Screened (Clojure JIRA)

[CLJ-1193] bigint, biginteger throw on double values outside of long range

Fri, 24 May 2013 11:46:12 -0500

This works fine:

user> (bigint (* 0.99 Long/MAX_VALUE))
9131138316486227968N

but this and any other Float or Double values outside the range of a
long throw an exception:

user> (bigint (* 1.01 Long/MAX_VALUE))
IllegalArgumentException Value out of range for long: 9.315605757223324E18 clojure.lang.RT.longCast (RT.java:1178)

Similarly for biginteger

[CLJ-1187] Clojure loses quoted metadata on empty literals

Fri, 24 May 2013 09:14:44 -0500

user=> (meta '^:foo [])
nil

while
user=> (meta '^:foo [1])
{:foo true}

This bug propagates to ^:const vars:
user=> (def ^:const foo ^:foo [])
#'user/foo
user=> (meta foo)
nil
user=> (meta @#'foo)
{:foo true}

[CLJ-1176] clojure.repl/source fails when read-eval bound to :unknown

Fri, 24 May 2013 09:42:41 -0500

clojure.repl/source is broken in Clojure 1.5.0 when *read-eval* is bound to :unknown, since source-fn reads without binding.

Reproduce:

Either set

:jvm-opts ["-Dclojure.read.eval=unknown"]

in Leiningen or eval at REPL:

(alter-var-root #'*read-eval* (constantly :unknown))

(use 'clojure.repl)
(source drop-last)

Expected:

Source of drop-last.

Actual:

RuntimeException Reading disallowed - *read-eval* bound to :unknown

[CLJ-1175] NPE in clojure.lang.Delay/deref

Fri, 24 May 2013 11:02:21 -0500

Summary: Delays get into a corrupt state if an exception thrown, allowing deref to behave differently on back-to-back calls. Patch causes Delay to rethrow the original exception, includes tests.

If a Delay wraps a function which throws an exception, then forcing that delay multiple times causes behavior which is, to me at least, surprising and unexpected: the first time it is forced, the expected exception is thrown; but after that it will behave as if all locals the expression refers to are nil. This can manifest in multiple ways, depending on the expression being delayed:

;; calling a local as a function causes an NPE inside clojure.lang.Delay
(let [f #(/ 1 0) d (delay (f))]
  [(try (force d) 
        (catch Exception e e))
   (try (force d)
        (catch Exception e e))])
[#<ArithmeticException java.lang.ArithmeticException: Divide by zero> 
 #<NullPointerException java.lang.NullPointerException>]

;; if nil is a valid value, this can cause subsequent forces to "succeed"
;; even though the first fails as it should
(let [x (java.util.Date.) d (delay (seq x))]
  [(try (force d) 
        (catch Exception e e))
   (try (force d)
        (catch Exception e e))])
[#<IllegalArgumentException java.lang.IllegalArgumentException: Don't know how to create ISeq from: java.util.Date> 
 nil]

The reason for this is that clojure.core/delay creates a ^:once function, promising to call it only once, so that the compiler can do locals-clearing on its lexical closure. However, when an exception is thrown the Delay object is left thinking it has never called the function, so when it is forced again it calls the function again, breaking its promise to the compiler and causing the function to run after its locals have been cleared.

In fact, once we realize that locals-clearing is involved, we can make the delay behave differently the first N times it is forced, instead of only the first time, by constructing an expression which throws an exception before using all of its locals:

(let [x (java.util.Date.) 
            y (Long. 1)
            d (delay (let [y (seq y)]
                       (cons (seq x) y)))]
  [(try (force d) 
        (catch Exception e e))
   (try (force d)
        (catch Exception e e))
   (try (force d)
        (catch Exception e e))])
[#<IllegalArgumentException java.lang.IllegalArgumentException: Don't know how to create ISeq from: java.lang.Long> 
 #<IllegalArgumentException java.lang.IllegalArgumentException: Don't know how to create ISeq from: java.util.Date>
 (nil)]

I'm not sure what the right fix for this issue is: perhaps the best choice is even to leave it alone, and just document that delay's behavior is undefined if the expression throws an exception. However, I propose making the Delay remember the first exception that was thrown, and rethrow it on subsequent force attempts. This makes sense, in a way: the "result" of the expression was this exception e being thrown, and so this should happen every time. It might be preferable to have the Delay retry the expression until it succeeds, but I don't believe this is possible without abandoning locals-clearing, which would cause perfectly-valid delays to now run out of memory, holding onto no-longer-needed locals just in case the expression needs to retry at some later date.

[CLJ-1171] Compiler macro for clojure.core/instance? disregards lexical shadows on class names

Fri, 24 May 2013 11:29:14 -0500

Summary

The compiler tries to emit jvm native instanceof expressions for direct clojure.core/instance? calls.
For that, it tries to resolve its first argument as a class name. However, it disregards lexical bindings when doing that.
This is incongruent to the default implementation in core.clj

Patches

[Stu] All three patches should be applied IMO.

0002 makes instance? respect lexical bindings
0003 makes instance?'s compiled form check arity, consistent with higher-order behavior
0001 has a minimal test for both 0002 and 0003.

Data

Test case

user=> (let [Long String] (instance? Long "abc"))
false
;; expected true as in
user=> (let [Long String] (apply instance? [Long "abc"]))
true

Culprit method

https://github.com/clojure/clojure/blob/4ccb10edbe66eae81336a4c0972050d9759b0bf7/src/jvm/clojure/lang/Compiler.java#L3578

List Discussion

https://groups.google.com/d/topic/clojure/mf25OlFRpa8/discussion

Tangent

This was discovered because the same compiler macro also omits the arity check implicit in the default definition. This could also conveniently be fixed when touching that method:

user=> (instance? String)
false
;; expected
user=> (apply instance? [String])
ArityException Wrong number of args (1) passed to: core$instance-QMARK-  clojure.lang.AFn.throwArity (AFn.java:437)

EDIT elaborated on ticket title and description; added tangent

[CLJ-1161] sources jar has bad versions.properties resource

Fri, 24 May 2013 11:13:59 -0500

Summary: patch leaves version.properties file out of sources JAR, where it causes problems for tools.

The "sources" jar (at least since Clojure 1.4 and including 1.5 RC) has a bad version.properties file in it. The resource clojure/version.properties is literally:

version=${version}

The regular Clojure jar has the correct version string in that resource.

I came across a problem when I was experimenting with the sources jar (as used by IDEs). I naively added the sources jar to my classpath, and Clojure died on start up. The bad clojure/versions.properties file was found first, which led to a parse error as the clojure version was being set.

[CLJ-1122] Add contributing.md file to github repository (shows clear message on issues/pull request create form)

Fri, 24 May 2013 08:24:38 -0500

This adds a clear message when someone wants to create a pull request/issue and invites the user to read the contribution guidelines: see https://github.com/blog/1184-contributing-guidelines.

The same thing could be done for all the clojure/* repositories.

The content of the file is just a markdown version of http://clojure.org/contributing

Preview here: https://github.com/mpenet/clojure/blob/aef170ca5eca1b71a2eb1ef320223d1277df0e5e/CONTRIBUTING.md

[CLJ-1121] -> and ->> have unexpected behavior when combined with unusual macros

Fri, 24 May 2013 08:48:42 -0500

My intuitive understanding of the classic threading macros is that the meaning of forms like (-> a b c) can be understood syntactically independent of the meaning of the symbols involved or the fact that the two threading macros are defined recursively. However the recursive definition breaks that expectation. After

(macroexpand-1 (macroexpand-1 '(-> a b c)))

=> (c (-> a b))

c is now in control if it is a macro, and is now seeing the argument (-> a b) rather than (b a) as would be the case if we had written (c (b a)) originally.

Admittedly I do not know of a realistic example where this is an important distinction (I noticed this when playing with a rather perverse use of ->> with macros from korma), but at the very least it means that the behavior of the threading macros isn't quite as easy to accurately explain as I thought it was.

[CLJ-944] compiler makes different concrete maps then the reader

Fri, 24 May 2013 09:03:46 -0500

I (Stu) agree with Nicola's assessment in the comments: the single real problem here is that the compiler's MapExpr parser emits maps differently from other map-makers, e.g. RT's factory functions.

Patch clj944-plus-tests does three things:

has the compiler emit map types consistent with the reader (like the previous "0002" patch)
adds tests
removes broken tests

Original report follows:

Hi guys, I am getting the following exception:

(.containsKey {:one 1} :one)
;=> ClassCastException clojure.lang.PersistentArrayMap cannot be cast to clojure.lang.PersistentHashMap

The map is a clojure.lang.PersistentArrayMap, which obviously has a containsKey method (https://github.com/clojure/clojure/blob/master/src/jvm/clojure/lang/PersistentArrayMap.java#L95).

Casting it works fine though:

(.containsKey ^clojure.lang.PersistentArrayMap {:one 1} :one)
;=> true

The mailing list suggest that the compiler injects an incorrect cast to clojure.lang.PersistentHashMap. In this case it should probably be cast to a clojure.lang.Associative, the highest common interface having the .containsKey method.

The problem is not present in Clojure 1.2.1.

[CLJ-850] Hinting the arg vector of a primitive-taking fn with a non-primitive type results in AbstractMethodError when invoked

Fri, 24 May 2013 12:12:20 -0500

Summary: The CLJ-850-conform-to-invokePrim.diff patch is constructed per Rich's feedback, and appears good to me [Stu].

See the following examples:

user=> (defn f1 ^String [^String s] s)
#'user/f1
user=> (f1 "foo")
"foo"
user=> (defn f2 ^long [^String s ^long i] i)
#'user/f2
user=> (f2 "foo" 1)
1
user=> (defn f3 ^String [^String s ^long i] s)                                       
#'user/f3
user=> (f3 "foo" 1)
AbstractMethodError user$f3.invokePrim(Ljava/lang/Object;J)Ljava/lang/Object;  user/eval8 (NO_SOURCE_FILE:6)

[CLJ-827] unsigned-bit-shift-right

Fri, 24 May 2013 14:24:09 -0500

Add a clojure equivalent of >>>.

A simple version of this is implemented here (https://github.com/joegallo/clojure/tree/unsigned-bit-shift-right), and just follows the example set by shift-right.

The downside of this implementation is that it treats all integer types as longs, and shifts them accordingly, which yields different results than you would get in java. A previous version of this did not have the same problem, when BitOps was its own thing. I'm not sure if this limitation is acceptable and appropriate, or needs to be worked around (my inclination is the latter).