The Makefile.in snippet is: ifeq (,$(filter beta release esr,$(MOZ_UPDATE_CHANNEL))) DEFINES += -DENABLE_TYPEDARRAY_MOVE endif ifeq (,$(filter aurora beta release esr,$(MOZ_UPDATE_CHANNEL))) DEFINES += -DENABLE_PARALLEL_JS endif MOZ_UPDATE_CHANNEL is usually default for non-official builds of any branch, which means both are enabled on such builds. This affects release, where the snippet is ifeq (,$(filter beta release esr,$(MOZ_UPDATE_CHANNEL))) DEFINES += -DENABLE_TYPEDARRAY_MOVE endif and beta, where it is ifeq (,$(filter beta release esr,$(MOZ_UPDATE_CHANNEL))) DEFINES += -DENABLE_TYPEDARRAY_MOVE DEFINES += -DENABLE_PARALLEL_JS endif On other branches, it's supposed to be enabled, so it's not really a problem.

Isn't it more serious than that, since on Aurora we set the channel on nightly builds and don't on on-push builds, so that sets us up for nightly-only failures, and on beta and release we set the channel on release builds, but not on on-push builds and there are no nightly builds, so that sets us up for totally unseen release failures. I'd think that means that the last time we actually ran tests on the disabled state that we ship was when those branches last built a nightly on Aurora.

Note there are many other places where MOZ_UPDATE_CHANNEL is used this way. I guess we have to change most of them, and forbid the use of MOZ_UPDATE_CHANNEL, it's just too much of a footgun. https://mxr.mozilla.org/mozilla-central/search?string=MOZ_UPDATE_CHANNEL

We should certainly use RELEASE_BUILD/NIGHTLY_BUILD where appropriate, instead of MOZ_UPDATE_CHANNEL.

See https://wiki.mozilla.org/Platform/Channel-specific_build_defines

Comment on attachment 753885 [details] [diff] [review] patch Review of attachment 753885 [details] [diff] [review]: ----------------------------------------------------------------- Looks right to me.

Comment on attachment 753885 [details] [diff] [review] patch Review of attachment 753885 [details] [diff] [review]: ----------------------------------------------------------------- ::: js/src/builtin/TestingFunctions.cpp @@ -168,5 @@ > #endif > if (!JS_SetProperty(cx, info, "oom-backtraces", &value)) > return false; > > -#ifdef ENABLE_PARALLEL_JS There's value in keeping ENABLE_PARALLEL_JS: you're not going to forget one when you want it to start riding the train.

Good point, here's a patch that addresses that.