good-fences
"Good fences make good neighbors." — Robert Frost, Mending Wall
What is good-fences?
Good-fences is a tool that allows you to segment a TypeScript project into conceptual areas and manage dependencies between those areas.
This is mostly a concern for large projects with many people working on them, where it is impossible for every developer to have a wholistic understanding of the entire codebase.
JavaScript's module system is a specialized form of anarchy because any file can import any other file, possibly allowing access to code that was really meant to be an internal implementation detail of some larger system.
Other languages have concepts like DLL boundaries and the internal
keyword to mitigate this.
Good-fences provides a way to enforce similar boundaries in the TypeScript world.
Getting started
-
Install:
npm install -g good-fences
-
Configure with one or more fence.json files (see below).
-
Run:
gf [options]
Or run good-fences programmatically via the API:
import { run } from 'good-fences'; run(options);
Also see the sample project which demonstrates how fences can be configured.
To run good-fences against the sample, just clone this repository and run npm run sample
.
Configuring good-fences
Good-fences is configured by creating fence.json files throughout your project's directory structure. This configuration file defines a "fence" around that directory (and any subdirectories). Within a fenced directory, modules may import each other without restriction—fences only control what passes in or out of them. Fences may be nested, so that a given directory may have two or more configuration files that apply to it.
A typical fence.json might look like the following.
{
"tags": [ "tag1", "tag2" ],
"exports": [
"index",
{
"modules": "internals/*",
"accessibleTo": "tag3"
}
],
"imports": [
"tag4",
"tag5"
],
"dependencies": [
"dependency1",
"dependency2/lib/**"
]
}
Tags
The tags
property can specify one or more tags to apply to all modules under this config's subdirectory.
Tags are a way of defining a class of files;
for example you might tag all your UI components with 'view', or you might have very granular tags for different areas within your application.
(Or both!)
Tags are used by the other config options to scope which modules are accessible to other modules.
Exports
The exports
property specifies what modules are accessible from the directory.
In other words, it allows you to keep private modules private.
If fence.json contains an exports
definition, then in order for any module outside the directory to import a module under the directory, there must be a matching export rule.
If there is no exports
definition, then all modules are considered exported.
The exports
property is an array of rules. A rule consists of:
- The
modules
glob string which resolves to one or more modules within the directory. (An asterisk ("*"
) indicates all modules under the directory.) - An optional
accessibleTo
property which is a tag (or array of tags) to which these modules are accessible. - If
accessibleTo
is not defined then there is no restriction on where these modules may be imported. (As a convenience, you can just provide a string as an export rule if you don't need to specifyaccessibleTo
.)
This is best demonstrated with an example:
"exports": [
"index",
{
"modules": "views/**/*",
"accessibleTo": "view"
},
{
"modules": "data/store",
"accessibleTo": [ "data", "view" ]
}
]
- The
index
module is accessible to all modules. - Modules under the
/view
directory are accessible to any module tagged with 'view'. - The
data/store
module is accessible to any module tagged with 'data' or 'view'.
Imports
The imports
property specifies what modules may be imported by modules in the directory.
This allows you to control your module graph by restricting unwanted dependencies.
(Note that imports
applies to your project code; for external dependencies see dependencies
below.)
If fence.json contains an imports
definition, then only imports with the given tags will be allowed.
If there is no imports
definition, then any module is free to be imported.
The imports
property is an array of tags:
"imports": [
"tag1",
"tag2"
]
In this case, modules tagged with either 'tag1' or 'tag2' may be imported.
Dependencies
The dependencies
property specifies what external dependencies (i.e. those installed under node_modules) may be imported by modules in the directory.
If fence.json contains an dependencies
definition, then only matching dependencies are allowed.
(If there is no dependencies
definition, then any dependency is free to be imported.)
The dependencies
property is an array of dependencies to allow. Each element can be a simple glob string or an object that allows for more configuration.
"dependencies": [
"dependency1",
"dependency2/lib/**",
{
"dependency": "dependency3",
"accessibleTo": "tag3"
}
]
- The index of
dependency1
may be imported. - Anything under
dependency2/lib
may be imported. - The index of
dependency3
may be imported, but only by modules with the 'tag3' tag.
Options
Project
Specify the tsconfig file to use for your project.
Default | CLI | API |
---|---|---|
./tsconfig.json |
--project <string> -p <string> |
project: string |
Root Directory
Specify the project root directory or directories.
These are the folders that will be scanned for fences, and if running with
--looseRootFileDiscovery
, the directories that will be scanned for source files.
Default | CLI | API |
---|---|---|
process.cwd() |
--rootDir <string...> -r <string...> |
`rootDir: string |
Ignore External Fences
Ignore external fences (e.g. those in node_modules
).
Default | CLI | API |
---|---|---|
false |
--ignoreExternalFences -i |
ignoreExternalFences: boolean |
Loose Root File Discovery
Discover sources from the root directories rather than discovering sources from the project file.
Default | CLI | API |
---|---|---|
false |
--looseRootFileDiscovery -x |
looseRootFileDiscovery: boolean |
Since Git Hash
Only run on files changed between the current git index and the given commit hash or reference name. If the git index is empty, good-fences will check against the current HEAD instead.
Default | CLI | API |
---|---|---|
undefined |
--sinceGitHash <string> -g <string> |
sinceGitHash: string |
Partial Check Limit
When running in a partial check (e.g. with --sinceGitHash
), the maximum number
of source files to check. If more than this number of files have changed in the
partial check (including fences and source files), good-fences will exit with
code 0. This is intended for using good-fences as a pre-commit check.
Default | CLI | API |
---|---|---|
undefined |
--partialCheckLimit <number> -l <number> |
partialCheckLimit: number |
Show Progress Bar
Whether a progress bar should be displayed on the process stderr during fence
checking. Does not show while discovering files, only while actually running
fences, so it may take several minutes to show on large projects not running
with --looseRootFileDiscovery
.
Default | CLI | API |
---|---|---|
false |
--progressBar <boolean> -p <boolean> |
maxConcurrentFenceJobs: boolean |
Max Concurrent Fence Jobs
The maximum number of fence jobs to run at the same time. Should be set below MFILE on your machine, as otherwise good-fences will hit EMFILE and crash out.
Default | CLI | API |
---|---|---|
6000 |
--maxConcurrentFenceJobs <number> -j <number> |
maxConcurrentFenceJobs: number |
Exclude Files Patten
Specify a pattern for file names that should be excluded from good-fences validation. For example *.test.ts to exclude test files.
Default | CLI | API |
---|---|---|
undefined |
--excludeFilesPattern <string> -p <string> |
excludeFilesPattern: string |
Return value
When running good-fences via the API, the results are returned in a structure like the following:
{
"errors": [
{
"message": "The error message",
"sourceFile": "The source file where the error was encountered",
"rawImport": "The offending import",
"fencePath": "The fence whose rule was violated",
"detailedMessage": "A human-friendly message that includes all of the above"
}
],
"warnings": [
{
"message": "The warning message",
"fencePath": "The fence which generated the warning",
"detailedMessage": "A human-friendly message that includes all of the above"
}
]
}