MDN/Development/Emscripten: Difference between revisions

From MozillaWiki
Jump to navigation Jump to search
No edit summary
No edit summary
Line 18: Line 18:
==What content should it have?==
==What content should it have?==


My experience and perspective is mainly from the "Game developer who has written a native game in C/C++ for Android/iOS/PC and wants to port it to FxOS/browser" viewpoint. There, the process comes with the following questions, roughly in the following chronological learning order:
Although Emscripten can be used to compile a number of different languages into JavaScript, it would make sense to start with C++, as that is the best supported and most useful target laguage for our current projects. The following quote from Jukka provides some good context for the purpose of these docs:
 
<blockquote>My experience and perspective is mainly from the "Game developer who has written a native game in C/C++ for Android/iOS/PC and wants to port it to FxOS/browser" viewpoint. There, the process comes with the following questions, roughly in the following chronological learning order:</blockquote>


(feasibility/disbelief) 1. Web doesn't do 3D well and I know web is slow. Is this port even feasible? What kind of sacrifices will I have to do in my existing game in terms of performance? How does web compare to native Android/iOS/PC?
(feasibility/disbelief) 1. Web doesn't do 3D well and I know web is slow. Is this port even feasible? What kind of sacrifices will I have to do in my existing game in terms of performance? How does web compare to native Android/iOS/PC?
Line 61: Line 63:


Currently we do have some examples of "how do I do X in Emscripten" as part of the unit test suite, but naturally that's not a good format for reading as a documentation/learning material. Some of those could be hoisted or used as a basis for a better set of example apps.
Currently we do have some examples of "how do I do X in Emscripten" as part of the unit test suite, but naturally that's not a good format for reading as a documentation/learning material. Some of those could be hoisted or used as a basis for a better set of example apps.
Best Regards,
    Jukka


==Raw ideas==
==Raw ideas==


==What should we talk to?==
==What should we talk to?==

Revision as of 11:24, 25 November 2013

Project statement

Emscripten is an LLVM to JavaScript compiler. It takes LLVM bytecode (which can be generated from C/C++ using Clang, or any other language that can be converted into LLVM bytecode) and compile that into JavaScript, which can be run on the web (or anywhere else JavaScript can run). Using Emscripten, you can

  • Compile C and C++ code into JavaScript and run that on the web
  • Run code originated in languages like Python on the web as well, for example by compiling CPython from C to JavaScript

This is a really cool and very useful technology, and it is essential for the success of some of Mozilla's current initiatives, such as web gaming. But the uses of Emscripten go beyond just building amazing games, therefore I have decided that Emscripten should have its own landing page on MDN, covering Emscripten in a more generic fashion.

Emscripten currently has a wiki on github that contains some great content, but this content could use some updating and reorganization, and we have lots of other ideas for content too, to tell the full story of how useful Emscripten is to interested parties such as C++ developers.

This document outlines the plans for an MDN resource covering Emcripten.

Where will this project live?

The obvious place would be a landing page at https://developer.mozilla.org/en-US/docs/Emscripten.

What content should it have?

Although Emscripten can be used to compile a number of different languages into JavaScript, it would make sense to start with C++, as that is the best supported and most useful target laguage for our current projects. The following quote from Jukka provides some good context for the purpose of these docs:

My experience and perspective is mainly from the "Game developer who has written a native game in C/C++ for Android/iOS/PC and wants to port it to FxOS/browser" viewpoint. There, the process comes with the following questions, roughly in the following chronological learning order:

(feasibility/disbelief) 1. Web doesn't do 3D well and I know web is slow. Is this port even feasible? What kind of sacrifices will I have to do in my existing game in terms of performance? How does web compare to native Android/iOS/PC?

(feasibility/disbelief) 2. Which browsers can I target with emscripten? ff/chrome/ie/opera/safari? desktop and mobile?

(feasibility research) 3. My game uses C/C++ libraries/technologies X, Y and Z. Are these supported? What do I need to rewrite?

(feasibility research) 4. I am looking an alternative to library X, which I know isn't supported. Which C/C++ technologies can I use instead? Can C/C++ do everything, or do I need to write JS to implement X/Y/Z?

(adoption,familiarization) 5. How do I get hands-on experience with the toolchain? Where is that minimal GL rendering sample that shows how to get stuff up and running?

(r&d port attempt) 6. How do I work my existing build system X to target emscripten? How do I co-maintain both native and emscripten targets simultaneously?

(r&d port attempt) 7. How do I deploy and run my app on a mobile device?

(r&d port attempt) 8. How do I do debugging?

(r&d port attempt) 9. Where is the compiler and toolchain documentation? Where is the graphics/WebGL/GLES2 info I need to know?

(disbelief/solving blockers) 10. I have learned about these hard problems (non-threading, async main loop, async file loading, memory model, networking, performance, ...) that I am now able to relate to my own codebase having attempted a port. What can I do? How do I architect around these problems?

(port development) 11. Where is the reference documentation for C/C++ library/api X in Emscripten?

(port development) 12. Where is the reference documentation for JS library/api/spec X?

(port development) 13. How do I develop a JS library to solve X and glue it to my C/C++ game?

(port development) 14. How do I manage FxOS app packaging and mobile runtime?

(finalization) 15. How do I do benchmarking, profiling and optimizations? How to minimize data sizes and optimize web transfers?

(finalization) 16. How do I interact/glue the C/C++ game experience with the browser environment? Web site development around the game shell page, etc?

Generally the devs are initially not familiar with the JS specs, and if they are presented e.g. with the Gamepad API or Web Audio API, they'll reject those docs since they are foremost looking for C/C++ solutions to their problems. This is partly because they have just learned about this Emscripten C/C++ compiler that they are now harnessing, and they are right now getting their initial problems solved in the C/C++ side without a single line of JS. Another reason is that devs might not know the JS syntax at all, or if they do, there's a considerable amount of new C/C++ <-> JS interop machinery that one needs to learn, and how to integrate JS into the native build arch, and so on, so being able to develop against JS side APIs will generally come much later during the porting cycle, when the landscape has been charted sufficiently enough to know that one can't (yet) solve all the problems with just native C/C++ code alone.

I think an important angle to cover, that we currently have a big glaring hole at in the current Emscripten docs at https://github.com/kripken/emscripten/wiki#technical-documentation is the answers to questions #3, #4 and #11. We should have strong documentation pages on each individual API/library we support directly in Emscripten, since a huge part of the Emscripten work (and one that stands us out most compared to nacl!) goes into making sure that existing C/C++ technologies like SDL, OpenAL, EGL, GLES2, etc. are directly supported. For each of these apis that are there, we should have a page "Support for X in Emscripten", similar to we now have for EGL https://github.com/kripken/emscripten/wiki/EGL-Support-in-Emscripten , since that will help people at the initial feasibility study phase and to deciding positively on whether they will take on the effort to try a port.

A second set of docs we should have more flesh in, is the set of emscripten-specific APIs that we have been developing. We don't have much of those, but generally this header file is the only doc available https://github.com/kripken/emscripten/blob/master/system/include/emscripten/emscripten.h .

Third, the set of big problems that everyone porting a native game will eventually run into (item #10 in the list above). Until we have good fixes on those, we should have good prose-like technical documentation that describes the problems with examples of what will not work, and how to refactor the code so that it is JS-friendly, etc.

Currently we do have some examples of "how do I do X in Emscripten" as part of the unit test suite, but naturally that's not a good format for reading as a documentation/learning material. Some of those could be hoisted or used as a basis for a better set of example apps.

Raw ideas

What should we talk to?