I spent too many years programming in all kinds of dynamically typed and statically typed systems and I kept a clear head and didn't cling to a blind religious persuasion. I've also watched the performance of static type enthusiasts closely.
I thought that when I updated a type that all of my running around fixing type errors had to do with me missing something but alas that was not the case. This is how the static enthusiasts do their work, all the while claiming that they are being more productive. When I call a static typer our on this the answer is always the same: "But you'd have to go update those case statements/maybe matches/type signatures anyway!" Meanwhile I'm over in a dynamic system not having to do any of that. Refactors are minimal not cross-the-entire-codebase.
Static type systems are training wheels. When a static type enthusiast takes his training wheels off, the bike falls over and s/he screams "See! I told you. I'm more productive with types!" And then they put the training wheels back on and think that they have proven that static verification is superior. What they don't know is that there is enormous missed opportunity once you learn how to ride a bike w/o training wheels.
The truth is you can produce much faster without a rules-based static type prover between you and your runtime environment if you know what you're doing. This should be obvious b/c in one world you have a filter/prover that you must past to ship. In the other world you do not. If the filter/prover was worth its weight then you should be able to clearly see the win. But in business systems the kinds of bugs found by filter/provers are of the "null pointer" variety -- the fastest and simplest bugs to fix when they are found; having a heavyweight filter/prover save me from these simplest of bugs to fix is simply not worth it.
If you have a place where you emit sum types, and you increase the set of possible cases you can emit then you do have to cross-the-entire-codebase to update all consumers of that sum type to handle the new case.
If you have a place where you consume product types, and you increase the set of fields that you inspect on that sum type, then you do have to cross-the-entire-codebase to update all producers of that product type to emit the new field.
I don't see how it could be any different. How would you handle this in Clojure?
Look I have been programming for 35+ years in both dynamic and typed languages. And my experience is that I am more productive when using a typed language. It sounds as if your experience is the opposite. I wonder why our experiences are so different? It would be interesting to find out.
I thought that when I updated a type that all of my running around fixing type errors had to do with me missing something but alas that was not the case. This is how the static enthusiasts do their work, all the while claiming that they are being more productive. When I call a static typer our on this the answer is always the same: "But you'd have to go update those case statements/maybe matches/type signatures anyway!" Meanwhile I'm over in a dynamic system not having to do any of that. Refactors are minimal not cross-the-entire-codebase.
Static type systems are training wheels. When a static type enthusiast takes his training wheels off, the bike falls over and s/he screams "See! I told you. I'm more productive with types!" And then they put the training wheels back on and think that they have proven that static verification is superior. What they don't know is that there is enormous missed opportunity once you learn how to ride a bike w/o training wheels.
The truth is you can produce much faster without a rules-based static type prover between you and your runtime environment if you know what you're doing. This should be obvious b/c in one world you have a filter/prover that you must past to ship. In the other world you do not. If the filter/prover was worth its weight then you should be able to clearly see the win. But in business systems the kinds of bugs found by filter/provers are of the "null pointer" variety -- the fastest and simplest bugs to fix when they are found; having a heavyweight filter/prover save me from these simplest of bugs to fix is simply not worth it.