workaround tty bugs on Node.js

[alexlamsl]

@kzc though #4529 (comment) didn't pan out, it dawned on me some of those recent false positives from ufuzz is likely courtesy of Node.js

[kzc]

At least something positive came out of it.

I did notice an usual number of failures in ufuzz testing. But when I looked at a few they seemed like const variable use before initialization issues and property references on undefined in the original test cases.

[alexlamsl]

Temporal Dead Zones ones are all caused by forward references through typeof fn == "function" && fn(...), and they are both rare, won't trigger false positives, and actually useful against things like #2750

Those cannot set property ones are caused by code like this:

var a = {};
a && ([ a.p ] = [ a = null ]);

These do trigger false positives from time to time, as the string representation may differ after AST_Sub.property gets uglified.

[kzc]

I've got to hand it to you, sifting through all those ufuzz failures is a lot of work. Better automation of test case reduction only leads to more cases to examine.

[alexlamsl]

Well at least these reports are full of verifiable/reproducible details... 😏

[kzc]

I'm still waiting for someone to file an UGLIFY_BUG_REPORT.

[alexlamsl]

May happen sooner if we add EXPORT_GPL() to the output... 🤔

[kzc]

I was unaware of that until I just googled it. Yes, that would work. :-)

[kzc]

Fewer ufuzz failures after this PR was merged.

[alexlamsl]

Yup, mystery solved − I was unable to recreate all those [object Object] locally even after 10MFuzz...