It seems to me that some of the metrics in the PP system are extremely abuseable, and end up shaping the meta significantly (people play what gives PP, people map what people play, so people map for PP). The 2 main culprits for metrics that affect this are max combo, and misses.
The PP system values misses evenly independantly of the number of objects in a map, using the formula 0.97^m (at least, this is the last value I saw, and what I have seen seems to reflect this fairly closely). If I play a short PP farm map with 150 objects, and miss 10% of them (effectively butchering the map, 15 total for the short map), my PP drops by 63%. A fairly significant loss, but if the base PP is high, retaining 37% of this base value can steal lead to a decent play. If I play a long map such as Yomi Yori Kikoyu, there are 3070 objects. If I miss 10% of them, I am essentially playing the map as well as the above map, which is 307 misses in this case (not even accounting for fatigue, you will likely miss an even higher % of notes in the longer song due to fatigue, yet instead of this being accounted for, it is punished, severely).
In this case, however, instead of retaining 37% of my PP, I retain less than 1% (I believe there ends up being a lower floor, or somthing similar, but this is hit long before 307 misses). In order to have the same effect as the shorter map, I would have to miss only 15 notes here as well. On the short map, hitting 90% of notes is equivalent to a 99.5% hit rate on the longer map.
The solution for this would be to have PP be effected by misses based on percentage misses as opposed to flat number of misses. An example suggestion of an adjusted formula could be:
(1-(30/max notes))^m
The other culprit (combo), is an even more abuseable metric, for 2 seperate reasons (length, and consistency). If a song has difficulty spikes spaced evenly throughout it, you will likely combo break periodically, never retaining a significant combo (significant meaning 50% of a song, as 50% of a song causes you to lose ~43% of your PP, and going below this causes huge losses). This means a PP map with 200 easy notes and 20 hard notes, with the 20 hard notes all at the end will allow players to miss almost all of the hard part, and still retain 90% combo, causing large PP gains, whereas a map that has 20 hard notes in bursts of 5 in 4 separate places a player may hit the same easy notes, and miss the same hard notes, but where the first player would only lose 8% of a songs base PP, the second player would lose 73% of it. This is not even accounting for the second point, length.
If a player hits the same portion of notes on a long song and a short song, you can expect their max combo to increase roughly with the log of the songs length, yet the formula expects this value to increase linearly. (It is way easier to hit a 75 note combo on a 150 note song than a 1500 note combo on a 3000 note song, just from an intuitive perspective, yet these award equal pp).
The solution to combo is to remove it from the PP calculations altogether. I realize that this might seem to be rather poor feedback, but for the above reasons I believe that combo is essentially impossible to gauge as to how it reflects a players performance, and mappers simply end up mapping short songs with 80% of the song being super easy and the last 20% being a massive diff spike to abuse this system.
In addition to the above 2 points, since short maps can be played much more quickly, more attempts in a similar period of time means you get more chances for a good run. I can play a 1 minute map 5 times in the same time I can play a 5 minute map, so it should be expected that a run on a 5 minute map would give more PP than a similar performance run on a 1 minute map (reward to time spend ration should be similar). The PP calculation for song length is shaped sort of like a graph with the formula 1/x^n. I'm not sure the exact values, but this could likely be tweaked a little bit so diminishing returns from extra PP from long songs tapers off a little slower.
The current meta focuses on very short songs with 75-80% easy sections, and short diff spikes at the end. It feels a bad that top 50 plays on the Promethean kings are giving single digit PP (1 play even giving 0 pp), and a DT play on Harumachi Clover where you miss 10 or 15 notes can give triple digit PP with much less skill required, and only takes 20 seconds, and this effects how mappers make their maps. Hopefully with a few tweaks to the system, the formula could better reflect how impressive a play on a given map actually was, which would allow mappers map different styles and lengths of maps without the player base ignoring them as much as they give no PP (unless you are cookiezi and can FC 3000 note maps with tons of jumps and deathstreams).
TLDR:
-change formula for misses to account for song length (ex, (1-(30/notes))^misses
-remove combo from the formula (too hard to account for, too abuseable)
-reward marathon maps slightly higher
EDIT:
Could also add a damping formula to prevent a small number of notes creating all the PP value in a song (ex below of how this can cause issues)
https://youtu.be/Vh15Q9Mdias
https://youtu.be/5iXTTfozN0U
https://youtu.be/CsqMGBJ1m6g (3.5k PP 1 button click)
https://www.youtube.com/watch?v=cjLABPThKuk (after more consideration, a damping formula is really needed to prevent this kind of abuse)