rxjs-tslint-rules

TSLint is deprecated. All of the rules in this package - with the exception of the RxJS-version-5-only rules - have equivalent ESLint rules in the eslint-plugin-rxjs and eslint-plugin-rxjs-angular packages.

If you've arrived here looking for the TSLint rules that automatically convert RxJS version 5 code to version 6, you can find those rules here: rxjs-tslint.

That said, if you've not already done so, you might want to checkout the rules in this package, too. Using them, you can avoid potential problems and questionable practices.

What is it?

rxjs-tslint-rules is set of TSLint rules to:

• help manage projects that use rxjs/add/... imports;
• enforce or disallow Finnish notation; and
• highlight other potential problems (see the rules for details).

Why might you need it?

When using imports that patch Observable:

import { Observable } from "rxjs/Observable";
import "rxjs/add/observable/of";
import "rxjs/add/operator/map";

TypeScript will see the merged declarations in all modules, making it difficult to find rxjs/add/... imports that are missing from modules in which patched observables and operators are used.

This can cause problems, as whether or not Observable is patched then depends upon the order in which the modules are executed.

The rules in this package can be used to highlight missing - or unused - imports and other potential problems with RxJS.

There are some examples of policies that can be implemented using particular rule combinations in:

• Managing RxJS Imports with TSLint; and
• Understanding Lettable Operators.
• TSLint Rules for Version 6

And Christian Liebel has written about his approach to importing RxJS in his blog post:

• Angular & TypeScript: How to Import RxJS Correctly?

Install

Install the package using NPM:

npm install rxjs-tslint-rules --save-dev

Update your tslint.json file to extend this package:

{
  "extends": [
    "rxjs-tslint-rules"
  ],
  "rules": {
    "rxjs-add": { "severity": "error" },
    "rxjs-no-unused-add": { "severity": "error" }
  }
}

Rules

WARNING: Before configuring any of the following rules, you should ensure that TSLint's no-unused-variable rule is not enabled in your configuration (or in any configuration that you extend). That rule has caused problems in the past - as it leaves the TypeScript program in an unstable state - and has a significant number of still-open issues. Consider using the no-unused-declaration rule from tslint-etc instead.

The package includes the following rules (none of which are enabled by default):

Options

rxjs-add

The rxjs-add rule takes an optional object with the property file. This is the path of the module - relative to the tsconfig.json - that imports the patched observables and operators.

For example:

"rules": {
  "rxjs-add": {
    "options": [{
      "allowElsewhere": false,
      "allowUnused": false,
      "file": "./source/rxjs-imports.ts"
    }],
    "severity": "error"
  }
}

Specifying the file option allows all of the patched observables and operators to be kept in a central location. Said module should be imported before other modules that use patched observables and operators. The importation of said module is not enforced; the rule only ensures that it imports observables and operators that are used in other modules.

If file is specified, the allowElsewhere and allowUnused options can be used to configure whether or not patched imports are allowed in other files and whether or not unused patched imports are allowed. Both allowElsewhere and allowUnused default to false.

Note that there is no file option for the rxjs-no-unused-add rule, so that rule should not be used in conjunction with the rxjs-add rule - if the file option is specified for the latter. Use the rxjs-add rule's allowUnused option instead.

If the file option is not specified, patched observables and operators must be imported in the modules in which they are used.

rxjs-ban-observables/operators

The rxjs-ban-observables and rxjs-ban-operators rules take an object containing keys that are the names of observables/operators and values that are either booleans or strings containing the explanation for the ban.

For example:

"rules": {
  "rxjs-ban-operators": {
    "options": [{
      "concat": "Use the concat factory function",
      "merge": "Use the merge factory function"
    }],
    "severity": "error"
  }
}

rxjs-finnish

The rxjs-finnish rule takes an optional object with optional functions, methods, parameters, properties and variables properties.

The properties are booleans and determine whether or not Finnish notation is enforced. All properties default to true.

For example, to enforce Finnish notation for variables only:

"rules": {
  "rxjs-finnish": {
    "options": [{
      "functions": false,
      "methods": false,
      "parameters": false,
      "properties": false,
      "variables": true
    }],
    "severity": "error"
  }
}

The options also support names and types properties that can be used to prevent the enforcement of Finnish notation for certain names or types. The properties themselves are objects with keys that are regular expressions and values that are booleans.

For example, the following configuration will not enforce Finnish notation for names ending with Stream or for the EventEmitter type:

"rules": {
  "rxjs-finnish": {
    "options": [{
      "names": {
          "Stream$": false
      },
      "types": {
          "^EventEmitter$": false
      }
    }],
    "severity": "error"
  }
}

If the types property is not specified, it will default to not enforcing Finnish notation for Angular's EventEmitter type.

rxjs-no-add and rxjs-no-patched

The rxjs-no-add and rxjs-no-patched rules take an optional object with the optional properties allowObservables and allowOperators. The properties can be specified as booleans - to allow or disallow all observables or operators - or as arrays of strings - to allow or disallow a subset of observables or operators.

For example:

"rules": {
  "rxjs-no-patched": {
    "options": [{
      "allowObservables": ["never", "throw"],
      "allowOperators": false
    }],
    "severity": "error"
  }
}

rxjs-no-exposed-subjects

The rule has an optional allowProtected property that can be specified - it defaults to false:

"rules": {
  "rxjs-no-exposed-subjects": {
    "options": [{
      "allowProtected": true
    }],
    "severity": "error"
  }
}

rxjs-no-sharereplay

This rule disallows the use of shareReplay as, prior to RxJS 6.4.0, it exhibited some surprising behaviour - see this PR for an explanation. In 6.4.0 and later, it's now possible to pass a configuration argument that includes a refCount property - so whether or not the implementation reference counts subscriptions can be determined by the caller.

The rule now has an optional allowConfig property that can be specified - it defaults to false - to allow shareReplay to be used as long as a config argument is passed:

"rules": {
  "rxjs-no-sharereplay": {
    "options": [{
      "allowConfig": true
    }],
    "severity": "error"
  }
}

rxjs-no-unsafe-catch

This rule disallows the usage of catch and catchError operators - in effects and epics - that are not within a flattening operator (switchMap, etc.). Such usage will see the effect or epic complete and stop dispatching actions after an error occurs. See Paul Lessing's article: Handling Errors in NgRx Effects.

The rule takes an optional object with an optional observable property. The property can be specifed as a regular expression string or as an array of words and is used to identify the action observables from which effects and epics are composed.

The following options are equivalent to the rule's default configuration:

"rules": {
  "rxjs-no-unsafe-catch": {
    "options": [{
      "observable": "action(s|\\$)?"
    }],
    "severity": "error"
  }
}

rxjs-no-unsafe-scope

The rule takes an optional object with optional allowDo, allowMethods, allowParameters, allowProperties, allowSubscribe and allowTap properties.

If the allowDo and allowTap options are true, the rule is not applied within do and tap operators respectively.

If the allowParameters option is true, referencing function parameters from outer scopes is allowed.

If the allowMethods option is true, calling methods via this is allowed.

If the allowProperties option is true, accessing properties via this is allowed.

If the allowSubscribe option is true, the rule is not applied within subscribe callbacks.

The following options are equivalent to the rule's default configuration:

"rules": {
  "rxjs-no-unsafe-scope": {
    "options": [{
      "allowDo": true,
      "allowMethods": true,
      "allowParameters": true,
      "allowProperties": false,
      "allowSubscribe": true,
      "allowTap": true
    }],
    "severity": "error"
  }
}

rxjs-no-unsafe-switchmap

The rxjs-no-unsafe-switchmap rule does its best to determine whether or not NgRx effects or redux-observable epics use the switchMap operator with actions for which it could be unsafe.

For example, it would be unsafe to use switchMap in an effect or epic that deletes a resource. If the user were to instigate another delete action whilst one was pending, the pending action would be cancelled and the pending delete might or might not occur. Victor Savkin has mentioned such scenarios in a tweet and I've written an article that's based on his tweet: Avoiding switchMap-Related Bugs.

The rule takes an optional object with optional allow, disallow and observable properties. The properties can be specifed as regular expression strings or as arrays of words.

If the allow option is specified, actions that do not match the regular expression or do not contain any of the specified words will effect an error if switchMap is used.

If the disallow option is specified, actions that match the regular expression or contain one of the specified words will effect an error if switchMap is used.

If neither option is specifed, the rule will default to a set of words are are likely to be present in any actions for which switchMap is unsafe.

The observable property is used to identify the action observables from which effects and epics are composed.

The following options are equivalent to the rule's default configuration:

"rules": {
  "rxjs-no-unsafe-switchmap": {
    "options": [{
      "disallow": ["add", "create", "delete", "post", "put", "remove", "set", "update"],
      "observable": "action(s|\\$)?"
    }],
    "severity": "error"
  }
}

To disallow or warn about all uses of switchMap within effects or epics, use a regular expression that will match all action types:

"rules": {
  "rxjs-no-unsafe-switchmap": {
    "options": [{
      "disallow": "."
    }],
    "severity": "error"
  }
}

rxjs-no-unsafe-takeuntil

The rule takes an optional object with optional alias and allow properties.

The alias property is an array containing the names of operators that aliases for takeUntil.

The allow property is an array containing the names of the operators that are allowed to follow takeUntil.

The following options are equivalent to the rule's default configuration:

"rules": {
  "rxjs-no-unsafe-takeuntil": {
    "options": [{
      "allow": ["count", "defaultIfEmpty", "endWith", "every", "finalize", "finally", "isEmpty", "last", "max", "min", "publish", "publishBehavior", "publishLast", "publishReplay", "reduce", "share", "shareReplay", "skipLast", "takeLast", "throwIfEmpty", "toArray"]
    }],
    "severity": "error"
  }
}

rxjs-prefer-angular-takeuntil

The rule takes an optional object with optional alias, checkDecorators and checkDestroy properties. The alias property is an array containing the names of operators that aliases for takeUntil. The checkDecorators property is an array containing the names of the decorators that determine whether or not a class is checked. And the checkDestroy property is a boolean that determines whether or not a Subject-based ngOnDestroy must be implemented.

rxjs-prefer-observer

The rule takes an optional object with an optional allowNext property. The property defaults to true, allowing a next callback to be passed instead of an observer. For more information, see this RxJS issue.

The following options are equivalent to the rule's default configuration:

"rules": {
  "rxjs-prefer-observer": {
    "options": [{
      "allowNext": true
    }],
    "severity": "error"
  }
}

rxjs-suffix-subjects

The rule takes an optional object with optional parameters, properties and variables properties. The properties are booleans and determine whether or not subjects used in those situations need a suffix.

parameters, properties and variables default to true, and the default suffix is 'Subject'.

The object also has optional types properties which are themselves objects containing keys that are regular expressions and values that are booleans - indicating whether suffixing is required for particular types.

The following options are equivalent to the rule's default configuration:

"rules": {
  "rxjs-suffix-subjects": {
    "options": [{
      "parameters": true,
      "properties": true,
      "suffix": "Subject",
      "variables": true
    }],
    "severity": "error"
  }
}

Gotchas

@angular/cli

Angular's CLI runs TSLint three times:

• first, with application files from src/ (using src/tsconfig.app.json);
• then with the test files from src/ (using src/tsconfig.spec.json);
• and, finally, with files from e2e/ (using e2e/tsconfig.e2e.json).

If you are using the file option of the rxjs-add rule to ensure patched observables and operators are kept in a central location, there are some configuration changes that you should make:

• I'd recommend switching off rxjs-add for the e2e linting, as the central file isn't necessary or appropriate. The simplest way to do this is to create an e2e/tslint.json file with the following content:

    {
      "extends": ["../tslint.json"],
      "rules": {
        "rxjs-add": { "severity": "off" }
      }
    }
  

• And, for the test linting, I'd recommend adding the central file to the TypeScript configuration. If the central file is, say, src/rxjs.imports.ts, add that file to the "files" in src/tsconfig.spec.json:

    "files": [
      "rxjs.imports.ts",
      "test.ts"
    ]
  

  Alternatively, you can import rxjs.imports.ts directly into tests.ts, like this:

    import "./rxjs.imports";
  

With these changes, the rule should play nice with the CLI's running of TSLint. If you are using "allowUnused": false and receive errors about unused operators, you should make sure that files in which those operators are used are imported into at least one test. (The rule will walk all files included in the TypeScript program - not just the specs - so if an unused error is effected, the file using the unused operator is not present in the program and needs to be imported into a test.)

If you experience difficulties in configuring the rules with an @angular/cli-generated application, there is an example in this repo of a working configuration. To see the configuration changes that were made to a vanilla CLI application, have a look at this commit.

Observable.create

Observable.create is declared as a Function, which means that its return type is any. This results in an observable that's not seen by the rules, as they use TypeScript's TypeChecker to determine whether or not a call involves an observable.

The rule implementations include no special handling for this case, so if spurious errors are effected due to Observable.create, explicit typing can resolve them. For example:

const ob: Observable<number> = Observable.create((observer: Observer<number>) => { ...