It doesn't work :(

Edit: I didn't read the code — this approach does work (it just uses too many queries in this instance)

For the edge case, here's what I did.

Terminology: Let the centroid be R, the subtrees be X,Y,Z. wlog let X be the biggest subtree.

This edge case occurs is at the point where SZ(X) = SZ(Y) + SZ(Z) + 1, we are currently in subtree Y, where ht(X) < ht(Y) < ht(Z). In this case, we will jump from Y, then to X, then to Z.

However, we cannot do this as ht(Z) + ht(X) > ht(Y) + ht(X). To resolve this, we make an additional jump from Y to Z.

This is my implementaton:

if(t.s!=MAJ && t.f!=-1 && SZ(res) && pd[res.back()] < t.f){
  	res.pb(t.s);
  	V[belg[t.s]].pop_back();
  	blk=-1;
}

Edit: tidied up my code a little bit, https://pastebin.com/Fq3hDrGB

Friday, 21 August 2020 19:00 (UTC+7) is the closing ceremony

Yes, but it is a bit simpler.

I think there is a solution that uses kruskal reconstruction tree to solve the problem

https://tlx.toki.id/problems/apio-2020