Enumerated types are very hard to bind to other languages (especially Rust), let's avoid them in that very important public API.

Comment on attachment 8748602 [details] [diff] [review] 0001-Do-not-use-enumerated-types-to-debug-CallArgs.patch Review of attachment 8748602 [details] [diff] [review]: ----------------------------------------------------------------- ::: js/public/CallArgs.h @@ +123,4 @@ > mutable bool usedRval_; > void setUsedRval() const { usedRval_ = true; } > void clearUsedRval() const { usedRval_ = false; } > +#endif I understand that the symbol are loaded from the parent class when we build with an optimized build. Still, I would prefer to have a #else which redefine the empty function in such case. Changing the lookup is a bad habit, and C++ has some corner case around looking up symbols. This is not the case, but I do not want such thing to become usual. @@ +236,3 @@ > private: > + friend CallReceiverBase<IncludeUsedRval> JS::CallReceiverFromVp(Value *vp); > + friend CallReceiverBase<IncludeUsedRval> JS::CallReceiverFromArgv(Value *argv); From what you told me you are going to investigate a similar issue in mfbt (why clang fails to bind this name to the name of the class), if you find a work-around in clang, we should not change these lines either.

Addressed :nbp's remarks.

nox, what exactly is hard to bind about this case? Is it that there's a template parameter that's an enum (I know Rust doesn't support type-parameters that are values), or something about the enum itself? Also: what's going on? I don't know if we've ever talked. How are these bindings being generated, what are they for, and how can we help?

FYI, a fresh laptop arrived just today \o/ so I probably should be able to get to this tomorrow.

I bitrotted the original pretty good doing simplifications in this area, so I should probably redo it for politeness. :-) I think the original had some extra unnecessary complexity to it, as well, that this version was able to remove. Win!

Pushed by jwalden@mit.edu: https://hg.mozilla.org/integration/mozilla-inbound/rev/e74111fa444f Make CallArgs's JS_DEBUG-only handling of asserting proper rval()/calleev() sequencing work without using enumerated types, for Servo bindings. r=nox/r=jwalden tag-team effort