Generic records causes very slow compile times

[arianvp]

apparently, large APIs using the Servant.API.Generic record mechanism are orders of magnitude slower to compile than their :<|> equivalent

https://twitter.com/mattoflambda/status/1041622124576661504

I'm not sure how much we can do about this, but we could at least benchmark, create a minimal example that triggers this behaviour, and submit it to GHC as a bug

[phadej]

FWIW, servant-generic is old package. Now the functionality is in servant itself (and servant-server etc). I think it's better to use Servant.API.Generic to be unambiguous.

[domenkozar]

Our server module that glues all generic api endpoints together compiles for ~30s on my machine (quad core) on GHC 8.6.4.

[domenkozar]

Note that #986 (comment) also hints that the slowness exists when not using generic records

[alpmestan]

Indeed I think Generic might merely make things a bit worse, and there is a serious slowness in #986.

[arianvp]

Do we want to perhaps merge / close this issue with #986 ?

Also; I wonder if having some servant API stuff as a test fixture for GHC is useful to track performance regressions?

[alpmestan]

Well, I'm specifically looking into #986 to figure out if GHC can do something better. servant would be too heavyweight a dependency but a smaller reproducer would be a great test case, once we identify the source of the problem.

We can leave this open as a reminder to check this again if/when we manage to improve the situation for #986.