Compute CAD ranks in pure SQL

[gregorbg]

This is a precursor to / offspring of #13266. With the proper indices, it runs in ~20 seconds on my local machine. Probably the biggest advantage in terms of runtime is that we don't need to manually "batch-insert" stuff anymore...

Need approval from WRT that the query is indeed correct. It is comforting that the CI tests are passing with this on my local machine.

[SAuroux]

@gregorbg I took a look now and...well, it works, I guess. 🙂

What deeply annoys me is that the following would just work, if people did not change regions:

Suggested change

	WITH current_person_regions AS (
	SELECT
	p.wca_id AS person_id,
	p.country_id AS current_country_id,
	c.continent_id AS current_continent_id
	FROM persons p
	JOIN countries c ON p.country_id = c.id
	WHERE p.sub_id = 1
	),
	world_stats AS (
	SELECT
	person_id,
	event_id,
	MIN(#{field}) AS world_best
	FROM #{concise_table_name}
	GROUP BY person_id, event_id
	),
	world_ranks AS (
	SELECT
	person_id,
	event_id,
	world_best,
	RANK() OVER (PARTITION BY event_id ORDER BY world_best) as world_rank
	FROM world_stats
	),
	continent_stats AS (
	SELECT
	person_id,
	event_id,
	continent_id,
	MIN(#{field}) AS continent_best
	FROM #{concise_table_name}
	GROUP BY person_id, event_id, continent_id
	),
	continent_ranks AS (
	SELECT
	person_id,
	event_id,
	continent_id,
	RANK() OVER (PARTITION BY continent_id, event_id ORDER BY continent_best) as continent_rank
	FROM continent_stats
	),
	country_stats AS (
	SELECT
	person_id,
	event_id,
	country_id,
	MIN(#{field}) AS country_best
	FROM #{concise_table_name}
	GROUP BY person_id, event_id, country_id
	),
	country_ranks AS (
	SELECT
	person_id,
	event_id,
	country_id,
	RANK() OVER (PARTITION BY country_id, event_id ORDER BY country_best) as country_rank
	FROM country_stats
	)
	SELECT
	wr.person_id,
	wr.event_id,
	wr.world_best AS best,
	wr.world_rank,
	COALESCE(cr.continent_rank, 0) AS continent_rank,
	COALESCE(cnr.country_rank, 0) AS country_rank
	FROM world_ranks wr
	INNER JOIN current_person_regions cps
	ON wr.person_id = cps.person_id
	LEFT JOIN continent_ranks cr
	ON wr.person_id = cr.person_id
	AND wr.event_id = cr.event_id
	AND cps.current_continent_id = cr.continent_id
	LEFT JOIN country_ranks cnr
	ON wr.person_id = cnr.person_id
	AND wr.event_id = cnr.event_id
	AND cps.current_country_id = cnr.country_id
	WITH current_person_regions AS (
	SELECT
	p.wca_id AS person_id,
	p.country_id AS current_country_id,
	c.continent_id AS current_continent_id
	FROM persons p
	JOIN countries c ON p.country_id = c.id
	WHERE p.sub_id = 1
	),
	world_stats AS (
	SELECT
	person_id,
	event_id,
	MIN(#{field}) AS world_best
	FROM #{concise_table_name}
	GROUP BY person_id, event_id
	)
	SELECT
	ws.person_id,
	ws.event_id,
	ws.world_best AS best,
	RANK() OVER (PARTITION BY ws.event_id ORDER BY ws.world_best) as world_rank,
	RANK() OVER (PARTITION BY ws.event_id, cps.current_continent_id ORDER BY ws.world_best) AS continent_rank,
	RANK() OVER (PARTITION BY ws.event_id, cps.current_country_id ORDER BY ws.world_best) AS country_rank
	FROM world_stats ws
	INNER JOIN current_person_regions cps
	ON ws.person_id = cps.person_id

Considering the rest are really rare edge cases, I really wonder if it should not be possible to achieve the correct outcome in a computationally much more efficient way.

[gregorbg]

Yeah, the nationality changes are exactly what almost drove me insane. In my opinion, the logic that "results which were achieved before the nationality change still count towards the old nationality's rankings" makes it pretty much impossible to use the "triple rank" approach which you have outlined.

But I will happily be proven wrong! 😄