Scheme is a great language. I found it to be one of the best "normal" programming languages for solving numeric stuff. The functions on top of its awesome numeric tower are something missing in pretty all other programming languages. Yeah, you can get it in your <language> by using a <lib>. You don't need libs in Scheme though.
Apropos libraries: there are no Scheme libraries with exception of commonly accepted SRFIs and the SLIB. If I look at the front page of SLIB, it tells everything: ``SLIB supports Bigloo, Chez, ELK 3.0, Gambit 4.0, ..." A great effort, will I have to input the same effort for my code? Yes!
Every Scheme implementation tries to specialize itself on a single approach, what is in theory really good - take best tool for the right job. The problem arises if you want to combine two approaches. You are locked in a room with a single implementation and either you select the best of the one or the best of the other. Don't get me wrong: for example C and C++ suffered for a long time of the same plague: code was bound to a single compiler or platform.
In contrast perl, java etc won of a singleton aproach - they ran everywhere and ran everyone's code. The reason the great code base exists is the fact that it was so easy to deploy own code. People wrote, posted and forgot about it. You cannot do that with Scheme. You can do that with PLT Scheme or Chicken Scheme or ... but you simply cannot do that with Scheme. You cannot extend R5RS without breaking compatibility. Take FFI for example.
Maybe in 20 additional years there will be a few generally accepted SRFIs defining standard minimal networking and FFI interfaces for that future RxRS. So that if you write a library only using that minimal subset, you will actually be able to write anything useful. You just cannot ignore those in the UI and Internet age. |