All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=918556

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=fab1683675872c97e0bea2b7e86f800f464881f1204d81df427e34c628a33e70


Bot(s) for this bug's original alert(s):

Google Chrome Win

I doubt given the range and the lag to the commit we can do anything. feel free to close.

+clemensh

I have a vague suspicion that it might be due to disabling noexcept on msvc in https://chromium-review.googlesource.com/c/v8/v8/+/1386869.

There is no similar regression in d8 on my linux workstation.

We should try to bisect to a single CL.
If it's really the noexcept annotation, we can do something about it. We would basically have to wrap each standard container and mark the constructors and assignment operators noexcept. That's doable with a single template. It would be a bit ugly because we have to identify all locations where it's actually needed. Some fields would be declared as:

base::NoExcept<std::vector<Foo>>

On non-msvc, that would be a nop-typedef:

namespace base {
template<typename T>
using NoExcept = T;
}

But as I said, before doing this, we should bisect the size increase.