Regression in modularize mode (currentSrc issue)

[bvibber]

Compiling ogv.js with emscripten sdk 1.38.10, I'm seeing a new failure mode with my demuxer module, which is loaded asynchronously into the web page context:

[Error] TypeError: null is not an object (evaluating 'ca.src')

This is in the (minified) setup code:

        if (n) {
            var da = this._currentScript || document.currentScript;
            0 !== da.src.indexOf("blob:") && (t = da.src.split("/").slice(0, -1).join("/") +
            "/")
        } else

source:

   if (ENVIRONMENT_IS_WEB) {
#if MODULARIZE
    // When MODULARIZE, this JS may be executed later, after document.currentScript is gone, so we send it
    // using this._currentScript.
#else
    var currentScript = document.currentScript;
#endif
     if (currentScript.src.indexOf('blob:') !== 0) {
       scriptDirectory = currentScript.src.split('/').slice(0, -1).join('/') + '/';
     }

Something about this isn't doing what I expect...

I notice at the end of the .js file is this which appears to be an attempt to capture the currentSource:

OGVDemuxerWebMW = OGVDemuxerWebMW.bind({
    _currentScript: typeof document !== 'undefined' ? document.currentScript : undefined
});

But that syntax for the Function.prototype.bind function doesn't appear to be documented on MDN, and it doesn't seem to do anything that I can tell on Safari 11.1.2.

[bvibber]

That bit appears to have come from d1f303a

[kripken]

Odd, isn't that bind usage basically the simple case shown here:

https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Function/bind

(of binding an object which is then this in the call)? Does that MDN example not work in Safari?

[kripken]

If that's not it, is it possible that the line with OGVDemuxerWebMW = OGVDemuxerWebMW.bind({ .. }) is not being called, somehow? If so, and it's calling the unbound version, then it would see a null there, which would seem to fit the symptoms (but I have no idea why that line wouldn't be reached, very odd).

[nazar-pc]

I suspect it is called, but asynchronously as part of a bigger bundle. You may want to adjust your build process or override Module.locateFile. On Emscripten side we should probably add a check whether currentScript is an object at all to avoid this error.

[kripken]

Hmm, we do the bind right after defining the function, so it seems like that has to happen.

And we do check if this['_currentScript'] exists, so I think that should handle the case of not having document or document.currentScript defined during binding.

Maybe another binding happens later, overriding this one?

[kripken]

Btw, this code slightly changed since the original landing - @Brion is this still an issue on latest incoming?

[bvibber]

(document and document.currentScript are defined at the time that bind() is called, I checked in the debugger)

[bvibber]

Ah, I understand what the bind() call is doing now -- it's putting the _currentScript property on the hidden this, not on the Module object... investigating further on my end.

[bvibber]

AHA!

My call to the OGVDemuerWebMW module function was made like this:

return new global[className](options);

which used to work fine. That "new" replaces the "this" parameter from the bind() call with a copy of the prototype, which has no _currentScript property.

Removing the "new" from the calls should get it working... recompiling now. :)

[bvibber]

Aaaaand it works now. :) As far as I know following documentation correctly in the initial call would have saved me here. :D

[kripken]

Great, thanks for the update @Brion!

I added a commit to #6894 to avoid crashing in either the case of document.scriptDirectory not existing (as @nazar-pc mentioned), or doing new Module() which overrides the object containing the helper property. In both cases we can't access the scriptDirectory but we should at least not crash.