@Esteban Küber to clarify, this PR isn't aiming to do the right thing per se, just to avoid an underflow?
cause it's not obviously right to me :)
It's only to stop the ICE from happening. The proper fix will be to rework the way the "unclosed angle brackets count" is kept and reset
I believe the problem is because there's a reset when changing from one parse scope to another, and currently it never recovers the previous count
the only reason this isn't a problem with debug assertions off, is because as long as all the angle bracket count is accounted for which won't emit any errors, the count will be undeflowed and will be an arbitrarily big number
You could have a case where the angle bracket count is mismatched and the error suggests removing 32K opening brackets, but that would be "unlikely" to happen
by changing it this way, it devolves to the prior behavior of only complaining about incorrect token with no suggestions, which I believe is reasonable.