Great point. I wish R had an interface to a well-typed language.
I'm surprised Microsoft hasn't done anything in this regard, despite the fact that they own R.
I'm surprised Microsoft hasn't done anything in this regard, despite the fact that they own R.
The most important errors in R are not at all related to data type safety (though NA/NULL handling is a common issue), they're about the statistical soundness of the analysis you're conducting. Experimental design, implicit assumptions and invariants, confounding factors, and the appropriateness of the statistical tests you're applying for the distribution of the data you're applying them to. It sounds like tea-lang helps with the latter part.
Microsoft definitely doesn't "own" R, just one distribution of R (Microsoft R Open) and a package repository (MRAN).
(I'm a Microsoft employee and heavy user of R at work)
You're spot on - Tea is focused on ensuring statistical soundness. Another way to think about this: Tea helps people analyzing data from studies to "maximize internal validity."
I hope to support integration with existing tools/common workflows in the future. Microsoft R Open definitely looks like an interesting target.
It'd be great to follow-up. I visit Microsoft in Redmond periodically. If you're interested, email me! (<my username> at cs.washington.edu)
I was thinking about whether it'd be possible to write something like Tea for R, but then I found there is no R package for Z3 nor any other SMT solver! So that would be an extra challenge...