Frequently Asked Questions
- Why did Google build Native Client?
- When should I use Portable Native Client instead of Native Client?
- When should I use Portable Native Client / Native Client?
- How fast does code run in Portable Native Client?
- Why use Portable Native Client instead of <technology X>?
- If I want direct access to the OS, should I use Native Client?
- Is Native Client open? Is it a standard?
- How can I contribute to Native Client?
- What are the supported instruction set architectures?
- Do I have to use C or C++? I’d really like to use another language.
- Do you only support Chrome? What about other browsers?
- What’s the difference between NPAPI and Pepper?
- Is NPAPI part of the Native Client SDK?
- Does Native Client support SIMD vector instructions?
- Can I use Native Client for 3D graphics?
- Does Native Client support concurrency/parallelism?
- How do I debug?
- How do I build x86-32, x86-64 or ARM
- How can my web application determine which
- Is it possible to build a Native Client module with just plain C (not C++)?
- What UNIX system calls can I make through Native Client?
- Is my favorite third-party library available for Native Client?
- Do all the files in an application need to be served from the same domain?
This document answers some frequently asked questions about Native Client (NaCl) and Portable Native Client (PNaCl, pronounced “pinnacle”). For a high-level overview of Native Client, see the Technical Overview.
If you have questions that aren’t covered in this FAQ:
What is Native Client Good For?
Why did Google build Native Client?
- Performance: Native Client modules run nearly as fast as native compiled code.
- Security: Native Client lets users run native compiled code in the browser with the same level of security and privacy as traditional web applications.
- Developers can leverage existing code, written in C/C++ or other languages, in their applications without forcing users to install a plugin.
Portability: Native Client and Portable Native Client applications can execute on:
- The Windows, Mac, Linux or ChromeOS operating systems.
- Processors with the x86-32, x86-64, or ARM instruction set architectures. Native Client also has experimental support for MIPS.
Portable Native client further enhances the above:
- Performance: Each PNaCl release brings with it more performance enhancements. Already-released applications get faster over time, conserving user’s battery.
- Security: Users are kept secure with an ever-improving sandbox model which adapts to novel attacks, without affecting already-released applications.
- Convenience: Developers only need to ship a single
.pexefile, not one
.nexefile per supported architecture.
- Portability: Developers and users don’t need to worry about already-released applications not working on new hardware: PNaCl already supports all architectures NaCl does, and as PNaCl evolves it gains support for new processors and fully uses their capabilities.
For more details, refer to the history behind and comparison of NaCl and PNaCl.
When should I use Portable Native Client instead of Native Client?
See NaCl and PNaCl. In short: PNaCl works on the Open Web platform delivered by Chrome whereas NaCl only works on the Chrome Web Store.
When should I use Portable Native Client / Native Client?
The following are some typical use cases. For details, see the Technical Overview.
- Porting existing applications or software components, written in C/C++ or virtual machines written in C/C++, for use in a web application.
Using compute-intensive applications, including threads and SIMD, such as:
- Scientific computing.
- Handling multimedia for a web application.
- Various aspects of web-based games, including physics engines and AI.
- Running untrusted code on a server or within an application (such as a plugin system for a game).
Portable Native Client and Native Client are versatile technologies which are used in many other contexts outside of Chrome.
How fast does code run in Portable Native Client?
Fast! The SPEC2k benchmarks (C, C++ and floating-point benchmarks) give the following overhead for optimized PNaCl compared to regular optimized LLVM:
Note that benchmark performance is sometimes bimodal, so different use cases are likely to achieve better or worse performance than the above averages. For example floating-point heavy code usually exhibits much lower overheads whereas very branch-heavy code often performs worse.
For details, see:
- PNaCl SIMD: Speed on the Web.
- Adapting Software Fault Isolation to Contemporary CPU Architectures (PDF).
- Native Client: A Sandbox for Portable, Untrusted x86 Code (PDF).
If your code isn’t performing as close to native speed as you’d expect, let us know!
Why use Portable Native Client instead of <technology X>?
Many other technologies can be compared to Portable Native Client: Flash, Java, Silverlight, ActiveX, .NET, asm.js, etc...
Different technologies have different strengths and weaknesses. In appropriate contexts, Portable Native Client can be faster, more secure, and/or more compatible across operating systems and architectures than other technologies.
Portable Native Client complement other technologies by giving web developers a new capability: the ability to run fast, secure native code from a web browser in an architecture-independent way.
If I want direct access to the OS, should I use Native Client?
Development Environments and Tools
What development environment and development operating system do you recommend?
You can develop on Windows, Mac, or Linux, and the resulting Native Client or Portable Native Client application will run inside the Google Chrome browser on all those platforms as well as ChromeOS. You can also develop on ChromeOS with Crouton or our experimental development environment which runs within NaCl, and we’re working on self-hosting a full development environment on Portable Native Client.
Any editor+shell combination should work as well as IDEs like Eclipse, Visual Studio with the Native Client Add-In on Windows, or Xcode on Mac OSX.
I’m not familiar with native development tools, can I still use the Native Client SDK?
You’ll need to learn how to use some tools (like GCC, LLVM, make, Eclipse, Visual Studio, or Xcode) before you can get very far with the SDK. Try seaching for an introduction to GCC.
Openness, and Supported Architectures and Languages
Is Native Client open? Is it a standard?
Native Client is completely open: the executable format is open and the source code is open. Right now the Native Client project is in its early stages, so it’s premature to consider Native Client for standardization.
We consistenly try to document our design and implementation and hope to standardize Portable Native Client when it gains more traction. A good example is our PNaCl bitcode reference manual.
How can I contribute to Native Client?
Read about contributor ideas.
What are the supported instruction set architectures?
Portable Native Client uses an architecture-independent format (the
.pexe) which can currently be translated to execute on processors with the x86-32, x86-64, and ARM instruction set architectures, as well as experimental support for MIPS. As new architectures come along and become popular we expect Portable Native Client to support them without developers having to recompile their code.
Native Client can currently execute on the same architectures as Portable Native Client but is only supported on the Chrome Web Store. Native Client’s
.nexe files are architecture-dependent and cannot adapt to new architectures without recompilation, we therefore deem them better suited to a web store than to the open web.
Do I have to use C or C++? I’d really like to use another language.
Do you only support Chrome? What about other browsers?
We aim to support multiple browsers. However, a number of features that we consider requirements for a production-quality system that keeps the user safe are difficult to implement without help from the browser. Specific examples are an out-of-process plugin architecture and appropriate interfaces for integrated 3D graphics. We have worked closely with Chromium developers to deliver these features and we are eager to collaborate with developers from other browsers.
What’s the difference between NPAPI and Pepper?
Pepper (also known as PPAPI) is a new API that lets Native Client modules communicate with the browser. Pepper supports various features that don’t have robust support in NPAPI, such as event handling, out-of-process plugins, and asynchronous interfaces. Native Client has transitioned from using NPAPI to using Pepper.
Is NPAPI part of the Native Client SDK?
NPAPI is not supported by the Native Client SDK, and is deprecated in Chrome.
Does Native Client support SIMD vector instructions?
Portable Native Client supports portable SIMD vectors, as detailed in Portable SIMD Vectors.
Native Client supports SSE, AVX1, FMA3 and AVX2 (except for VGATHER) on x86 and NEON on ARM.
Can I use Native Client for 3D graphics?
Yes. Native Client supports OpenGL ES 2.0.
Some GL extensions are exposed to Native Client applications, see the GLES2 file. This file is part of the GL wrapper supplied by the library
ppapi_gles2 which you’ll want to include in your project. In most cases extensions map to extensions available on other platforms, or differ very slightly (if they differ, the extension is usually CHROMIUM or ANGLE instead of EXT).
Does Native Client support concurrency/parallelism?
Native Client and Portable Native Client both support pthreads, C11/C++11 threads, and low-level synchronization primitives (mutex, barriers, atomic read/modify/write, compare-and-exchange, etc...), thus allowing your Native Client application to utilize several CPU cores. Note that this allows you to modify datastructures concurrently without needing to copy them, which is often a limitation of shared-nothing systems. For more information see memory model and atomics and threading.
Do Native Client modules have access to external devices?
At this time Native Client modules do not have access to serial ports, camera devices, or microphones: Native Client can only use native resources that today’s browsers can access. However, we intend to recommend such features to the standards bodies and piggyback on their efforts to make these resources available inside the browser.
Security and Privacy
What happens to my data when I use Native Client?
Users can opt-in to sending usage statistics and crash information in Chrome, which includes usage statistics and crash information about Native Client. Crashes in your code won’t otherwise send your information to Google: Google counts the number of such crashes, but does so anonymously without sending your application’s data or its debug information.
How does Native Client prevent sandboxed code from doing Bad Things?
Native Client’s sandbox works by validating the untrusted code (the compiled Native Client module) before running it. The validator checks the following:
- Data integrity: No loads or stores are permitted outside of the data sandbox. In particular this means that once loaded into memory, the binary is not writable. This is enforced by operating system protection mechanisms. While new instructions can be inserted at runtime to support things like JIT compilers, such instructions will be subject to runtime verification according to the following constraints before they are executed.
- No unsafe instructions: The validator ensures that the Native Client application does not contain any unsafe instructions. Examples of unsafe instructions are
- Control flow integrity: The validator ensures that all direct and indirect branches target a safe instruction.
The beauty of the Native Client sandbox is in reducing “safe” code to a few simple rules that can be verified by a small trusted validator: the compiler isn’t trusted. The same applies to Portable Native Client where even the
.nexe translator, a simplified compiler backend, isn’t trusted: it is validated before executing, and so is its output.
In addition to static analysis of untrusted code, the Native Client runtime also includes an outer sandbox that mediates system calls. For more details about both sandboxes, see Native Client: A Sandbox for Portable, Untrusted x86 Code (PDF).
How does Google know that the safety measures in Native Client are sufficient?
Google has taken several steps to ensure that Native Client’s security works, including:
- Open source, peer-reviewed papers describing the design.
- A security contest.
- Multiple internal and external security reviews.
- The ongoing vigilance of our engineering and developer community.
How do I debug?
How do I build x86-32, x86-64 or ARM
By default, the applications in the
/examples folder create architecture-independent
.pexe for Portable Native Client. To generate a
.nexe targeting one specific architecture using the Native Client or Portable Native Client toolchains, see the Building instructions.
How can my web application determine which
.nexe to load?
Your application does not need to make the decision of loading an x86-32, x86-64 or ARM
.nexe explicitly—the Native Client runtime examines a manifest file (
.nmf) to pick the right
.nexe file for a given user. You can generate a manifest file using a Python script that’s included in the SDK (see the
Makefile in any of the SDK examples for an illustration of how to do so). Your HTML file specifies the manifest filename in the
src attribute of the
<embed> tag. You can see the way the pieces fit together by examining the examples included in the SDK.
Is it possible to build a Native Client module with just plain C (not C++)?
Yes. See the
"Hello, World!" in C example in the SDK under
examples/tutorial/using_ppapi_simple/, or the Game of Life example under
What UNIX system calls can I make through Native Client?
Native Client doesn’t directly expose any system calls from the host OS because of the inherent security risks and because the resulting application would not be portable across operating systems. Instead, Native Client provides portable cross-OS abstractions wrapping or proxying OS functionality or emulating UNIX system calls. For example, Native Client provides an
mmap() system call that behaves much like the standard UNIX
mmap() system call.
Is my favorite third-party library available for Native Client?
Google has ported several third-party libraries to Native Client; such libraries are available in the webports project. We encourage you to contribute libraries to webports, and/or to host your own ported libraries, and to let the team know about it on native-client-discuss when you do. You can also read through contributor ideas to find ideas of new projects to port.
Do all the files in an application need to be served from the same domain?
.pexe files must either be served from the same origin as the embedding page or an origin that has been configured correctly using CORS.
For applications installed from the Chrome Web Store the Web Store manifest must include the correct, verified domain of the embedding page.
Do I have to do anything special to make my application run on different operating systems?
No. Native Client and Portable Native Client applications run without modification on all supported operating systems.
However, to run on different instruction set architectures (such as x86-32, x86-64 or ARM), you currently have to either:
- Use Portable Native Client.
- Build and supply a separate
.nexefile for each architecture, and make them available on the Chrome Web Store. See target architectures for details about which
.nexefiles will run on which architectures.
How easy is it to port my existing native code to Native Client?
In most cases you won’t have to rewrite much, if any, code. The Native Client-specific tools, such as
x86_64-nacl-g++, take care of most of the necessary changes. You may need to make some changes to your operating system calls and interactions with external devices to work with the web. Porting existing Linux libraries is generally straightforward, with large libraries often requiring no source change.
The following kinds of code may be more challenging to port:
- Code that does direct TCP or UDP networking. For security reasons these APIs are only available to Chrome apps after asking for the appropriate permissions, not on the open web. Native Client is otherwise restricted to the networking APIs available in the browser. You may want to use to nacl_io library to use POSIX-like sockets.
- Code that creates processes, including UNIX
fork, won’t function as-is. However, threads are supported. You can nonetheless create new
<embed>tags in your HTML page to launch new PNaCl processes. You can even use new
.pexefiles that your existing
.pexesaved in a local filesystem. This is somewhat akin to
execve, but the process management has to go through
write/ ...). See Using NaCl I/O for more details.
.pexe isn’t loading, help!
- You must use Google Chrome version 31 or greater for Portable Native Client. Find your version of chrome by opening
about:chrome, and update Chrome if you are on an older version. If you’re already using a recent version, open
about:componentsand “Check for update” for PNaCl. Note that on ChromeOS PNaCl is always up to date, whereas on other operating systems it updates shortly after Chrome updates.
- A PNaCl
.pexemust be compiled with pepper_31 SDK or higher. Update your bundles and make sure you’re using a version of Chrome that matches the SDK version.
navigator.mimeTypes['application/x-pnacl'] !== undefined. This is preferred over checking the Chrome version.
.nexe files never finish loading. What gives?
Here are ways to resolve some common problems that can prevent loading:
- You must use Google Chrome version 14 or greater for Native Client.
- If you haven’t already done so, enable the Native Client flag in Google Chrome. Type
about:flagsin the Chrome address bar, scroll down to “Native Client”, click the “Enable” link, scroll down to the bottom of the page, and click the “Relaunch Now” button (all browser windows will restart).
- Verify that the Native Client plugin is enabled in Google Chrome. Type
about:pluginsin the Chrome address bar, scroll down to “Native Client”, and click the “Enable” link. (You do not need to relaunch Chrome after you enable the Native Client plugin).
- Make sure that the
.nexefiles are being served from a web server. Native Client uses the same-origin security policy, which means that modules will not load in pages opened with the
file://protocol. In particular, you can’t run the examples in the SDK by simply dragging the HTML files from the desktop into the browser. See Running Native Client Applications for instructions on how to run the httpd.py mini-server included in the SDK.
.nexefiles must have been compiled using SDK version 0.5 or greater.
- You must load the correct
.nexefile for your machine’s specific instruction set architecture (x86-32, x86-64 or ARM). You can ensure you’re loading the correct
.nexefile by building a separate
.nexefor each architecture, and using a
.nmfmanifest file to let the browser select the correct
.nexefile. Note: the need to select a processor-specific
.nexegoes away with Portable Native Client.
- If things still aren’t working, ask for help!