_dir);
523 }
524 }
525 }
526 }
527
528 let themes_dir = extension_path.join("themes");
529 if themes_dir.exists() {
530 for entry in fs::read_dir(&themes_dir).context("failed to list themes dir")? {
531 let entry = entry?;
532 let theme_path = entry.path();
533 if theme_path.extension() == Some("json".as_ref()) {
534 let relative_theme_path = theme_path.strip_prefix(extension_path)?.to_path_buf();
535 if !manifest.themes.contains(&relative_theme_path) {
536 manifest.themes.push(relative_theme_path);
537 }
538 }
539 }
540 }
541
542 let icon_themes_dir = extension_path.join("icon_themes");
543 if icon_themes_dir.exists() {
544 for entry in fs::read_dir(&icon_themes_dir).context("failed to list icon themes dir")? {
545 let entry = entry?;
546 let icon_theme_path = entry.path();
547 if icon_theme_path.extension() == Some("json".as_ref()) {
548 let relative_icon_theme_path =
549 icon_theme_path.strip_prefix(extension_path)?.to_path_buf();
550 if !manifest.icon_themes.contains(&relative_icon_theme_path) {
551 manifest.icon_themes.push(relative_icon_theme_path);
552 }
553 }
554 }
555 }
556
557 let snippets_json_path = extension_path.join("snippets.json");
558 if snippets_json_path.exists() {
559 manifest.snippets = Some(snippets_json_path);
560 }
561
562 // For legacy extensions on the v0 schema (aka, using `extension.json`), we want to populate the grammars in
563 // the manifest using the contents of the `grammars` directory.
564 if manifest.schema_version.is_v0() {
565 let grammars_dir = extension_path.join("grammars");
566 if grammars_dir.exists() {
567 for entry in fs::read_dir(&grammars_dir).context("failed to list grammars dir")? {
568 let entry = entry?;
569 let grammar_path = entry.path();
570 if grammar_path.extension() == Some("toml".as_ref()) {
571 #[derive(Deserialize)]
572 struct GrammarConfigToml {
573 pub repository: String,
574 pub commit: String,
575 #[serde(default)]
576 pub path: Option<String>,
577 }
578
579 let grammar_config = fs::read_to_string(&grammar_path)?;
580 let grammar_config: GrammarConfigToml = toml::from_str(&grammar_config)?;
581
582 let grammar_name = grammar_path
583 .file_stem()
584 .and_then(|stem| stem.to_str())
585 .context("no grammar name")?;
586 if !manifest.grammars.contains_key(grammar_name) {
587 manifest.grammars.insert(
588 grammar_name.into(),
589 GrammarManifestEntry {
590 repository: grammar_config.repository,
591 rev: grammar_config.commit,
592 path: grammar_config.path,
593 },
594 );
595 }
596 }
597 }
598 }
599 }
600
601 Ok(())
602}