Home

Awesome

@mapbox/node-pre-gyp

@mapbox/node-pre-gyp makes it easy to publish and install Node.js C++ addons from binaries

Build status

@mapbox/node-pre-gyp stands between npm and node-gyp and offers a cross-platform method of binary deployment.

Special note on previous package

On Feb 9th, 2021 @mapbox/node-pre-gyp@1.0.0 was released. Older, unscoped versions that are not part of the @mapbox org are deprecated and only @mapbox/node-pre-gyp will see updates going forward. To upgrade to the new package do:

npm uninstall node-pre-gyp --save
npm install @mapbox/node-pre-gyp --save

Features

For a hello world example of a module packaged with node-pre-gyp see https://github.com/springmeyer/node-addon-example and the wiki for real world examples.

Credits

FAQ

See the Frequently Ask Questions.

Depends

Install

node-pre-gyp is designed to be installed as a local dependency of your Node.js C++ addon and accessed like:

./node_modules/.bin/node-pre-gyp --help

But you can also install it globally:

npm install @mapbox/node-pre-gyp -g

Usage

Commands

View all possible commands:

node-pre-gyp --help

You can also chain commands:

node-pre-gyp clean build unpublish publish info

Options

Options include:

Both --build-from-source and --fallback-to-build can be passed alone or they can provide values. You can pass --fallback-to-build=false to override the option as declared in package.json. In addition to being able to pass --build-from-source you can also pass --build-from-source=myapp where myapp is the name of your module.

For example: npm install --build-from-source=myapp. This is useful if:

Configuring

This is a guide to configuring your module to use node-pre-gyp.

1) Add new entries to your package.json

This looks like:

    "dependencies"  : {
      "@mapbox/node-pre-gyp": "1.x"
    },
    "devDependencies": {
      "aws-sdk": "2.x"
    }
    "scripts": {
        "install": "node-pre-gyp install --fallback-to-build"
    },
    "binary": {
        "module_name": "your_module",
        "module_path": "./lib/binding/",
        "host": "https://your_module.s3-us-west-1.amazonaws.com"
    }

For a full example see node-addon-examples's package.json.

Let's break this down:

Note: in the past we recommended putting @mapbox/node-pre-gyp in the bundledDependencies, but we no longer recommend this. In the past there were npm bugs (with node versions 0.10.x) that could lead to node-pre-gyp not being available at the right time during install (unless we bundled). This should no longer be the case. Also, for a time we recommended using "preinstall": "npm install @mapbox/node-pre-gyp" as an alternative method to avoid needing to bundle. But this did not behave predictably across all npm versions - see https://github.com/mapbox/node-pre-gyp/issues/260 for the details. So we do not recommend using preinstall to install @mapbox/node-pre-gyp. More history on this at https://github.com/strongloop/fsevents/issues/157#issuecomment-265545908.

The binary object has three required properties
module_name

The name of your native node module. This value must:

module_path

The location your native module is placed after a build. This should be an empty directory without other Javascript files. This entire directory will be packaged in the binary tarball. When installing from a remote package this directory will be overwritten with the contents of the tarball.

Note: This property supports variables based on Versioning.

host

A url to the remote location where you've published tarball binaries (must be https not http).

It is highly recommended that you use Amazon S3. The reasons are:

Why then not require S3? Because while some applications using node-pre-gyp need to distribute binaries as large as 20-30 MB, others might have very small binaries and might wish to store them in a GitHub repo. This is not recommended, but if an author really wants to host in a non-S3 location then it should be possible.

It should also be mentioned that there is an optional and entirely separate npm module called node-pre-gyp-github which is intended to complement node-pre-gyp and be installed along with it. It provides the ability to store and publish your binaries within your repositories GitHub Releases if you would rather not use S3 directly. Installation and usage instructions can be found here, but the basic premise is that instead of using the node-pre-gyp publish command you would use node-pre-gyp-github publish.

The binary object other optional S3 properties

If you are not using a standard s3 path like bucket_name.s3(.-)region.amazonaws.com, you might get an error on publish because node-pre-gyp extracts the region and bucket from the host url. For example, you may have an on-premises s3-compatible storage server, or may have configured a specific dns redirecting to an s3 endpoint. In these cases, you can explicitly set the region and bucket properties to tell node-pre-gyp to use these values instead of guessing from the host property. The following values can be used in the binary section:

host

The url to the remote server root location (must be https not http).

bucket

The bucket name where your tarball binaries should be located.

region

Your S3 server region.

s3ForcePathStyle

Set s3ForcePathStyle to true if the endpoint url should not be prefixed with the bucket name. If false (default), the server endpoint would be constructed as bucket_name.your_server.com.

The binary object has optional properties
remote_path

It is recommended that you customize this property. This is an extra path to use for publishing and finding remote tarballs. The default value for remote_path is "" meaning that if you do not provide it then all packages will be published at the base of the host. It is recommended to provide a value like ./{name}/v{version} to help organize remote packages in the case that you choose to publish multiple node addons to the same host.

Note: This property supports variables based on Versioning.

package_name

It is not recommended to override this property unless you are also overriding the remote_path. This is the versioned name of the remote tarball containing the binary .node module and any supporting files you've placed inside the module_path directory. Unless you specify package_name in your package.json then it defaults to {module_name}-v{version}-{node_abi}-{platform}-{arch}.tar.gz which allows your binary to work across node versions, platforms, and architectures. If you are using remote_path that is also versioned by ./{module_name}/v{version} then you could remove these variables from the package_name and just use: {node_abi}-{platform}-{arch}.tar.gz. Then your remote tarball will be looked up at, for example, https://example.com/your-module/v0.1.0/node-v11-linux-x64.tar.gz.

Avoiding the version of your module in the package_name and instead only embedding in a directory name can be useful when you want to make a quick tag of your module that does not change any C++ code. In this case you can just copy binaries to the new version behind the scenes like:

aws s3 sync --acl public-read s3://mapbox-node-binary/sqlite3/v3.0.3/ s3://mapbox-node-binary/sqlite3/v3.0.4/

Note: This property supports variables based on Versioning.

2) Add a new target to binding.gyp

node-pre-gyp calls out to node-gyp to compile the module and passes variables along like module_name and module_path.

A new target must be added to binding.gyp that moves the compiled .node module from ./build/Release/module_name.node into the directory specified by module_path.

Add a target like this at the end of your targets list:

    {
      "target_name": "action_after_build",
      "type": "none",
      "dependencies": [ "<(module_name)" ],
      "copies": [
        {
          "files": [ "<(PRODUCT_DIR)/<(module_name).node" ],
          "destination": "<(module_path)"
        }
      ]
    }

For a full example see node-addon-example's binding.gyp.

3) Dynamically require your .node

Inside the main js file that requires your addon module you are likely currently doing:

var binding = require('../build/Release/binding.node');

or:

var bindings = require('./bindings')

Change those lines to:

var binary = require('@mapbox/node-pre-gyp');
var path = require('path');
var binding_path = binary.find(path.resolve(path.join(__dirname,'./package.json')));
var binding = require(binding_path);

For a full example see node-addon-example's index.js

4) Build and package your app

Now build your module from source:

npm install --build-from-source

The --build-from-source tells node-pre-gyp to not look for a remote package and instead dispatch to node-gyp to build.

Now node-pre-gyp should now also be installed as a local dependency so the command line tool it offers can be found at ./node_modules/.bin/node-pre-gyp.

5) Test

Now npm test should work just as it did before.

6) Publish the tarball

Then package your app:

./node_modules/.bin/node-pre-gyp package

Once packaged, now you can publish:

./node_modules/.bin/node-pre-gyp publish

Currently the publish command pushes your binary to S3. This requires:

You can also host your binaries elsewhere. To do this requires:

7) Automate builds

Now you need to publish builds for all the platforms and node versions you wish to support. This is best automated.

8) You're done!

Now publish your module to the npm registry. Users will now be able to install your module from a binary.

What will happen is this:

  1. npm install <your package> will pull from the npm registry
  2. npm will run the install script which will call out to node-pre-gyp
  3. node-pre-gyp will fetch the binary .node module and unpack in the right place
  4. Assuming that all worked, you are done

If a a binary was not available for a given platform and --fallback-to-build was used then node-gyp rebuild will be called to try to source compile the module.

9) One more option

It may be that you want to work with two s3 buckets, one for staging and one for production; this arrangement makes it less likely to accidentally overwrite a production binary. It also allows the production environment to have more restrictive permissions than staging while still enabling publishing when developing and testing.

The binary.host property can be set at execution time. In order to do so all of the following conditions must be true.

If any of these checks fail then the operation will not perform execution time determination of the s3 target.

If the command being executed is either "publish" or "unpublish" then the default is set to binary.staging_host. In all other cases the default is binary.production_host.

The command-line options --s3_host=staging or --s3_host=production override the default. If s3_host is present and not staging or production an exception is thrown.

This allows installing from staging by specifying --s3_host=staging. And it requires specifying --s3_option=production in order to publish to, or unpublish from, production, making accidental errors less likely.

Node-API Considerations

Node-API, which was previously known as N-API, is an ABI-stable alternative to previous technologies such as nan which are tied to a specific Node runtime engine. Node-API is Node runtime engine agnostic and guarantees modules created today will continue to run, without changes, into the future.

Using node-pre-gyp with Node-API projects requires a handful of additional configuration values and imposes some additional requirements.

The most significant difference is that an Node-API module can be coded to target multiple Node-API versions. Therefore, an Node-API module must declare in its package.json file which Node-API versions the module is designed to run against. In addition, since multiple builds may be required for a single module, path and file names must be specified in way that avoids naming conflicts.

The napi_versions array property

A Node-API module must declare in its package.json file, the Node-API versions the module is intended to support. This is accomplished by including an napi-versions array property in the binary object. For example:

"binary": {
    "module_name": "your_module",
    "module_path": "your_module_path",
    "host": "https://your_bucket.s3-us-west-1.amazonaws.com",
    "napi_versions": [1,3]
  }

If the napi_versions array property is not present, node-pre-gyp operates as it always has. Including the napi_versions array property instructs node-pre-gyp that this is a Node-API module build.

When the napi_versions array property is present, node-pre-gyp fires off multiple operations, one for each of the Node-API versions in the array. In the example above, two operations are initiated, one for Node-API version 1 and second for Node-API version 3. How this version number is communicated is described next.

The napi_build_version value

For each of the Node-API module operations node-pre-gyp initiates, it ensures that the napi_build_version is set appropriately.

This value is of importance in two areas:

  1. The C/C++ code which needs to know against which Node-API version it should compile.
  2. node-pre-gyp itself which must assign appropriate path and file names to avoid collisions.

Defining NAPI_VERSION for the C/C++ code

The napi_build_version value is communicated to the C/C++ code by adding this code to the binding.gyp file:

"defines": [
    "NAPI_VERSION=<(napi_build_version)",
]

This ensures that NAPI_VERSION, an integer value, is declared appropriately to the C/C++ code for each build.

Note that earlier versions of this document recommended defining the symbol NAPI_BUILD_VERSION. NAPI_VERSION is preferred because it used by the Node-API C/C++ headers to configure the specific Node-API versions being requested.

Path and file naming requirements in package.json

Since node-pre-gyp fires off multiple operations for each request, it is essential that path and file names be created in such a way as to avoid collisions. This is accomplished by imposing additional path and file naming requirements.

Specifically, when performing Node-API builds, the {napi_build_version} text configuration value must be present in the module_path property. In addition, the {napi_build_version} text configuration value must be present in either the remote_path or package_name property. (No problem if it's in both.)

Here's an example:

"binary": {
    "module_name": "your_module",
    "module_path": "./lib/binding/napi-v{napi_build_version}",
    "remote_path": "./{module_name}/v{version}/{configuration}/",
    "package_name": "{platform}-{arch}-napi-v{napi_build_version}.tar.gz",
    "host": "https://your_bucket.s3-us-west-1.amazonaws.com",
    "napi_versions": [1,3]
  }

Supporting both Node-API and NAN builds

You may have a legacy native add-on that you wish to continue supporting for those versions of Node that do not support Node-API, as you add Node-API support for later Node versions. This can be accomplished by specifying the node_napi_label configuration value in the package.json binary.package_name property.

Placing the configuration value node_napi_label in the package.json binary.package_name property instructs node-pre-gyp to build all viable Node-API binaries supported by the current Node instance. If the current Node instance does not support Node-API, node-pre-gyp will request a traditional, non-Node-API build.

The configuration value node_napi_label is set by node-pre-gyp to the type of build created, napi or node, and the version number. For Node-API builds, the string contains the Node-API version nad has values like napi-v3. For traditional, non-Node-API builds, the string contains the ABI version with values like node-v46.

Here's how the binary configuration above might be changed to support both Node-API and NAN builds:

"binary": {
    "module_name": "your_module",
    "module_path": "./lib/binding/{node_napi_label}",
    "remote_path": "./{module_name}/v{version}/{configuration}/",
    "package_name": "{platform}-{arch}-{node_napi_label}.tar.gz",
    "host": "https://your_bucket.s3-us-west-1.amazonaws.com",
    "napi_versions": [1,3]
  }

The C/C++ symbol NAPI_VERSION can be used to distinguish Node-API and non-Node-API builds. The value of NAPI_VERSION is set to the integer Node-API version for Node-API builds and is set to 0 for non-Node-API builds.

For example:

#if NAPI_VERSION
// Node-API code goes here
#else
// NAN code goes here
#endif

Two additional configuration values

The following two configuration values, which were implemented in previous versions of node-pre-gyp, continue to exist, but have been replaced by the node_napi_label configuration value described above.

  1. napi_version If Node-API is supported by the currently executing Node instance, this value is the Node-API version number supported by Node. If Node-API is not supported, this value is an empty string.

  2. node_abi_napi If the value returned for napi_version is non empty, this value is 'napi'. If the value returned for napi_version is empty, this value is the value returned for node_abi.

These values are present for use in the binding.gyp file and may be used as {napi_version} and {node_abi_napi} for text substitution in the binary properties of the package.json file.

S3 Hosting

You can host wherever you choose but S3 is cheap, node-pre-gyp publish expects it, and S3 can be integrated well with Travis.ci to automate builds for OS X and Ubuntu, and with Appveyor to automate builds for Windows. Here is an approach to do this:

First, get setup locally and test the workflow:

1) Create an S3 bucket

And have your key and secret key ready for writing to the bucket.

It is recommended to create a IAM user with a policy that only gives permissions to the specific bucket you plan to publish to. This can be done in the IAM console by: 1) adding a new user, 2) choosing Attach User Policy, 3) Using the Policy Generator, 4) selecting Amazon S3 for the service, 5) adding the actions: DeleteObject, GetObject, GetObjectAcl, ListBucket, HeadBucket, PutObject, PutObjectAcl, 6) adding an ARN of arn:aws:s3:::bucket/* (replacing bucket with your bucket name), and finally 7) clicking Add Statement and saving the policy. It should generate a policy like:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Sid": "objects",
            "Effect": "Allow",
            "Action": [
                "s3:PutObject",
                "s3:GetObjectAcl",
                "s3:GetObject",
                "s3:DeleteObject",
                "s3:PutObjectAcl"
            ],
            "Resource": "arn:aws:s3:::your-bucket-name/*"
        },
        {
            "Sid": "bucket",
            "Effect": "Allow",
            "Action": "s3:ListBucket",
            "Resource": "arn:aws:s3:::your-bucket-name"
        },
        {
            "Sid": "buckets",
            "Effect": "Allow",
            "Action": "s3:HeadBucket",
            "Resource": "*"
        }
    ]
}

2) Install node-pre-gyp

Either install it globally:

npm install node-pre-gyp -g

Or put the local version on your PATH

export PATH=`pwd`/node_modules/.bin/:$PATH

3) Configure AWS credentials

It is recommended to configure the AWS JS SDK v2 used internally by node-pre-gyp by setting these environment variables:

But also you can also use the Shared Config File mentioned in the AWS JS SDK v2 docs

4) Package and publish your build

Install the aws-sdk:

npm install aws-sdk

Then publish:

node-pre-gyp package publish

Note: if you hit an error like Hostname/IP doesn't match certificate's altnames it may mean that you need to provide the region option in your config.

Appveyor Automation

Appveyor can build binaries and publish the results per commit and supports:

For an example of doing this see node-sqlite3's appveyor.yml.

Below is a guide to getting set up:

1) Create a free Appveyor account

Go to https://ci.appveyor.com/signup/free and sign in with your GitHub account.

2) Create a new project

Go to https://ci.appveyor.com/projects/new and select the GitHub repo for your module

3) Add appveyor.yml and push it

Once you have committed an appveyor.yml (appveyor.yml reference) to your GitHub repo and pushed it AppVeyor should automatically start building your project.

4) Create secure variables

Encrypt your S3 AWS keys by going to https://ci.appveyor.com/tools/encrypt and hitting the encrypt button.

Then paste the result into your appveyor.yml

environment:
  AWS_ACCESS_KEY_ID:
    secure: Dn9HKdLNYvDgPdQOzRq/DqZ/MPhjknRHB1o+/lVU8MA=
  AWS_SECRET_ACCESS_KEY:
    secure: W1rwNoSnOku1r+28gnoufO8UA8iWADmL1LiiwH9IOkIVhDTNGdGPJqAlLjNqwLnL

NOTE: keys are per account but not per repo (this is difference than Travis where keys are per repo but not related to the account used to encrypt them).

5) Hook up publishing

Just put node-pre-gyp package publish in your appveyor.yml after npm install.

6) Publish when you want

You might wish to publish binaries only on a specific commit. To do this you could borrow from the Travis CI idea of commit keywords and add special handling for commit messages with [publish binary]:

SET CM=%APPVEYOR_REPO_COMMIT_MESSAGE%
if not "%CM%" == "%CM:[publish binary]=%" node-pre-gyp --msvs_version=2013 publish

If your commit message contains special characters (e.g. &) this method might fail. An alternative is to use PowerShell, which gives you additional possibilities, like ignoring case by using ToLower():

ps: if($env:APPVEYOR_REPO_COMMIT_MESSAGE.ToLower().Contains('[publish binary]')) { node-pre-gyp --msvs_version=2013 publish }

Remember this publishing is not the same as npm publish. We're just talking about the binary module here and not your entire npm package.

Travis Automation

Travis can push to S3 after a successful build and supports both:

For an example of doing this see node-add-example's .travis.yml.

Note: if you need 32 bit binaries, this can be done from a 64 bit Travis machine. See the node-sqlite3 scripts for an example of doing this.

Below is a guide to getting set up:

1) Install the Travis gem

gem install travis

2) Create secure variables

Make sure you run this command from within the directory of your module.

Use travis-encrypt like:

travis encrypt AWS_ACCESS_KEY_ID=${node_pre_gyp_accessKeyId}
travis encrypt AWS_SECRET_ACCESS_KEY=${node_pre_gyp_secretAccessKey}

Then put those values in your .travis.yml like:

env:
  global:
    - secure: F+sEL/v56CzHqmCSSES4pEyC9NeQlkoR0Gs/ZuZxX1ytrj8SKtp3MKqBj7zhIclSdXBz4Ev966Da5ctmcTd410p0b240MV6BVOkLUtkjZJyErMBOkeb8n8yVfSoeMx8RiIhBmIvEn+rlQq+bSFis61/JkE9rxsjkGRZi14hHr4M=
    - secure: o2nkUQIiABD139XS6L8pxq3XO5gch27hvm/gOdV+dzNKc/s2KomVPWcOyXNxtJGhtecAkABzaW8KHDDi5QL1kNEFx6BxFVMLO8rjFPsMVaBG9Ks6JiDQkkmrGNcnVdxI/6EKTLHTH5WLsz8+J7caDBzvKbEfTux5EamEhxIWgrI=

More details on Travis encryption at http://about.travis-ci.org/docs/user/encryption-keys/.

3) Hook up publishing

Just put node-pre-gyp package publish in your .travis.yml after npm install.

OS X publishing

If you want binaries for OS X in addition to linux you can enable multi-os for Travis

Use a configuration like:


language: cpp

os:
- linux
- osx

env:
  matrix:
    - NODE_VERSION="4"
    - NODE_VERSION="6"

before_install:
- rm -rf ~/.nvm/ && git clone --depth 1 https://github.com/creationix/nvm.git ~/.nvm
- source ~/.nvm/nvm.sh
- nvm install $NODE_VERSION
- nvm use $NODE_VERSION

See Travis OS X Gotchas for why we replace language: node_js and node_js: sections with language: cpp and a custom matrix.

Also create platform specific sections for any deps that need install. For example if you need libpng:

- if [ $(uname -s) == 'Linux' ]; then apt-get install libpng-dev; fi;
- if [ $(uname -s) == 'Darwin' ]; then brew install libpng; fi;

For detailed multi-OS examples see node-mapnik and node-sqlite3.

Travis OS X Gotchas

First, unlike the Travis Linux machines, the OS X machines do not put node-pre-gyp on PATH by default. To do so you will need to:

export PATH=$(pwd)/node_modules/.bin:${PATH}

Second, the OS X machines do not support using a matrix for installing different Node.js versions. So you need to bootstrap the installation of Node.js in a cross platform way.

By doing:

env:
  matrix:
    - NODE_VERSION="4"
    - NODE_VERSION="6"

before_install:
 - rm -rf ~/.nvm/ && git clone --depth 1 https://github.com/creationix/nvm.git ~/.nvm
 - source ~/.nvm/nvm.sh
 - nvm install $NODE_VERSION
 - nvm use $NODE_VERSION

You can easily recreate the previous behavior of this matrix:

node_js:
  - "4"
  - "6"

4) Publish when you want

You might wish to publish binaries only on a specific commit. To do this you could borrow from the Travis CI idea of commit keywords and add special handling for commit messages with [publish binary]:

COMMIT_MESSAGE=$(git log --format=%B --no-merges -n 1 | tr -d '\n')
if [[ ${COMMIT_MESSAGE} =~ "[publish binary]" ]]; then node-pre-gyp publish; fi;

Then you can trigger new binaries to be built like:

git commit -a -m "[publish binary]"

Or, if you don't have any changes to make simply run:

git commit --allow-empty -m "[publish binary]"

WARNING: if you are working in a pull request and publishing binaries from there then you will want to avoid double publishing when Travis CI builds both the push and pr. You only want to run the publish on the push commit. See https://github.com/Project-OSRM/node-osrm/blob/8eb837abe2e2e30e595093d16e5354bc5c573575/scripts/is_pr_merge.sh which is called from https://github.com/Project-OSRM/node-osrm/blob/8eb837abe2e2e30e595093d16e5354bc5c573575/scripts/publish.sh for an example of how to do this.

Remember this publishing is not the same as npm publish. We're just talking about the binary module here and not your entire npm package. To automate the publishing of your entire package to npm on Travis see http://about.travis-ci.org/docs/user/deployment/npm/

Versioning

The binary properties of module_path, remote_path, and package_name support variable substitution. The strings are evaluated by node-pre-gyp depending on your system and any custom build flags you passed.

The options are visible in the code at https://github.com/mapbox/node-pre-gyp/blob/612b7bca2604508d881e1187614870ba19a7f0c5/lib/util/versioning.js#L114-L127

Download binary files from a mirror

S3 is broken in China for the well known reason.

Using the npm config argument: --{module_name}_binary_host_mirror can download binary files through a mirror, - in module_name will be replaced with _.

e.g.: Install v8-profiler from npm.

$ npm install v8-profiler --profiler_binary_host_mirror=https://npm.taobao.org/mirrors/node-inspector/

e.g.: Install canvas-prebuilt from npm.

$ npm install canvas-prebuilt --canvas_prebuilt_binary_host_mirror=https://npm.taobao.org/mirrors/canvas-prebuilt/