Home

Awesome

go-pdfium

Go Reference Build Status codecov

:rocket: Easy to use PDF library using Go and PDFium :rocket:

A fast, multi-threaded and easy to use PDF library for Go applications.

Features

PDFium

This project uses the PDFium C++ library by Google (https://pdfium.googlesource.com/pdfium/) to process the PDF documents. Therefor this project could also be called a binding.

Please be aware that PDFium comes with the Apache License 2.0 license.

Single/Multi-threading

Since PDFium is not a multithreaded C++ library, we can not directly make it multithreaded by calling it from Go's subroutines.

To solve this, this library has 3 different implementations that you can use to call PDFium:

Both single_threaded and multi_threaded requires PDFium to be installed on your machine for your platform during compilation and runtime, it also requires CGO to work on the platform you're compiling to.

webassembly does not need any external dependencies and also does not require CGO to work. However, Wazero currently only has compiler support for amd64 and arm64, meaning it will be using the interpreter on other architectures which will be much, much slower.

All implementations use exactly the same interface, so there won't be any code changes for you to switch between them.

All implementations are thread/subroutine safe, this has been guaranteed by locking the instance that's doing your work while it's doing PDFium operations. New operations will wait until the lock becomes available again.

Be aware that PDFium could use quite some memory depending on the size of the PDF and the requests that you do, so be aware of the amount of workers that you configure.

Implementations

Single/Multi-threading through CGO

Single-threading in CGO works by directly calling the PDFium library from the same process. Single-threaded might be preferred if the caller is managing the workers themselves and does not want the overhead of another process. Be aware that since PDFium is C++, we can't handle segfaults caused by PDFium, which may cause your process to be killed. So using this library in the multi-threaded way, with only 1 worker, can still have some benefits, since it can automatically recover from things like segfaults.

For CGO we have implemented multi-threading using HashiCorp's Go Plugin System, which allows us to launch separate PDFium worker processes, and then route the requests through the different workers. This also makes it a bit more safe to use PDFium, as it's less likely to segfaults or corrupt your main Go application. The Plugin system provides the communication between the processes using gRPC, however, when implementing this library, you won't really see anything of that. From the outside it will look like normal Go code. The inter-process communication does come with a cost as it has to serialize/deserialize input/output as it moves between the main process and the PDFium workers.

Prerequisites (CGO)

To use this Go library, you will need the actual PDFium library to run it and have it available through pkgconfig.

Get the library

You can try to compile PDFium yourself, but you can also use pre-compiled binaries, for example from: https://github.com/bblanchon/pdfium-binaries/releases

If you use a pre-compiled library, make sure to extract it somewhere logical, for example /opt/pdfium.

Configure pkg-config

Create/edit file /usr/lib/pkgconfig/pdfium.pc

prefix={path}
libdir={path}/lib
includedir={path}/include

Name: PDFium
Description: PDFium
Version: 6721
Requires:

Libs: -L${libdir} -lpdfium
Cflags: -I${includedir}

Replace {path} with the path you extracted/compiled pdfium in.

Make sure you extend your library path when running:

export LD_LIBRARY_PATH={path}/lib

You can do this globally or just in your editor.

this can globally be done on ubuntu by editing ~/.profile and adding the line in this file. reloading for bash can be done by relogging or running source ~/.profile can be used to test the change for a terminal

Getting started (CGO)

To get started, make sure that you create a separate package in your application that will start the worker.

The examples below can also be found in the examples folder.

Single-threaded through CGO

For single threaded implementations we just have to initialize the library.

pdfium/renderer/renderer.go

package renderer

import (
	"log"

	"github.com/klippa-app/go-pdfium"
	"github.com/klippa-app/go-pdfium/single_threaded"
)

// Be sure to close pools/instances when you're done with them.
var pool pdfium.Pool
var instance pdfium.Pdfium

func init() {
	// Init the PDFium library and return the instance to open documents.
	pool = single_threaded.Init(single_threaded.Config{})

	var err error
	instance, err = pool.GetInstance(time.Second * 30)
	if err != nil {
		log.Fatal(err)
	}
}
Multi-threaded through CGO
Worker package

This package has to be named main to make it available as a binary. The plugin system will use this to start new PDFium workers. Example:

pdfium/worker/main.go

package main

import (
	"github.com/klippa-app/go-pdfium/multi_threaded/worker"
)

func main() {
	worker.StartWorker(nil)
}
Worker configuration

To actually start workers, you will have to init the PDFium library somewhere, this also allows you to dynamically start workers when needed. The best location to add this is in the init() of a package that is going to call the PDFium library. Example:

pdfium/renderer/renderer.go

package renderer

import (
	"log"

	"github.com/klippa-app/go-pdfium"
	"github.com/klippa-app/go-pdfium/multi_threaded"
)

// Be sure to close pools/instances when you're done with them.
var pool pdfium.Pool
var instance pdfium.Pdfium

func init() {
	// Init the PDFium library and return the instance to open documents.
	// You can tweak these configs to your need. Be aware that workers can use quite some memory.
	pool = multi_threaded.Init(multi_threaded.Config{
		MinIdle:  1, // Makes sure that at least x workers are always available
		MaxIdle:  1, // Makes sure that at most x workers are ever available
		MaxTotal: 1, // Maxium amount of workers in total, allows the amount of workers to grow when needed, items between total max and idle max are automatically cleaned up, while idle workers are kept alive so they can be used directly.
		Command: multi_threaded.Command{
			BinPath: "go",                                     // Only do this while developing, on production put the actual binary path in here. You should not want the Go runtime on production.
			Args:    []string{"run", "pdfium/worker/main.go"}, // This is a reference to the worker package, this can be left empty when using a direct binary path.
		},
	})

	var err error
	instance, err = pool.GetInstance(time.Second * 30)
	if err != nil {
		log.Fatal(err)
	}
}
Get page count
package renderer

import (
	"io/ioutil"
	"log"

	"github.com/klippa-app/go-pdfium"
	"github.com/klippa-app/go-pdfium/requests"
)

// Insert the single/multi-threaded init() here.

func main() {
	filePath := "example.pdf"
	pageCount, err := getPageCount(filePath)
	if err != nil {
		log.Fatal(err)
	}

	log.Printf("The PDF %s has %d pages", filePath, pageCount)
}

func getPageCount(filePath string) (int, error) {
	// Load the PDF file into a byte array.
	pdfBytes, err := ioutil.ReadFile(filePath)
	if err != nil {
		return 0, err
	}

	// Open the PDF using PDFium (and claim a worker)
	doc, err := instance.OpenDocument(&requests.OpenDocument{
		File: &pdfBytes,
	})
	if err != nil {
		return 0, err
	}

	// Always close the document, this will release its resources.
	defer instance.FPDF_CloseDocument(&requests.FPDF_CloseDocument{
		Document: doc.Document,
	})

	pageCount, err := instance.FPDF_GetPageCount(&requests.FPDF_GetPageCount{
		Document: doc,
	})
	if err != nil {
		return 0, err
	}

	return pageCount.PageCount, nil
}
Render a page
package renderer

import (
	"image/png"
	"io/ioutil"
	"log"
	"os"

	"github.com/klippa-app/go-pdfium"
	"github.com/klippa-app/go-pdfium/requests"
)

// Insert the single/multi-threaded init() here.

func main() {
	filePath := "example.pdf"
	output := "example.pdf.png"
	err := renderPage(filePath, 1, output)
	if err != nil {
		log.Fatal(err)
	}
}

func renderPage(filePath string, page int, output string) error {
	// Load the PDF file into a byte array.
	pdfBytes, err := ioutil.ReadFile(filePath)
	if err != nil {
		return err
	}

	// Open the PDF using PDFium (and claim a worker)
	doc, err := instance.OpenDocument(&requests.OpenDocument{
		File: &pdfBytes,
	})
	if err != nil {
		return err
	}

	// Always close the document, this will release its resources.
	defer instance.FPDF_CloseDocument(&requests.FPDF_CloseDocument{
		Document: doc.Document,
	})

	// Render the page in DPI 200.
	pageRender, err := instance.RenderPageInDPI(&requests.RenderPageInDPI{
		DPI: 200, // The DPI to render the page in.
		Page: requests.Page{
			ByIndex: &requests.PageByIndex{
				Document: doc,
				Index:    0,
			},
		}, // The page to render, 0-indexed.
	})
	if err != nil {
		return err
	}

	// Write the output to a file.
	f, err := os.Create(output)
	if err != nil {
		return err
	}
	defer f.Close()

	err = png.Encode(f, pageRender.Image)
	if err != nil {
		return err
	}

	return nil
}

Experimental (CGO)

Some newer API's by PDFium are marked as experimental. We do have support for these functions, but because they are prone to change we do not compile with support for it by default. This is to keep support for most PDFium versions by default.

If you call any API methods marked as experimental, the call will result in an error

Adding support for the experimental API is quite easy. All you have to do is give the build tag pdfium_experimental when running go build or go run, like this: go run -tags pdfium_experimental {package}/{file.go} or go build -tags pdfium_experimental {package}/{file.go}.

We actively monitor PDFium API additions/changes/deletions and apply them in the code-base. When API methods become non-experimental, we will make them available in the default configuration.

WebAssembly

Recently we have added support for a non-cgo implementation using WebAssembly, we use the Wazero runtime for running WebAssembly within Go. The comes with quite some advantages:

Of course there are also some disadvantages:

Please be aware that Wazero comes with the Apache License 2.0 license.

Path handling (WebAssembly)

Because you can tell Wazero which folders have to be mounted in WebAssembly, you have full control over the filesystem.

By default, go-pdfium will mount the full root disk in Wazero on non-Windows environments. On Windows environments, go-pdfium will get the volume of the current working directory and mount that as the root.

You can change this behaviour by overwriting FSConfig in the pool setup.

All paths given to go-pdfium in WebAssembly mode have to be in POSIX style and have to be absolute, so for example: /home/user/Downloads/file.pdf. If you have mounted /home/user/on the root, then the path you would have to give is /Downloads/file.pdf, this is the same on Windows, so no backward slashes or volume names in paths.

You can set your own mounts by overwriting FSConfig in the pool setup.

Getting started (WebAssembly)

The examples below can also be found in the examples folder.

To start go-pdfium workers, you will have to init the go-pdfium worker pool somewhere, this also allows you to dynamically start workers when needed. The best location to add this is in the init() of a package that is going to call the PDFium library. Example:

pdfium/renderer/renderer.go

package renderer

import (
	"log"

	"github.com/klippa-app/go-pdfium"
	"github.com/klippa-app/go-pdfium/webassembly"
)

// Be sure to close pools/instances when you're done with them.
var pool pdfium.Pool
var instance pdfium.Pdfium

func init() {
	// Init the PDFium library and return the instance to open documents.
	// You can tweak these configs to your need. Be aware that workers can use quite some memory.
	pool, err = webassembly.Init(webassembly.Config{
		MinIdle:  1, // Makes sure that at least x workers are always available
		MaxIdle:  1, // Makes sure that at most x workers are ever available
		MaxTotal: 1, // Maxium amount of workers in total, allows the amount of workers to grow when needed, items between total max and idle max are automatically cleaned up, while idle workers are kept alive so they can be used directly.
	})

	var err error
	instance, err = pool.GetInstance(time.Second * 30)
	if err != nil {
		log.Fatal(err)
	}
}
Get page count
package renderer

import (
	"io/ioutil"
	"log"

	"github.com/klippa-app/go-pdfium"
	"github.com/klippa-app/go-pdfium/requests"
)

// Insert the webassembly init() here.

func main() {
	filePath := "example.pdf"
	pageCount, err := getPageCount(filePath)
	if err != nil {
		log.Fatal(err)
	}

	log.Printf("The PDF %s has %d pages", filePath, pageCount)
}

func getPageCount(filePath string) (int, error) {
	// Load the PDF file into a byte array.
	pdfBytes, err := ioutil.ReadFile(filePath)
	if err != nil {
		return 0, err
	}

	// Open the PDF using PDFium (and claim a worker)
	doc, err := instance.OpenDocument(&requests.OpenDocument{
		File: &pdfBytes,
	})
	if err != nil {
		return 0, err
	}

	// Always close the document, this will release its resources.
	defer instance.FPDF_CloseDocument(&requests.FPDF_CloseDocument{
		Document: doc.Document,
	})

	pageCount, err := instance.FPDF_GetPageCount(&requests.FPDF_GetPageCount{
		Document: doc.Document,
	})
	if err != nil {
		return 0, err
	}

	return pageCount.PageCount, nil
}

Render a page

package renderer

import (
	"image/png"
	"io/ioutil"
	"log"
	"os"

	"github.com/klippa-app/go-pdfium"
	"github.com/klippa-app/go-pdfium/requests"
)

// Insert the webassembly init() here.

func main() {
	filePath := "example.pdf"
	output := "example.pdf.png"
	err := renderPage(filePath, 1, output)
	if err != nil {
		log.Fatal(err)
	}
}

func renderPage(filePath string, page int, output string) error {
	// Load the PDF file into a byte array.
	pdfBytes, err := ioutil.ReadFile(filePath)
	if err != nil {
		return err
	}

	// Open the PDF using PDFium (and claim a worker)
	doc, err := instance.OpenDocument(&requests.OpenDocument{
		File: &pdfBytes,
	})
	if err != nil {
		return err
	}

	// Always close the document, this will release its resources.
	defer instance.FPDF_CloseDocument(&requests.FPDF_CloseDocument{
		Document: doc.Document,
	})

	// Render the page in DPI 200.
	pageRender, err := instance.RenderPageInDPI(&requests.RenderPageInDPI{
		DPI: 200, // The DPI to render the page in.
		Page: requests.Page{
			ByIndex: &requests.PageByIndex{
				Document: doc.Document,
				Index:    0,
			},
		}, // The page to render, 0-indexed.
	})
	if err != nil {
		return err
	}

	// The Render* methods return a cleanup function that has to be called when
	// using webassembly to make sure resources are cleaned up. Do this after
	// you are done with the returned image object.
	defer pageRender.Cleanup()

	// Write the output to a file.
	f, err := os.Create(output)
	if err != nil {
		return err
	}
	defer f.Close()

	err = png.Encode(f, pageRender.Result.Image)
	if err != nil {
		return err
	}

	return nil
}

Experimental (WebAssembly)

Some newer API's by PDFium are marked as experimental. The WebAssembly build has support for all of them.

We actively monitor PDFium API additions/changes/deletions and apply them in the code-base.

The WebAssembly build will always be the latest PDFium version that we added support for.

io.ReadSeeker and io.Writer

Document loading allows you to load a document with a io.ReadSeeker. Please be aware that this only works efficiently when using the single-threaded or WebAssembly usage, as that lives in the same process. For multi-threaded usage this will just load in the complete file and pass the bytes through the gRPC interface.

Document/image saving allows you to save using a io.Writer. Please be aware this only works when using the single-threaded or WebAssembly usage. It's not possible to encode the io.Writer with gRPC. Or share it between processes for that matter.

Improving JPEG rendering speed

By default, this library renders images with the image/jpeg package that comes with Go to make distribution as simple as possible. However, this package is quite slow compared to other native libraries like libjpeg and libjpeg-turbo, you can enable the usage of libjpeg-turbo by using the build tag pdfium_use_turbojpeg, this will require you to have the package libturbojpeg-dev installed during build time and the libturbojpeg package during runtime and build time.

Speed improvements that can be expected are significant, for example: on a simple PDF the full process of rendering a page is 3x as fast compared to a build without libjpeg-turbo.

This is supported in both the CGO and WebAssembly implementation, please note that the WebAssembly implementation also uses CGO for libjpeg-turbo for now. There are plans to compile libjpeg-turbo to WebAssembly for the WebAssembly implementation to keep the WebAssembly implementation actually full WebAssembly.

Support Policy

We offer an API stability promise with semantic versioning. In other words, we promise to not break any exported function signature without incrementing the major version. New features and behaviors happen with a minor version increment, e.g. 1.0.11 to 1.1.0. We also fix bugs or change internal details with a patch version, e.g. 1.0.0 to 1.0.1. Upgrades of the supported PDFium version will cause a minor version update.

Go

This project will support the last 3 version of Go, this means that if the last version of Go is 1.21, our go.mod will be set to Go 1.19, and our CI tests will be run on Go 1.19, 1.20 and 1.21. It won't mean that the library won't work with older versions of Go, but it will tell you what to expect of the supported Go versions. If we change the supported Go versions, we will make that a minor version upgrade. This policy allows you to not be forced to the latest Go version for a pretty long time, but it still allows us to use new language features in a pretty reasonable time-frame.

About Klippa

Founded in 2015, Klippa's goal is to digitize & automate administrative processes with modern technologies. We help clients enhance the effectiveness of their organization by using machine learning and OCR. Since 2015, more than a thousand happy clients have used Klippa's software solutions. Klippa currently has an international team of 50 people, with offices in Groningen, Amsterdam and Brasov.

License

The MIT License (MIT)