Build CsoundAC for WebAssembly

[gogins]

Currently there are two methods of generating bindings automatically, WebIDL and Embind. Calling JavaScript from C++ is supported by Embind, and this seems to provide somewhat more functionality. I also have already used Embind for Csound itself and it works just fine.

For now, I will proceed on the assumption of using Embind. However, there are issues.

The major issue is the Eigen matrix class used by CsoundAC. This class in itself is just plain wonderful, but it has a complex interface and appears everywhere in the CsoundAC interface.

But the payoff here is huge, once done I can do all new CsoundAC code in C++ and use JavaScript only for pieces.

[gogins]

To do:

• [x] Identify Eigen calls made from CsoundAC. So far this does not seem to be needed.
• [x] Investigate modularizing both csound-extended and CsoundAC, or incorporating CsoundAC in csound-extended. It will be in csound-extended.

[gogins]

Actually what I really want is to use CsoundAC instead of the Silencio JavaScript stuff on Android. I don't care that much how it happens but I need to work on the same pieces in both the Android and the desktop environments. Take a look at https://github.com/kivy/python-for-android and see if that helps.

[gogins]

It's not clear how to configure Eigen3 for Emscripten. It was used in ammo.js by @kripken but has been replaced by custom code. It may be necessary to link or copy the Eigen directory to Emscripten's include directory. Then too I will have to rewrite or conditionally omit code that uses FLTK and perhaps some other libraries.

[gogins]

I created this link:

mkg@xenakis:~/emsdk/upstream/emscripten/system/include$ ls -ll
total 132
drwxr-xr-x 2 mkg mkg  4096 Apr 16 17:48 AL
drwxr-xr-x 3 mkg mkg  4096 Apr 16 17:48 compat
drwxr-xr-x 2 mkg mkg  4096 Apr 16 17:48 EGL
lrwxrwxrwx 1 mkg mkg    19 Jun  1 09:32 eigen3 -> /usr/include/eigen3

[gogins]

Creating the link in ~/emsdk/upstream/emscripten/system/include/libcxx did the job, so now libcsoundac-static.a compiles. Getting this to build and link with the other stuff is now the job. Also it is not clear how much work this actually will turn out to be.

For the SWIG-generated Python interface, a good deal of code wrapping pointers and references to Eigen matrices is generated. I do not need the Eigen API but I do need Eigen objects.

But probably I don't actually need the Eigen objects. I don't need them in Python although they have SWIG proxies generated for them more or less by default.

[gogins]

I have wrongly put CsoundAC in the CsoundAudioProcessor, it needs to go in the CsoundAudioNode. But I am not sure how this works out with the MusicModel class embedding Csound. I may have to change that.

I think this might be a big problem. Theoretically I can push it all down into the CsoundAudioProcessor but the amount of messaging clue code becomes nightmarish.

[gogins]

I have refactored csound::System to deal with the message printing code in a more cross-platform way. Things are now building, and I can proceed to actually wrapping stuff.

[gogins]

I'm running into problems with Embind. Specifically, I have properly defined CellAdd as having Cell as a base class but Embind is not seeing that.

Somehow I missed this which I will now evaluate: https://github.com/charto/nbind. Oops, looks like nbind is declining and Embind is keeping on going.

[gogins]

Closed by mistake...

Disabling RTTI is no help, RTTI is needed.

Going to try a test case of an inheritance chain. A chain of two derivations works fine. Try defining them out of order. Still works.

[gogins]

Oops, it's just I got the base class of CellAdd wrong. It's Node, not Cell.

[gogins]

I may now be running into this in Scale::modulations which is one of the coolest things in CsoundAC failing to compile: https://github.com/emscripten-core/emscripten/issues/9005. Or this: https://github.com/emscripten-core/emscripten/issues/9485.

[gogins]

Trying to build with C++ 17 now. No help.

[gogins]

Got rid of the class method overloads and it works. I just spell out different method names. I will try to get rid of all the overloads in the Embind definitions for CsoundAC.

[gogins]

I have substantially completed the Embind wrappers for CsoundAC. I had to change a number of things in the CsoundAC C++ source code to simplify this job. To do:

• [x] Test a C++ composition that uses CsoundAC to see if it still works as it used to. This is Parachronic_v2-issue140.cpp. Sounds the same, sonogram looks the same, OK.
• [x] Run the ChordSpace test suite carefully. Still OK as far as I can tell. May need more testing.
• [x] Try to rewrite a C++ style piece with a music graph, etc. in JavaScript. Proof of concept completed and runs.
• [x] Write something to produce the 3-D piano roll score display in the browser from a CsoundAC Score. For now the existing JavaScript code in Silencio will be adapted. But in future this should be done in C++ using OpenGL ES 2.0. Wrote a JavaScript PIanoRoll helper class.
• [x] Port ScoreGraphs.js to CsoundAC. Not doing this now.
• [x] Port ParametricLindenmayerSystem4.js to CsoundAC. Not doing this now.
• [x] Create an issue to package this in the Csound for Android app.

[gogins]

Can't load CsoundAC.js in NW.js due to __dirname being referenced in generated code. See https://github.com/emscripten-core/emscripten/issues/10381.

But here's a workaround, setting that variable before loading CsoundAC, in the body of the page:

<body>
<script>
try {
    var fs = require("fs");
    var __dirname = fs.realpathSync.native(".");
} catch(e) {
    console.log(e);
}
</script>
<script src="CsoundAC.js"></script>
<script src="CsoundAudioNode.js"></script>
<script>
let CsoundAC;
(async function() { CsoundAC =  await createCsoundAC(); CsoundAC.hello(); }());
</script>

This doesn't affect the loading of CsoundAC.js in the browser.

[gogins]

This workaround seems to have quit working.

[gogins]

No, the workaround is still working.