There is already the atcoder library which implements a lot of data structures and algorithms.

Should we also ban vectors and sets and let everybody code their own?

Unironically yes. Maybe we won't get tons of retarded blogs like "why my solution with unordered_map fails with TL" or "why my solution with deque fails with ML" if people actually implement and understand how data structures work.

Don't forget to add these algs to the library:

• 5D incremental convex hull in $$$O(n\log n)$$$ + 5D onion decomposition in $$$O(n\log^2 n)$$$
• 3D Voronoi diagram in expected $$$O(n\log n)$$$
• 4D online half-hyperspace intersection in $$$O(\log n)$$$ per query
• Fast arbitrary precision floating point numbers for super precise calculations
• Bigint as a consequence of the previous point
• FFT, Newton's method, Burnikel-Ziegler-Verfahren algorithm as a consequence of the previous point
• KD tree, Vantage point tree, R-tree for getting AC with $$$O(n^2)$$$ solutions
• Visual debugging tool for all that mess

these are only $$$1\%$$$ of "Geometry" topic. Good luck in implementation xD xD xD.

You will still need to explore what does the codeforeces library have then. Someone will know about some functions, someone won't. Still unfair xd

Also, we need that library for all the possible contest languages? Creating a large library only for one language is also very unfair

I think it can be one large library where we all as users can contribute to improve it, it might be a great idea

I would say that having a library prepared beforehand is fair game. Here are my arguments:

• Everything templateable is already on the internet. If you don't have your own Dijkstra implementation, you are not on much worse position than other better prepared contestants. Searching for it would immediately yield you a code in your prefered language. In the nearest months, maybe even AI will be able to provide implementations for those boilerplate algorithms and data structures.

• A template library is part of contestant preparation. Sure, you can use segment tree from the internet but having your own implementation can make you adaptable to more exotic use cases (let's say that the operation is xor by x, query max which you are unlikely to find on the internet).

• Even if you forbid any outer source of code or library, one could argue that better prepared contestants have the code memorized and that's unfair because now they can implement some DS faster then others.

• Preparing a library is a natural part of CP journey: while solving a set like CSES or studying classical textbooks like Cormen you discover a broad lore of widely applicable algorithms. While training, you are encouraged to implement them yourself and to save the implementations for later use.

• I can't imagine implementing rare, large templates each time from scratch (like max flow, convex hull trick, simplex etc.). It would be a source of lots of bugs, WAs and frustration.

• What about the overal template with macros, typedefs etc? Is the usage of such template unfair as well?

• [edit] One more: selecting a template from library is rarely a crucial part of solving a problem. The library is intended to help with the most mundane and reapetable parts of problemsolving.

I feel like you are shifting the point of a standard library a little bit. You don't need to create your own library to use algorithms. You can use somebody else's codebase. Or atcoder library, for example. The only advantage of a codeforces library would be that you don't need to copy-paste this code into your source file so it would save you a minute. I am not against it but I don't see it as any kind of important thing. I've never used atcoder library myself but I appreciate the fact that when someone uses it, it makes it easier to see what is actually new in the solution and what is standard. See where the actual solution idea is.

You are talking about the fact that using plain c++ is not a part of the sport but in my opinion, it is a part of the sport! Yes, std::set is a data structure, but I get much more joy if I can solve a problem using std::set instead of a segment tree. This is a so-called "no data structures" achievement. There are these artificial "constraints" but they are the beauty of the sport. Plus, I really enjoy reading the libraries of other people sometimes. It is very cool to see how people put a lot of thought into trying to figure out the most general and easy-to-use black-box implementations of algorithms and data structures.

Imo, part of the fun of competitive programming is googling similar problems during contests and copy-pasting their solutions together. A standard library would kinda take away from that.

I think it's up to contestants to prepare their library in their own familiar coding style.

Atcoder's library is convenient and fast, but many files are ridiculously long because of extreme optimizations. It makes little sense to print it and bring into ICPC contest floors: you won't want to type pages of convolution template before starting a polynomial problem.

Our team have two different segtree templates, each written by one of the two main coders.

I think the library isn't necessary. In fact most basic templates are very short, so the extra time spent on writing templates is not significant compared to the time you actually use to figure out the solution(I can write dij in 1~2min and segtree in 3min). For complicated/long templates(e.g. floorsum/FFT/flows/bignum/toptree), everyone uses their own library, so this makes no difference.