Skip to content

Performance

Andrew Casey edited this page Aug 2, 2021 · 59 revisions

🛑 Read carefully before you jump to conclusions on this page!

There are easy ways to configure TypeScript to ensure faster compilations and editing experiences. The earlier that these practices can be adopted, the better. Beyond best-practices, there are some common techniques for investigating slow compilations/editing experiences, some common fixes, and some common ways of helping the TypeScript team investigate the issues as a last resort.

Writing Easy-to-Compile Code

Preferring Interfaces Over Intersections

Much of the time, a simple type alias to an object type acts very similarly to an interface.

interface Foo { prop: string }

type Bar = { prop: string };

However, and as soon as you need to compose two or more types, you have the option of extending those types with an interface, or intersecting them in a type alias, and that's when the differences start to matter.

Interfaces create a single flat object type that detects property conflicts, which are usually important to resolve! Intersections on the other hand just recursively merge properties, and in some cases produce never. Interfaces also display consistently better, whereas type aliases to intersections can't be displayed in part of other intersections. Type relationships between interfaces are also cached, as opposed to intersection types as a whole. A final noteworthy difference is that when checking against a target intersection type, every constituent is checked before checking against the "effective"/"flattened" type.

For this reason, extending types with interfaces/extends is suggested over creating intersection types.

- type Foo = Bar & Baz & {
-     someProp: string;
- }
+ interface Foo extends Bar, Baz {
+     someProp: string;
+ }

Using Type Annotations

Adding type annotations, especially return types, can save the compiler a lot of work. In part, this is because named types tend to be more compact than anonymous types (which the compiler might infer), which reduces the amount of time spent reading and writing declaration files (e.g. for incremental builds). Type inference is very convenient, so there's no need to do this universally - however, it can be a useful thing to try if you've identified a slow section of your code.

- import { otherFunc } from "other";
+ import { otherFunc, otherType } from "other";

- export function func() {
+ export function func(): otherType {
      return otherFunc();
  }

Preferring Base Types Over Unions

Union types are great - they let you express the range of possible values for a type.

interface WeekdaySchedule {
  day: "Monday" | "Tuesday" | "Wednesday" | "Thursday" | "Friday";
  wake: Time;
  startWork: Time;
  endWork: Time;
  sleep: Time;
}

interface WeekendSchedule {
  day: "Saturday" | "Sunday";
  wake: Time;
  familyMeal: Time;
  sleep: Time;
}

declare function printSchedule(schedule: WeekdaySchedule | WeekendSchedule);

However, they also come with a cost. Every time an argument is passed to printSchedule, it has to be compared to each element of the union. For a two-element union, this is trivial and inexpensive. However, if your union has more than a dozen elements, it can cause real problems in compilation speed. For instance, to eliminate redundant members from a union, the elements have to be compared pairwise, which is quadratic. This sort of check might occur when intersecting large unions, where intersecting over each union member can result in enormous types that then need to be reduced. One way to avoid this is to use subtypes, rather than unions.

interface Schedule {
  day: "Monday" | "Tuesday" | "Wednesday" | "Thursday" | "Friday" | "Saturday" | "Sunday";
  wake: Time;
  sleep: Time;
}

interface WeekdaySchedule extends Schedule {
  day: "Monday" | "Tuesday" | "Wednesday" | "Thursday" | "Friday";
  startWork: Time;
  endWork: Time;
}

interface WeekendSchedule extends Schedule {
  day: "Saturday" | "Sunday";
  familyMeal: Time;
}

declare function printSchedule(schedule: Schedule);

A more realistic example of this might come up when trying to model every built-in DOM element type. In this case, it would be preferable to create a base HtmlElement type with common members, which DivElement, ImgElement, etc. extend, rather than to create the exhaustive union DivElement | /*...*/ | ImgElement | /*...*/.

Using Project References

New Code

When building up any codebase of a non-trivial size with TypeScript, it is helpful to organize the codebase into several independent projects. Each project has its own tsconfig.json that has dependencies on other projects. This can be helpful to avoid loading too many files in a single compilation, and also makes certain codebase layout strategies easier to put together.

There are some very basic ways of organizing a codebase into projects. As an example, one might be a program with a project for the client, a project for the server, and a project that's shared between the two.

              ------------
              |          |
              |  Shared  |
              ^----------^
             /            \
            /              \
------------                ------------
|          |                |          |
|  Client  |                |  Server  |
-----^------                ------^-----

Tests can also be broken into their own project.

              ------------
              |          |
              |  Shared  |
              ^-----^----^
             /      |     \
            /       |      \
------------  ------------  ------------
|          |  |  Shared  |  |          |
|  Client  |  |  Tests   |  |  Server  |
-----^------  ------------  ------^-----
     |                            |
     |                            |
------------                ------------
|  Client  |                |  Server  |
|  Tests   |                |  Tests   |
------------                ------------

You can read up more about project references here.

Existing Code

When a workspace becomes so large that it's hard for the editor to handle (and you've used performance tracing to confirm that there are no hotspots, making scale the most likely culprit), it can be helpful to break it down into a collection of projects that reference each other. If you're working in a monorepo, this can be as simple as creating a project for each package and mirroring the package dependency graph in project references. Otherwise the process is more ad hoc - you may be able to follow the directory structure or you may have to use carefully chosen include and exclude globs. Some things to keep in mind:

  • Aim for evenly-sized projects - avoid having a single humongous project with lots of tiny satellites
  • Try to group together files that will be edited together - this will limit the number of projects the editor needs to load
  • Separating out test code can help prevent product code from accidentally depending on it

Performance Considerations

As with any encapsulation mechanism, projects come with a cost. For example, if all projects depend on the same packages (e.g. a popular UI framework), some parts of that package's types will be checked repeatedly - once for each project consuming them. Empirically, it seems that (for a workspace with more than one project) 5-20 projects is an appropriate range - fewer may result in editor slowdowns and more may result in excessive overhead. Some good reasons to split out a project:

  • It has a different output location (e.g. because it's a package in a monorepo)
  • It requires different settings (e.g. lib or moduleResolution)
  • It contains global declarations that you want to scope (either for encapsulation or to limit expensive global rebuilds)
  • The editor's language service runs out of memory when trying to process the code as a single project
    • In this case, you will want to set "disableReferencedProjectLoad": true and "disableSolutionSearching": true to limit project loading while editing

Configuring tsconfig.json or jsconfig.json

TypeScript and JavaScript users can always configure their compilations with a tsconfig.json file. jsconfig.json files can also be used to configure the editing experience for JavaScript users.

Specifying Files

You should always make sure that your configuration files aren't including too many files at once.

Within a tsconfig.json, there are two ways to specify files in a project.

  • the files list
  • the include and exclude lists

The primary difference between the two is that files expects a list of file paths to source files, and include/exclude use globbing patterns to match against files.

While specifying files will allow TypeScript to quickly load up files up directly, it can be cumbersome if you have many files in your project without just a few top-level entry-points. Additionally, it's easy to forget to add new files to your tsconfig.json, which means that you might end up with strange editor behavior where those new files are incorrectly analyzed. All this can be cumbersome.

include/exclude help avoid needing to specify these files, but at a cost: files must be discovered by walking through included directories. When running through a lot of folders, this can slow compilations down. Additionally, sometimes a compilation will include lots of unnecessary .d.ts files and test files, which can increase compilation time and memory overhead. Finally, while exclude has some reasonable defaults, certain configurations like mono-repos mean that a "heavy" folders like node_modules can still end up being included.

For best practices, we recommend the following:

  • Specify only input folders in your project (i.e. folders whose source code you want to include for compilation/analysis).
  • Don't mix source files from other projects in the same folder.
  • If keeping tests in the same folder as other source files, give them a distinct name so they can easily be excluded.
  • Avoid large build artifacts and dependency folders like node_modules in source directories.

Note: without an exclude list, node_modules is excluded by default; as soon as one is added, it's important to explicitly add node_modules to the list.

Here is a reasonable tsconfig.json that demonstrates this in action.

{
    "compilerOptions": {
        // ...
    },
    "include": ["src"],
    "exclude": ["**/node_modules", "**/.*/"],
}

Controlling @types Inclusion

By default, TypeScript automatically includes every @types package that it finds in your node_modules folder, regardless of whether you import it. This is meant to make certain things "just work" when using Node.js, Jasmine, Mocha, Chai, etc. since these tools/packages aren't imported - they're just loaded into the global environment.

Sometimes this logic can slow down program construction time in both compilation and editing scenarios, and it can even cause issues with multiple global packages with conflicting declarations, causing errors like

Duplicate identifier 'IteratorResult'.
Duplicate identifier 'it'.
Duplicate identifier 'define'.
Duplicate identifier 'require'.

In cases where no global package is required, the fix is as easy as specifying an empty field for the "types" option in a tsconfig.json/jsconfig.json

// src/tsconfig.json
{
   "compilerOptions": {
       // ...

       // Don't automatically include anything.
       // Only include `@types` packages that we need to import.
       "types" : []
   },
   "files": ["foo.ts"]
}

If you still need a few global packages, add them to the types field.

// tests/tsconfig.json
{
   "compilerOptions": {
       // ...

       // Only include `@types/node` and `@types/mocha`.
       "types" : ["node", "mocha"]
   },
   "files": ["foo.test.ts"]
}

Incremental Project Emit

The --incremental flag allows TypeScript to save state from the last compilation to a .tsbuildinfo file. This file is used to figure out the smallest set of files that might to be re-checked/re-emitted since it last ran, much like how TypeScript's --watch mode works.

Incremental compiles are enabled by default when using the composite flag for project references, but can bring the same speed-ups for any project that opts in.

Skipping .d.ts Checking

By default, TypeScript performs a full re-check of all .d.ts files in a project to find issues and inconsistencies; however, this is typically unnecessary. Most of the time, the .d.ts files are known to already work - the way that types extend each other was already verified once, and declarations that matter will be checked anyway.

TypeScript provides the option to skip type-checking of the .d.ts files that it ships with (e.g. lib.d.ts) using the skipDefaultLibCheck flag.

Alternatively, you can also enable the skipLibCheck flag to skip checking all .d.ts files in a compilation.

These two options can often hide misconfiguration and conflicts in .d.ts files, so we suggest using them only for faster builds.

Using Faster Variance Checks

Is a list of dogs a list of animals? That is, is List<Dog> assignable to List<Animals>? The straightforward way to find out is to do a structural comparison of the types, member by member. Unfortunately, this can be very expensive. However, if we know enough about List<T>, we can reduce this assignability check to determining whether Dog is assignable to Animal (i.e. without considering each member of List<T>). (In particular, we need to know the variance of the type parameter T.) The compiler can only take full advantage of this potential speedup if the strictFunctionTypes flag is enabled (otherwise, it uses the slower, but more lenient, structural check). For this reason, we recommend building with --strictFunctionTypes (which is enabled by default under --strict).

Configuring Other Build Tools

TypeScript compilation is often performed with other build tools in mind - especially when writing web apps that might involve a bundler. While we can only make suggestions for a few build tools, ideally these techniques can be generalized.

Make sure that in addition to reading this section, you read up about performance in your choice of build tool - for example:

Concurrent Type-Checking

Type-checking typically requires information from other files, and can be relatively expensive compared to other steps like transforming/emitting code. Because type-checking can take a little bit longer, it can impact the inner development loop - in other words, you might experience a longer edit/compile/run cycle, and this might be frustrating.

For this reason, some build tools can run type-checking in a separate process without blocking emit. While this means that invalid code can run before TypeScript reports an error in your build tool, you'll often see errors in your editor first, and you won't be blocked for as long from running working code.

An example of this in action is the fork-ts-checker-webpack-plugin plugin for Webpack, or awesome-typescript-loader which also sometimes does this.

Isolated File Emit

By default, TypeScript's emit requires semantic information that might not be local to a file. This is to understand how to emit features like const enums and namespaces. But needing to check other files to generate the output for an arbitrary file can make emit slower.

The need for features that need non-local information is somewhat rare - regular enums can be used in place of const enums, and modules can be used instead of namespaces. For that reason, TypeScript provides the isolatedModules flag to error on features powered by non-local information. Enabling isolatedModules means that your codebase is safe for tools that use TypeScript APIs like transpileModule or alternative compilers like Babel.

As an example, the following code won't properly work at runtime with isolated file transforms because const enum values are expected to be inlined; but luckily, isolatedModules will tell us that early on.

// ./src/fileA.ts

export declare const enum E {
    A = 0,
    B = 1,
}

// ./src/fileB.ts

import { E } from "./fileA";

console.log(E.A);
//          ~
// error: Cannot access ambient const enums when the '--isolatedModules' flag is provided.

Remember: isolatedModules doesn't automatically make code generation faster - it just tells you when you're about to use a feature that might not be supported. The thing you're looking for is isolated module emit in different build tools and APIs.

Isolated file emit can be leveraged by using the following tools:

Investigating Issues

There are certain ways to get hints of what might be going wrong.

Disabling Editor Plugins

Editor experiences can be impacted by plugins. Try disabling plugins (especially JavaScript/TypeScript-related plugins) to see if that fixes any issues in performance and responsiveness.

Certain editors also have their own troubleshooting guides for performance, so consider reading up on them. For example, Visual Studio Code has its own page for Performance Issues as well.

extendedDiagnostics

You can run TypeScript with --extendedDiagnostics to get a printout of where the compiler is spending its time.

Files:                         6
Lines:                     24906
Nodes:                    112200
Identifiers:               41097
Symbols:                   27972
Types:                      8298
Memory used:              77984K
Assignability cache size:  33123
Identity cache size:           2
Subtype cache size:            0
I/O Read time:             0.01s
Parse time:                0.44s
Program time:              0.45s
Bind time:                 0.21s
Check time:                1.07s
transformTime time:        0.01s
commentTime time:          0.00s
I/O Write time:            0.00s
printTime time:            0.01s
Emit time:                 0.01s
Total time:                1.75s

Note that Total time won't be the sum of all times preceding it, since there is some overlap and some work is not instrumented.

The most relevant information for most users is:

Field Meaning
Files the number of files that the program is including (use --listFiles to see what they are).
I/O Read time time spent reading from the file system - this includes traversing include'd folders.
Parse time time spent scanning and parsing the program
Program time combined time spent performing reading from the file system, scanning and parsing the program, and other calculation of the program graph. These steps are intermingled and combined here because files need to be resolved and loaded once they're included via imports and exports.
Bind time Time spent building up various semantic information that is local to a single file.
Check time Time spent type-checking the program.
transformTime time Time spent rewriting TypeScript ASTs (trees that represent source files) into forms that work in older runtimes.
commentTime Time spent calculating comments in output files.
I/O Write time Time spent writing/updating files on disk.
printTime Time spent calculating the string representation of an output file and emitting it to disk.

Things that you might want to ask given this input:

showConfig

It's not always obvious what settings a compilation is being run with when running tsc, especially given that tsconfig.jsons can extend other configuration files. showConfig can explain what tsc will calculate for an invocation.

tsc --showConfig

# or to select a specific config file...

tsc --showConfig -p tsconfig.json

traceResolution

Running with traceResolution can help explain why a file was included in a compilation. The emit is somewhat verbose, so you might want to redirect output to a file.

tsc --traceResolution > resolution.txt

If you find a file that shouldn't be present, you may need to look into fixing up your include/exclude lists in your tsconfig.json, or alternatively, you might need to adjust other settings like types, typeRoots, or paths.

Running tsc Alone

Much of the time, users run into slow performance using 3rd party build tools like Gulp, Rollup, Webpack, etc. Running with tsc --extendedDiagnostics to find major discrepancies between using TypeScript and the tool can indicate external misconfiguration or inefficiencies.

Some questions to keep in mind:

  • Is there a major difference in build times between tsc and the build tool with TypeScript integration?
  • If the build tool provides diagnostics, is there a difference between TypeScript's resolution and the build tool's?
  • Does the build tool have its own configuration that could be the cause?
  • Does the build tool have configuration for its TypeScript integration that could be the cause? (e.g. options for ts-loader?)

Upgrading Dependencies

Sometimes TypeScript's type-checking can be impacted by computationally intensive .d.ts files. This is rare, but can happen. Upgrading to a newer version of TypeScript (which can be more efficient) or to a newer version of an @types package (which may have reverted a regression) can often solve the issue.

Performance Tracing

In some cases, the approaches above might not give you enough insight to understand why TypeScript feels slow. TypeScript 4.1 and higher provides a --generateTrace option that can give you a sense of the work the compiler is spending time on. --generateTrace will create an output file that can be analyzed within Edge or Chrome.

Ideally, TypeScript will be able to compile your project without any errors, though it's not a strict requirement for tracing. Once you're ready to get a trace, you can run TypeScript with the --generateTrace flag.

tsc -p ./some/project/src/tsconfig.json --generateTrace tracing_output_folder

You can then

  1. Visit about://tracing on Edge/Chrome,
  2. Click on the Load button at the top left,
  3. Open the generated JSON file (trace.*.json) in your output directory.

You can read more about performance tracing in more detail here.

⚠ Warning: A performance trace may include information from your workspace, including file paths and source code. If you have any concerns about posting this publicly on GitHub, let us know and you can share the details privately.

⚠ Warning: The format of performance trace files is not stable, and may change from version to version.

Common Issues

Once you've trouble-shooted, you might want to explore some fixes to common issues. If the following solutions don't work, it may be worth filing an issue.

Misconfigured include and exclude

As mentioned above, the include/exclude options can be misused in several ways.

Problem Cause Fix
node_modules was accidentally included from deeper folder exclude was not set "exclude": ["**/node_modules", "**/.*/"]
node_modules was accidentally included from deeper folder "exclude": ["node_modules"] "exclude": ["**/node_modules", "**/.*/"]
Hidden dot files (e.g. .git) were accidentally included "exclude": ["**/node_modules"] "exclude": ["**/node_modules", "**/.*/"]
Unexpected files are being included. include was not set "include": ["src"]

Filing an Issue

If your project is already properly and optimally configured, you may want to file an issue.

The best reports of performance issues contain easily obtainable and minimal reproductions of the problem. In other words, a codebase that can easily be cloned over git that contains only a few files. They require either no external integration with build tools - they can either be invoked via tsc or use isolated code which consumes the TypeScript API. Codebases that require complex invocations and setups cannot be prioritized.

We understand that this is not always easy to achieve - specifically, because it is hard to isolate the source of a problem within a codebase, and because sharing intellectual property may be an issue. In some cases, the team will be willing to send a non-disclosure agreement (NDA) if we believe the issue is highly impactful.

Regardless of whether a reproduction is possible, following these directions when filing issues will help us provide you with performance fixes.

Reporting Compiler Performance Issues

Sometimes you'll witness performance issues in both build times as well as editing scenarios. In these cases, it's best to focus on the TypeScript compiler.

First, a nightly version of TypeScript should be used to ensure you're not hitting a resolved issue:

npm install --save-dev typescript@next

# or

yarn add typescript@next --dev

A compiler perf issue should include

  • The version of TypeScript that was installed (i.e. npx tsc -v or yarn tsc -v)
  • The version of Node on which TypeScript ran (i.e. node -v)
  • The output of running with extendedDiagnostics (tsc --extendedDiagnostics -p tsconfig.json)
  • Ideally, a project that demonstrates the issues being encountered.
  • Output logs from profiling the compiler (isolate-*-*-*.log and *.cpuprofile files)

Providing Performance Traces

Performance traces are meant to help teams figure out build performance issues in their own codebases; however, they can also be useful for the TypeScript team in diagnosing and fixing issues. See the above section on performance traces and continue reading more on our dedicated performance tracing page.

Profiling the Compiler

It is important to provide the team with diagnostic traces by running Node.js v10+ with the --trace-ic flag alongside TypeScript with the --generateCpuProfile flag:

node --trace-ic ./node_modules/typescript/lib/tsc.js --generateCpuProfile profile.cpuprofile -p tsconfig.json

Here ./node_modules/typescript/lib/tsc.js can be replaced with any path to where your version of the TypeScript compiler is installed, and tsconfig.json can be any TypeScript configuration file. profile.cpuprofile is an output file of your choice.

This will generate two files:

  • --trace-ic will emit to a file of the isolate-*-*-*.log (e.g. isolate-00000176DB2DF130-17676-v8.log).
  • --generateCpuProfile will emit to a file with the name of your choice. In the above example, it will be a file named profile.cpuprofile.

⚠ Warning: These files may include information from your workspace, including file paths and source code. Both of these files are readable as plain-text, and you can modify them before attaching them as part of a GitHub issue. (e.g. to scrub them of file paths that may expose internal-only information).

However, if you have any concerns about posting these publicly on GitHub, let us know and you can share the details privately.

Reporting Editing Performance Issues

Perceived editing performance is frequently impacted by a number of things, and the only thing within the TypeScript team's control is the performance of the JavaScript/TypeScript language service, as well as the integration between that language service and certain editors (i.e. Visual Studio, Visual Studio Code, Visual Studio for Mac, and Sublime Text). Ensure that all 3rd-party plugins are turned off in your editor to determine whether there is an issue with TypeScript itself.

Editing performance issues are slightly more involved, but the same ideas apply: clone-able minimal repro codebases are ideal, and though in some cases the team will be able to sign an NDA to investigate and isolate issues.

Including the output from tsc --extendedDiagnostics is always good context, but taking a TSServer trace is the most helpful.

Taking a TSServer Log

Collecting a TSServer Log in Visual Studio Code

  1. Open up your command palette and either
    • open your global settings by entering Preferences: Open User Settings
    • open your local project by entering Preferences: Open Workspace Settings
  2. Set the option "typescript.tsserver.log": "verbose",
  3. Restart VS Code and reproduce the problem
  4. In VS Code, run the TypeScript: Open TS Server log command
  5. This should open the tsserver.log file.

⚠ Warning: A TSServer log may include information from your workspace, including file paths and source code. If you have any concerns about posting this publicly on GitHub, let us know and you can share the details privately.

Clone this wiki locally