JavaScript: Improve detection of UMD modules.

[max-schaefer]

Specifically, we now detect this pattern:

(function (global, factory) {
    checkForNodeJS() ? module.exports = factory(...) :
    checkForAmd() ? define(..., factory) :
    global.mymodule = factory(...)
}(this, function (...) { ... });

We previously missed it because we expected the define to appear as an expression statement, but here it is nested a bit more deeply.

I fixed this by using the inVoidContext predicate we use in a few other queries, which however needed a bit of extra work. I'm curious to see whether that has any knock-on effects on performance or results, if it does I'm happy to switch to a more targeted fix.

[erik-krogh]

LGTM assuming an evaluation works out.
(Let me know if you want me to do one for you).

[max-schaefer]

Let me know if you want me to do one for you

That would be fantastic, thanks! (The language tests pass.)

[erik-krogh]

Let me know if you want me to do one for you

That would be fantastic, thanks! (The language tests pass.)

I've started one. It should appear as a backref shortly.

[max-schaefer]

The evaluation revealed significant performance regressions that need looking into.

[max-schaefer]

I took a quick look at the regression, but didn't make much progress. It's easy to reproduce by, for example, running ClientSideUrlRedirect.ql on openlayers/ol2, which goes from 2s to 105s. We seem to be evaluating a whole load of predicates we previously didn't evaluate (or perhaps they were very quick, I don't know how to read logs anymore these days). I can't justify spending more time on this right now, so I'll just leave this open as a draft in case anyone else wants to pick it up later.

[erik-krogh]

I've looked a bit at this, and I think I we could merge it now.

New evaluation.

Some of the performance regression seem to have disappeared from just waiting (core improvements?).
But there is still a performance regression left.

However, the cause seems to be that we recognize more flowsteps.
There are several new results, and all the new results (I sampled a few) seem to be from us having a more complete callgraph, which is caused from recognizing more UMD modules.
Some of those new results are eventually FPs (e.g. flow through _.escape(..) in adobe/brackets), but the new edge in the callgraph is good.

I think the trade-off is worth it.

[max-schaefer]

I think we could merge it now

Ah, self-fixing performance problems! Those are the best.

[max-schaefer]

@erik-krogh, could you re-approve so I can auto-merge?