ClojureScript

1.10.597 Release

18 November 2019
ClojureScript Team

Noteworthy Changes

  • The behavior of set/union and into is now aligned with Clojure.

  • subvec argument checking is now aligned with Clojure.

  • apply vector on an array now properly clones the array.

Google Closure Namespace Analysis

The compiler now produces analysis metadata for Google Closure namespaces. This means that for these namespaces:

  • REPL facilities like doc, dir, apropos, etc., will now work.

  • Argument lists are available, thus enabling arity checking.

  • Return types are available, enhancing type inference.

  • Warnings on private var usage will be generated.

To illustrate, let’s (require '[goog.crypt :as crypt]) and explore it at the REPL:

Now, (dir crypt) will list the functions in that namespace:

byteArrayToHex
byteArrayToString
hexToByteArray
...

Docstrings are available; (doc crypt/hexToByteArray) produces:

-------------------------
goog.crypt/hexToByteArray
([hexString])
  /**
 * Converts a hex string into an integer array.
...

Arity information is available. Passing an incorrect argument count to one of the functions, produces an arity warning. For example, (crypt/hexToByteArray "abc" 123) generates:

WARNING: Wrong number of args (2) passed to goog.crypt/hexToByteArray at line 1 <cljs repl>

Type Inference Improvements

Several improvements to ClojureScript’s type inference are in this release.

Direct field access for keyword lookup on records

This is easily explained by way of example:

(defrecord Complex [re im])

(let [x (->Complex 1.1 2.7)]
   (:re x))

The code generated for the last expression will be x.re. This can be anywhere between 66% and 450% faster.

count specializations for string and array

If you apply count to a value that is statically inferred to be either a string or an array, the JavaScript generated will involve direct access to the length field instead of a runtime call to count.

For example (count "abc") will cause "abc".length to be emitted. Depending on context, this could be several orders of magnitude faster.

simple- / qualified- predicate-induced inference

If simple-keyword? or qualified-keyword? is satisfied for a local then that local is inferred to be a keyword. Similarly simple-symbol? or qualified-symbol? results in a local being as a symbol.

This essentially broadens the existing predicate-induced inferrence for keyword? and symbol? to these additional core predicates.

Thread predicate-induced inference through and

This type inference improvement is perhaps best explained by way of an example. For the following code

(and (string? x) (zero? (count x)))

the compiler will now know that, if the first clause in the and above is satisfied, then, in the second clause, x must be of type string. Combined with the count specialization mentioned above, this causes efficient JavaScript to be generated:

typeof x === "string" && x.length === 0

Not inferring on implements?

This is an internal, yet important, optimization for the standard library: When the implements? predicate is satisfied on a local, the compiler generates more efficient function dispatch code for expressions involving that local.

Improperly widened cross-param loop / recur inference

A corner-case involving loop / recur widening inference was fixed where incorrect inferrence can occur if a loop-bound local is used as a recur target.

Dynamic Vars are now properly inferred as having type any

An inference bug was fixed where dynamic Vars were incorrectly inferred as having the type of their initialization value. This was incorrect because dynamic Vars can be re-bound at runtime with values of types that differ from the initialization value type.

Performance Improvements

Optimize assoc on IAssociative values

An optimization in the core assoc function makes it faster when assoc ing onto IAssociative values (the common case).

For example, assoc ing a key-value onto a map can be 24% faster in V8 and 11% faster in JavaScript core, a great perf boost for this frequently used core function.

Tag coll as not-native in ci-reduce

This is an important internal optimization affecting the standard library which improves performance when reducing collections which are IIndexed and ICounted.

Improve perf of cljs.source-map.base64/encode

This improves the performance of a function heavily used in the generation of source maps, improving the performance by 17% in one of our measurements.

Change List

For a complete list of updates in ClojureScript 1.10.597 see Changes.

Contributors

Thanks to all of the community members who contributed to ClojureScript 1.10.597:

  • Dieter Komendera

  • Erik Assum

  • Herald

  • Martin Kavalar

  • Martin Kučera

  • Michiel Borkent

  • Roman Liutikov

  • Seçkin Kükrer

  • Thomas Mulvaney