Following up after the thread has moved on -- but I think if you spend some time in the "tidyverse" and read the Advanced R / R4DS books, you'll find that the t

Following up after the thread has moved on -- but I think if you spend some time in the "tidyverse" and read the Advanced R / R4DS books, you'll find that the tidyverse extends the language's core competencies in a way that fixes a lot of the inconsistencies, and it is possible for people to basically never have to interact with base R if they don't want to. We train our incoming PhDs in R focused on the tidyverse and they generally get the hang of it pretty quickly. The problem you're highlighting was pretty bad 5-8 years ago where core R developers were pushing three different object models. These days pretty much everyone has agreed to use S3. The bigger challenge in data science is that statisticians are generally shitty programmers and so many of the packages are very much v0.1 releases designed to accompany a journal article and never iterated on. The reverse tends to be true in Python -- the notorious thing with scikit-learn defaulting to penalized logistic regression last I checked is insane. If none of this is news to you, then agree to sort-of disagree for sure, but if you haven't checked out R4DS or spent much time with modern packages you'll see that you can pretty safely leave behind a lot of the stuff you used to hate. I re-wrote the analysis for a paper I wrote I think 4 years ago, and I was surprised at how little code didn't change in that time.