Yep, something like this should work i suppose.

Hm, and the msvc failures seem real, probably need to explicitly cast something in tests from size_t to int64_t.

Looks like more casts are needed :)

Hmm, those msvc failure look weird. I don't think this PR can cause them.

Hmm, those msvc failure look weird. I don't think this PR can cause them.

I reproduced the failure on my Windows computer. I don’t understand why, but it seems like the failure is caused by my PR.

e1eaa52 may help.