3 This document describes the high-level architecture of rust-analyzer.
4 If you want to familiarize yourself with the code base, you are just in the right place!
6 You might also enjoy ["Explaining Rust Analyzer"](https://www.youtube.com/playlist?list=PLhb66M_x9UmrqXhQuIpWC5VgTdrGxMx3y) series on YouTube.
7 It goes deeper than what is covered in this document, but will take some time to watch.
9 See also these implementation-related blog posts:
11 * https://rust-analyzer.github.io/blog/2019/11/13/find-usages.html
12 * https://rust-analyzer.github.io/blog/2020/07/20/three-architectures-for-responsive-ide.html
13 * https://rust-analyzer.github.io/blog/2020/09/16/challeging-LR-parsing.html
14 * https://rust-analyzer.github.io/blog/2020/09/28/how-to-make-a-light-bulb.html
15 * https://rust-analyzer.github.io/blog/2020/10/24/introducing-ungrammar.html
17 For older, by now mostly outdated stuff, see the [guide](./guide.md) and [another playlist](https://www.youtube.com/playlist?list=PL85XCvVPmGQho7MZkdW-wtPtuJcFpzycE).
22 ![](https://user-images.githubusercontent.com/4789492/107129398-0ab70f00-687a-11eb-9bfc-d4eb023aec06.png)
24 On the highest level, rust-analyzer is a thing which accepts input source code from the client and produces a structured semantic model of the code.
26 More specifically, input data consists of a set of test files (`(PathBuf, String)` pairs) and information about project structure, captured in the so called `CrateGraph`.
27 The crate graph specifies which files are crate roots, which cfg flags are specified for each crate and what dependencies exist between the crates.
28 This is the input (ground) state.
29 The analyzer keeps all this input data in memory and never does any IO.
30 Because the input data is source code, which typically measures in tens of megabytes at most, keeping everything in memory is OK.
32 A "structured semantic model" is basically an object-oriented representation of modules, functions and types which appear in the source code.
33 This representation is fully "resolved": all expressions have types, all references are bound to declarations, etc.
34 This is derived state.
36 The client can submit a small delta of input data (typically, a change to a single file) and get a fresh code model which accounts for changes.
38 The underlying engine makes sure that model is computed lazily (on-demand) and can be quickly updated for small modifications.
42 `crates/rust-analyzer/src/bin/main.rs` contains the main function which spawns LSP.
43 This is *the* entry point, but it front-loads a lot of complexity, so it's fine to just skim through it.
45 `crates/rust-analyzer/src/handlers.rs` implements all LSP requests and is a great place to start if you are already familiar with LSP.
47 `Analysis` and `AnalysisHost` types define the main API for consumers of IDE services.
51 This section talks briefly about various important directories and data structures.
52 Pay attention to the **Architecture Invariant** sections.
53 They often talk about things which are deliberately absent in the source code.
55 Note also which crates are **API Boundaries**.
56 Remember, [rules at the boundary are different](https://www.tedinski.com/2018/02/06/system-boundaries.html).
60 This is rust-analyzer's "build system".
61 We use cargo to compile rust code, but there are also various other tasks, like release management or local installation.
62 They are handled by Rust code in the xtask directory.
70 rust-analyzer independent libraries which we publish to crates.io.
71 It's not heavily utilized at the moment.
75 It is a hand-written recursive descent parser, which produces a sequence of events like "start node X", "finish node Y".
77 [kotlin's parser](https://github.com/JetBrains/kotlin/blob/4d951de616b20feca92f3e9cc9679b2de9e65195/compiler/frontend/src/org/jetbrains/kotlin/parsing/KotlinParsing.java),
78 which is a good source of inspiration for dealing with syntax errors and incomplete input.
79 Original [libsyntax parser](https://github.com/rust-lang/rust/blob/6b99adeb11313197f409b4f7c4083c2ceca8a4fe/src/libsyntax/parse/parser.rs) is what we use for the definition of the Rust language.
80 `TreeSink` and `TokenSource` traits bridge the tree-agnostic parser from `grammar` with `rowan` trees.
82 **Architecture Invariant:** the parser is independent of the particular tree structure and particular representation of the tokens.
83 It transforms one flat stream of events into another flat stream of events.
84 Token independence allows us to parse out both text-based source code and `tt`-based macro input.
85 Tree independence allows us to more easily vary the syntax tree implementation.
86 It should also unlock efficient light-parsing approaches.
87 For example, you can extract the set of names defined in a file (for typo correction) without building a syntax tree.
89 **Architecture Invariant:** parsing never fails, the parser produces `(T, Vec<Error>)` rather than `Result<T, Error>`.
93 Rust syntax tree structure and parser.
94 See [RFC](https://github.com/rust-lang/rfcs/pull/2256) and [./syntax.md](./syntax.md) for some design notes.
96 - [rowan](https://github.com/rust-analyzer/rowan) library is used for constructing syntax trees.
97 - `ast` provides a type safe API on top of the raw `rowan` tree.
98 - `ungrammar` description of the grammar, which is used to generate `syntax_kinds` and `ast` modules, using `cargo test -p xtask` command.
100 Tests for ra_syntax are mostly data-driven.
101 `test_data/parser` contains subdirectories with a bunch of `.rs` (test vectors) and `.txt` files with corresponding syntax trees.
102 During testing, we check `.rs` against `.txt`.
103 If the `.txt` file is missing, it is created (this is how you update tests).
104 Additionally, running the xtask test suite with `cargo test -p xtask` will walk the grammar module and collect all `// test test_name` comments into files inside `test_data/parser/inline` directory.
106 To update test data, run with `UPDATE_EXPECT` variable:
109 env UPDATE_EXPECT=1 cargo qt
112 After adding a new inline test you need to run `cargo test -p xtask` and also update the test data as described above.
114 Note [`api_walkthrough`](https://github.com/rust-analyzer/rust-analyzer/blob/2fb6af89eb794f775de60b82afe56b6f986c2a40/crates/ra_syntax/src/lib.rs#L190-L348)
115 in particular: it shows off various methods of working with syntax tree.
117 See [#93](https://github.com/rust-analyzer/rust-analyzer/pull/93) for an example PR which fixes a bug in the grammar.
119 **Architecture Invariant:** `syntax` crate is completely independent from the rest of rust-analyzer. It knows nothing about salsa or LSP.
120 This is important because it is possible to make useful tooling using only the syntax tree.
121 Without semantic information, you don't need to be able to _build_ code, which makes the tooling more robust.
122 See also https://web.stanford.edu/~mlfbrown/paper.pdf.
123 You can view the `syntax` crate as an entry point to rust-analyzer.
124 `syntax` crate is an **API Boundary**.
126 **Architecture Invariant:** syntax tree is a value type.
127 The tree is fully determined by the contents of its syntax nodes, it doesn't need global context (like an interner) and doesn't store semantic info.
128 Using the tree as a store for semantic info is convenient in traditional compilers, but doesn't work nicely in the IDE.
129 Specifically, assists and refactors require transforming syntax trees, and that becomes awkward if you need to do something with the semantic info.
131 **Architecture Invariant:** syntax tree is built for a single file.
132 This is to enable parallel parsing of all files.
134 **Architecture Invariant:** Syntax trees are by design incomplete and do not enforce well-formedness.
135 If an AST method returns an `Option`, it *can* be `None` at runtime, even if this is forbidden by the grammar.
139 We use the [salsa](https://github.com/salsa-rs/salsa) crate for incremental and on-demand computation.
140 Roughly, you can think of salsa as a key-value store, but it can also compute derived values using specified functions.
141 The `base_db` crate provides basic infrastructure for interacting with salsa.
142 Crucially, it defines most of the "input" queries: facts supplied by the client of the analyzer.
143 Reading the docs of the `base_db::input` module should be useful: everything else is strictly derived from those inputs.
145 **Architecture Invariant:** particularities of the build system are *not* the part of the ground state.
146 In particular, `base_db` knows nothing about cargo.
147 For example, `cfg` flags are a part of `base_db`, but `feature`s are not.
148 A `foo` feature is a Cargo-level concept, which is lowered by Cargo to `--cfg feature=foo` argument on the command line.
149 The `CrateGraph` structure is used to represent the dependencies between the crates abstractly.
151 **Architecture Invariant:** `base_db` doesn't know about file system and file paths.
152 Files are represented with opaque `FileId`, there's no operation to get an `std::path::Path` out of the `FileId`.
154 ### `crates/hir_expand`, `crates/hir_def`, `crates/hir_ty`
156 These crates are the *brain* of rust-analyzer.
157 This is the compiler part of the IDE.
159 `hir_xxx` crates have a strong [ECS](https://en.wikipedia.org/wiki/Entity_component_system) flavor, in that they work with raw ids and directly query the database.
160 There's little abstraction here.
161 These crates integrate deeply with salsa and chalk.
163 Name resolution, macro expansion and type inference all happen here.
164 These crates also define various intermediate representations of the core.
166 `ItemTree` condenses a single `SyntaxTree` into a "summary" data structure, which is stable over modifications to function bodies.
168 `DefMap` contains the module tree of a crate and stores module scopes.
170 `Body` stores information about expressions.
172 **Architecture Invariant:** these crates are not, and will never be, an api boundary.
174 **Architecture Invariant:** these crates explicitly care about being incremental.
175 The core invariant we maintain is "typing inside a function's body never invalidates global derived data".
176 i.e., if you change the body of `foo`, all facts about `bar` should remain intact.
178 **Architecture Invariant:** hir exists only in context of particular crate instance with specific CFG flags.
179 The same syntax may produce several instances of HIR if the crate participates in the crate graph more than once.
183 The top-level `hir` crate is an **API Boundary**.
184 If you think about "using rust-analyzer as a library", `hir` crate is most likely the façade you'll be talking to.
186 It wraps ECS-style internal API into a more OO-flavored API (with an extra `db` argument for each call).
188 **Architecture Invariant:** `hir` provides a static, fully resolved view of the code.
189 While internal `hir_*` crates _compute_ things, `hir`, from the outside, looks like an inert data structure.
191 `hir` also handles the delicate task of going from syntax to the corresponding `hir`.
192 Remember that the mapping here is one-to-many.
193 See `Semantics` type and `source_to_def` module.
195 Note in particular a curious recursive structure in `source_to_def`.
196 We first resolve the parent _syntax_ node to the parent _hir_ element.
197 Then we ask the _hir_ parent what _syntax_ children does it have.
198 Then we look for our node in the set of children.
200 This is the heart of many IDE features, like goto definition, which start with figuring out the hir node at the cursor.
201 This is some kind of (yet unnamed) uber-IDE pattern, as it is present in Roslyn and Kotlin as well.
205 The `ide` crate builds on top of `hir` semantic model to provide high-level IDE features like completion or goto definition.
206 It is an **API Boundary**.
207 If you want to use IDE parts of rust-analyzer via LSP, custom flatbuffers-based protocol or just as a library in your text editor, this is the right API.
209 **Architecture Invariant:** `ide` crate's API is build out of POD types with public fields.
210 The API uses editor's terminology, it talks about offsets and string labels rather than in terms of definitions or types.
211 It is effectively the view in MVC and viewmodel in [MVVM](https://en.wikipedia.org/wiki/Model%E2%80%93view%E2%80%93viewmodel).
212 All arguments and return types are conceptually serializable.
213 In particular, syntax trees and hir types are generally absent from the API (but are used heavily in the implementation).
214 Shout outs to LSP developers for popularizing the idea that "UI" is a good place to draw a boundary at.
216 `ide` is also the first crate which has the notion of change over time.
217 `AnalysisHost` is a state to which you can transactionally `apply_change`.
218 `Analysis` is an immutable snapshot of the state.
220 Internally, `ide` is split across several crates. `ide_assists`, `ide_completion` and `ide_ssr` implement large isolated features.
221 `ide_db` implements common IDE functionality (notably, reference search is implemented here).
222 The `ide` contains a public API/façade, as well as implementation for a plethora of smaller features.
224 **Architecture Invariant:** `ide` crate strives to provide a _perfect_ API.
225 Although at the moment it has only one consumer, the LSP server, LSP *does not* influence its API design.
226 Instead, we keep in mind a hypothetical _ideal_ client -- an IDE tailored specifically for rust, every nook and cranny of which is packed with Rust-specific goodies.
228 ### `crates/rust-analyzer`
230 This crate defines the `rust-analyzer` binary, so it is the **entry point**.
231 It implements the language server.
233 **Architecture Invariant:** `rust-analyzer` is the only crate that knows about LSP and JSON serialization.
234 If you want to expose a data structure `X` from ide to LSP, don't make it serializable.
235 Instead, create a serializable counterpart in `rust-analyzer` crate and manually convert between the two.
237 `GlobalState` is the state of the server.
238 The `main_loop` defines the server event loop which accepts requests and sends responses.
239 Requests that modify the state or might block user's typing are handled on the main thread.
240 All other requests are processed in background.
242 **Architecture Invariant:** the server is stateless, a-la HTTP.
243 Sometimes state needs to be preserved between requests.
244 For example, "what is the `edit` for the fifth completion item of the last completion edit?".
245 For this, the second request should include enough info to re-create the context from scratch.
246 This generally means including all the parameters of the original request.
248 `reload` module contains the code that handles configuration and Cargo.toml changes.
249 This is a tricky business.
251 **Architecture Invariant:** `rust-analyzer` should be partially available even when the build is broken.
252 Reloading process should not prevent IDE features from working.
254 ### `crates/toolchain`, `crates/project_model`, `crates/flycheck`
256 These crates deal with invoking `cargo` to learn about project structure and get compiler errors for the "check on save" feature.
258 They use `crates/path` heavily instead of `std::path`.
259 A single `rust-analyzer` process can serve many projects, so it is important that server's current directory does not leak.
261 ### `crates/mbe`, `crates/tt`, `crates/proc_macro_api`, `crates/proc_macro_srv`
263 These crates implement macros as token tree -> token tree transforms.
264 They are independent from the rest of the code.
266 `tt` crate defined `TokenTree`, a single token or a delimited sequence of token trees.
267 `mbe` crate contains tools for transforming between syntax trees and token tree.
268 And it also handles the actual parsing and expansion of declarative macro (a-la "Macros By Example" or mbe).
270 For proc macros, the client-server model are used.
271 We pass an argument `--proc-macro` to `rust-analyzer` binary to start a separate process (`proc_macro_srv`).
272 And the client (`proc_macro_api`) provides an interface to talk to that server separately.
274 And then token trees are passed from client, and the server will load the corresponding dynamic library (which built by `cargo`).
275 And due to the fact the api for getting result from proc macro are always unstable in `rustc`,
276 we maintain our own copy (and paste) of that part of code to allow us to build the whole thing in stable rust.
278 **Architecture Invariant:**
279 Bad proc macros may panic or segfault accidentally. So we run it in another process and recover it from fatal error.
280 And they may be non-deterministic which conflict how `salsa` works, so special attention is required.
284 This crate is responsible for parsing, evaluation and general definition of `cfg` attributes.
286 ### `crates/vfs`, `crates/vfs-notify`
288 These crates implement a virtual file system.
289 They provide consistent snapshots of the underlying file system and insulate messy OS paths.
291 **Architecture Invariant:** vfs doesn't assume a single unified file system.
292 i.e., a single rust-analyzer process can act as a remote server for two different machines, where the same `/tmp/foo.rs` path points to different files.
293 For this reason, all path APIs generally take some existing path as a "file system witness".
297 This crate contains various non-rust-analyzer specific utils, which could have been in std, as well
298 as copies of unstable std items we would like to make use of already, like `std::str::split_once`.
302 This crate contains utilities for CPU and memory profiling.
305 ## Cross-Cutting Concerns
307 This sections talks about the things which are everywhere and nowhere in particular.
309 ### Stability Guarantees
311 One of the reasons rust-analyzer moves relatively fast is that we don't introduce new stability guarantees.
312 Instead, as much as possible we leverage existing ones.
316 * The `ide` API of rust-analyzer are explicitly unstable, but the LSP interface is stable, and here we just implement a stable API managed by someone else.
317 * Rust language and Cargo are stable, and they are the primary inputs to rust-analyzer.
318 * The `rowan` library is published to crates.io, but it is deliberately kept under `1.0` and always makes semver-incompatible upgrades
320 Another important example is that rust-analyzer isn't run on CI, so, unlike `rustc` and `clippy`, it is actually ok for us to change runtime behavior.
322 At some point we might consider opening up APIs or allowing crates.io libraries to include rust-analyzer specific annotations, but that's going to be a big commitment on our side.
326 * `rust-project.json` is a de-facto stable format for non-cargo build systems.
327 It is probably ok enough, but was definitely stabilized implicitly.
328 Lesson for the future: when designing API which could become a stability boundary, don't wait for the first users until you stabilize it.
329 By the time you have first users, it is already de-facto stable.
330 And the users will first use the thing, and *then* inform you that now you have users.
331 The sad thing is that stuff should be stable before someone uses it for the first time, or it should contain explicit opt-in.
332 * We ship some LSP extensions, and we try to keep those somewhat stable.
333 Here, we need to work with a finite set of editor maintainers, so not providing rock-solid guarantees works.
337 Some components in this repository are generated through automatic processes.
338 Generated code is updated automatically on `cargo test`.
339 Generated code is generally committed to the git repository.
341 In particular, we generate:
343 * API for working with syntax trees (`syntax::ast`, the [`ungrammar`](https://github.com/rust-analyzer/ungrammar) crate).
344 * Various sections of the manual:
350 * Documentation tests for assists
352 See the `sourcegen` crate for details.
354 **Architecture Invariant:** we avoid bootstrapping.
355 For codegen we need to parse Rust code.
356 Using rust-analyzer for that would work and would be fun, but it would also complicate the build process a lot.
357 For that reason, we use syn and manual string parsing.
361 Let's say that the IDE is in the process of computing syntax highlighting, when the user types `foo`.
363 `rust-analyzer`s answer is that the highlighting process should be cancelled -- its results are now stale, and it also blocks modification of the inputs.
365 The salsa database maintains a global revision counter.
366 When applying a change, salsa bumps this counter and waits until all other threads using salsa finish.
367 If a thread does salsa-based computation and notices that the counter is incremented, it panics with a special value (see `Canceled::throw`).
368 That is, rust-analyzer requires unwinding.
370 `ide` is the boundary where the panic is caught and transformed into a `Result<T, Cancelled>`.
374 Rust Analyzer has three interesting [system boundaries](https://www.tedinski.com/2018/04/10/making-tests-a-positive-influence-on-design.html) to concentrate tests on.
376 The outermost boundary is the `rust-analyzer` crate, which defines an LSP interface in terms of stdio.
377 We do integration testing of this component, by feeding it with a stream of LSP requests and checking responses.
378 These tests are known as "heavy", because they interact with Cargo and read real files from disk.
379 For this reason, we try to avoid writing too many tests on this boundary: in a statically typed language, it's hard to make an error in the protocol itself if messages are themselves typed.
380 Heavy tests are only run when `RUN_SLOW_TESTS` env var is set.
382 The middle, and most important, boundary is `ide`.
383 Unlike `rust-analyzer`, which exposes API, `ide` uses Rust API and is intended for use by various tools.
384 A typical test creates an `AnalysisHost`, calls some `Analysis` functions and compares the results against expectation.
386 The innermost and most elaborate boundary is `hir`.
387 It has a much richer vocabulary of types than `ide`, but the basic testing setup is the same: we create a database, run some queries, assert result.
389 For comparisons, we use the `expect` crate for snapshot testing.
391 To test various analysis corner cases and avoid forgetting about old tests, we use so-called marks.
392 See the `marks` module in the `test_utils` crate for more.
394 **Architecture Invariant:** rust-analyzer tests do not use libcore or libstd.
395 All required library code must be a part of the tests.
396 This ensures fast test execution.
398 **Architecture Invariant:** tests are data driven and do not test the API.
399 Tests which directly call various API functions are a liability, because they make refactoring the API significantly more complicated.
400 So most of the tests look like this:
404 fn check(input: &str, expect: expect_test::Expect) {
405 // The single place that actually exercises a particular API
410 check("foo", expect![["bar"]]);
415 check("spam", expect![["eggs"]]);
417 // ...and a hundred more tests that don't care about the specific API at all.
420 To specify input data, we use a single string literal in a special format, which can describe a set of rust files.
421 See the `Fixture` its module for fixture examples and documentation.
423 **Architecture Invariant:** all code invariants are tested by `#[test]` tests.
424 There's no additional checks in CI, formatting and tidy tests are run with `cargo test`.
426 **Architecture Invariant:** tests do not depend on any kind of external resources, they are perfectly reproducible.
429 ### Performance Testing
431 TBA, take a look at the `metrics` xtask and `#[test] fn benchmark_xxx()` functions.
435 **Architecture Invariant:** core parts of rust-analyzer (`ide`/`hir`) don't interact with the outside world and thus can't fail.
436 Only parts touching LSP are allowed to do IO.
438 Internals of rust-analyzer need to deal with broken code, but this is not an error condition.
439 rust-analyzer is robust: various analysis compute `(T, Vec<Error>)` rather than `Result<T, Error>`.
441 rust-analyzer is a complex long-running process.
442 It will always have bugs and panics.
443 But a panic in an isolated feature should not bring down the whole process.
444 Each LSP-request is protected by a `catch_unwind`.
445 We use `always` and `never` macros instead of `assert` to gracefully recover from impossible conditions.
449 rust-analyzer is a long-running process, so it is important to understand what's going on inside.
450 We have several instruments for that.
452 The event loop that runs rust-analyzer is very explicit.
453 Rather than spawning futures or scheduling callbacks (open), the event loop accepts an `enum` of possible events (closed).
454 It's easy to see all the things that trigger rust-analyzer processing, together with their performance
456 rust-analyzer includes a simple hierarchical profiler (`hprof`).
457 It is enabled with `RA_PROFILE='*>50'` env var (log all (`*`) actions which take more than `50` ms) and produces output like:
460 85ms - handle_completion
461 68ms - import_on_the_fly
462 67ms - import_assets::search_for_relative_paths
463 0ms - crate_def_map:wait (804 calls)
464 0ms - find_path (16 calls)
465 2ms - find_similar_imports (1 calls)
466 0ms - generic_params_query (334 calls)
467 59ms - trait_solve_query (186 calls)
468 0ms - Semantics::analyze_impl (1 calls)
469 1ms - render_resolution (8 calls)
470 0ms - Semantics::analyze_impl (5 calls)
473 This is cheap enough to enable in production.
476 Similarly, we save live object counting (`RA_COUNT=1`).
477 It is not cheap enough to enable in prod, and this is a bug which should be fixed.
481 rust-analyzer strives to be as configurable as possible while offering reasonable defaults where no configuration exists yet.
482 There will always be features that some people find more annoying than helpful, so giving the users the ability to tweak or disable these is a big part of offering a good user experience.
483 Mind the code--architecture gap: at the moment, we are using fewer feature flags than we really should.
487 In Rust, it is easy (often too easy) to add serialization to any type by adding `#[derive(Serialize)]`.
488 This easiness is misleading -- serializable types impose significant backwards compatability constraints.
489 If a type is serializable, then it is a part of some IPC boundary.
490 You often don't control the other side of this boundary, so changing serializable types is hard.
492 For this reason, the types in `ide`, `base_db` and below are not serializable by design.
493 If such types need to cross an IPC boundary, then the client of rust-analyzer needs to provide custom, client-specific serialization format.
494 This isolates backwards compatibility and migration concerns to a specific client.
496 For example, `rust-project.json` is it's own format -- it doesn't include `CrateGraph` as is.
497 Instead, it creates a `CrateGraph` by calling appropriate constructing functions.