A lot of ggplot2 code is boilerplate code, that is, repeated section of code with little to no variation.
Think of all the geom_*() or stat_*() or scale_*() constructors, which are all very similar.
It seems to me that this pattern of duplicated code is susceptible to accruing tiny inconsistencies.

Probably, a lot of this code could be generated (semi-)automatically. This would:

• Reduce duplications
• Decrease inconsistencies
• Be easier to maintain
• Be beneficial for extensions too

Notably #5484 started replacing a lot of duplicated scales_*() code, but I don't think this is systemic enough.