I recall Chouser (corrections welcome) noting that in Clojure code, parentheses are used primarily for "forms" (applications of special forms, macros and functions) and perhaps a good goal would be to avoid using them elsewhere. Doing so would more strongly enforce the close association between parentheses and "calls".
I experimented with this in an "ns" form and found that because of the coolness of the seq abstraction, it appears that all uses of lists in an "ns" form can be replaced by vectors and still work the same.
Here's an example:Currently the canonical form for ns is a mixture of lists and vectors within its clauses:
(ns clojure.contrib.miglayout.internal
(:import (java.awt Component))
(:use (clojure.contrib
[except :only (throwf)]
[fcase :only (fcase)])))
If we reserved parentheses only for the ns form itself, this could
become:
(ns clojure.contrib.miglayout.internal
[:import [java.awt Component]]
[:use [clojure.contrib
[except :only [throwf]]
[fcase :only [fcase]]]])
The ns form is an interesting case because Rich has expressed a vision
that it will be parsed and used not only by Clojure itself, but also
by tools/IDEs to determine what a source file contains and what it
depends on. As such, its "acceptable syntax" is governed (even more
than elsewhere in Clojure code) not primarily by "what works in
Clojure", but rather by "what is its canonical, documented syntax".
Should we consider changing the canonical form for ns to use vectors for all of its sequential sub-forms?
Benefits:
- canonical form would be easier to describe
- less to remember and get right than with the current mixed
syntax
- makes it clear that (:use ...), for example, is not a "call"
- supports the "parentheses imply form application" notion
Downside:
- Nested uses of square brackets are arguably less readable than a
mixture.
- matching brackets by eye is easier when there's a variety
--Steve
smime.p7s
Description: S/MIME cryptographic signature
