Here's my counterproposal to ModuleFactoringShinn, with rationales:
- Move (scheme io) back to the base.
- It provides string-ports and the shared function of all ports, and there is no reason why all Schemes shouldn't provide these. Note that (scheme io) was never actually proposed or voted on, and its presence in the first draft is an error by the editors.
- Don't provide an umbrella module for I/O.
- Files, reading Scheme objects, and writing Scheme objects are orthogonal. Demanding that your implementation provide all of them as a matter of course, across the whole range of Schemes, doesn't make much sense. We already have a ticket allowing Schemes to pre-import whichever modules they want in the REPL.
- Leave interaction-environment in the (scheme repl) module.
- Alex wants to move this to (scheme eval), but the essence of (scheme repl) is that it provides not only evaluation of expressions, but a mutable global environment. There is no reason why a Scheme can't provide the first without the second.
- Leave case-lambda in the base module.
- It captures a useful pattern and can be optimized by a slick compiler or by hand.
- Move (scheme multiple-values) back to the base.
- Multiple values are easy to hack an implementation for (lists, or a list wrapped in a record if you insist), and at least exact-integer-sqrt depends on them. Alex wants to move this procedure to a separate module so it can depend on (scheme multiple-values), but I think that's piling Pelion on Ossa.
- Rename (scheme unicode) to (scheme char).
- Alex and I agree on this one. The essence of this module is that it does things which, on a full-Unicode system, require a big table. Making it a module means that you can have full Unicode without the big table unless specific code requires it.
- Move char-alphabetic?, char-numeric?, char-upper-case?, char-lower-case? and char-whitespace? to (scheme char).
- Alex and I agree on this one; not doing it in the first place was an oversight on my part.
- Move the normalization procedures to (scheme char normalization).
- I still prefer to keep the names of WG1 modules simple, but okay, fine.
Not part of the proposal
Alex says that depending on discussion, he would consider moving
the following out of the core:
- syntax-rules
- define-record-type
- blob procedures
- string-ref and string-set! (but possibly he means string-set! and string-fill?)
The first three are IMO important parts of Scheme and cheap to provide directly on all reasonable systems. There is no reason why all Schemes can't support them, which I consider the most important factor in moving things to modules. It doesn't make sense for all Schemes to provide inexact arithmetic (which is mostly about having floating-point instructions available) or files, so they belong in modules.
I am no fan of mutable strings, but reluctantly I conclude that programs depend on them and we can't get rid of them, so all Schemes might as well provide them. Hence no module.