Clojure JIRA

[CLJ-423] make sure future clears closed-overs

Tue, 24 Aug 2010 09:55:00 -0500

As lazy-seq and delay already do.

[CLJ-422] code in finally clauses is sometimes run twice

Tue, 24 Aug 2010 00:55:00 -0500

When an exception is thrown in a finally clause, the code inside that finally clause is run twice:

(try (prn 1) (finally (prn 2) (prn 3) (throw (Exception.))))

That prints 1 2 3 2 3, when it should simply print 1 2 3. This has been the case since revision:5e9f2b293b307aa7953cd390360d24549e542b92

But simply reverting that commit causes problems in other cases, such as:

(try (prn 1) (throw (Exception. "one"))
(catch Exception e (throw (Exception. "two")))
(finally (prn 2) (prn 3)))

...which should print 1 2 3 as well, but used to print just 1 instead.

This was discussed in http://groups.google.com/group/clojure/browse_thread/thread/43d2dcb2505e8915

[CLJ-418] fix record equality with other maps

Tue, 24 Aug 2010 17:42:00 -0500

Record equality is non-symmetric with maps, due to record equality using type and map equality not. Make it so record equiv uses type but equals doesn't, and make Util.equiv use IPersistentCollection's equiv if on either side. Add marker interface for records. Make APersistentMap check for record marker interface.

[CLJ-417] Add more information to changes.txt for case and vector-of

Tue, 24 Aug 2010 00:42:00 -0500

The information for case and vector-of could stand to be a little more detailed. I've attached a diff (against the RC1 changes.txt) that adds a little more information.
:f

[CLJ-413] clojure.java.shell/sh in-enc doesn't default to UTF-8

Tue, 24 Aug 2010 00:35:00 -0500

clojure.java.shell/sh was recently updated (changed) to say that it defaults to UTF-8 for both in-enc and out-enc, but it does not do so for in-enc.

On line 122 of the current build:

(with-open [osw (OutputStreamWriter. (.getOutputStream proc) ^String in-enc)]

in-enc is nil by default, but OutputStreamWriter needs to be passed "UTF-8" instead if in-enc is nil. I would add a patch but I still need to submit the CLA.

[CLJ-407] Function names not munged in :gen-class

Tue, 24 Aug 2010 18:32:00 -0500

A function name with a dash in the :methods list isn't properly munged when using compile/gen-class. This results in a method in the byte code that is not callable from Java.

Offending code:

(ns foo.bar
(:gen-class
:methods [[foo-bar [] void]]))

(defn -foo-bar [] (println "a"))

[CLJ-404] Making a writer from a Socket with clojure.java.io/writer fails

Tue, 24 Aug 2010 00:22:00 -0500

In clojure.java.io, IOFactory is extended to java.net.Sockets. The make-input-stream method is implemented correctly, but the make-output-stream method is missing. Actually, there is another key in the map named :output-stream. That is probably some old code lying around that didn't get refactored.

How to reproduce

(use 'clojure.java.io)
(writer (java.net.Socket. "google.com" 80))

When evaluated, the following is thrown: java.lang.IllegalArgumentException: Cannot open <#<Socket Socket[addr=google.com/66.102.13.147,port=80,localport=55335]>> as an OutputStream. (NO_SOURCE_FILE:0)
Any host/port combination that is open will do. I chose Google's HTTP server as an arbitrary example of an open port.

Expected Output

An instance of java.io.BufferedWriter.

Affected Versions

All versions of Clojure 1.2 since commit cd8fc7c11213fbe99355 "promote clojure.contrib.io to clojure.java.io (with api changes)" by Stuart Halloway.

Group Discussions

I mentioned this on the Clojure mail list at the end of this thread: http://groups.google.com/group/clojure/browse_thread/thread/898f5dbe14057277/8fe69b1d2104f8b4?lnk=gst&q=code+review#8fe69b1d2104f8b4

Proposed Solution

I signed the CA and posted it in the middle of June, but my name has not appeared on the list of contributors yet. I take the liberty of demonstrating how simple the fix should be:
<pre>diff --git a/src/clj/clojure/java/io.clj b/src/clj/clojure/java/io.clj
index 4d6c551..0151482 100644
— a/src/clj/clojure/java/io.clj
+++ b/src/clj/clojure/java/io.clj
@@ -260,7 +260,7 @@
(extend Socket
IOFactory
(assoc default-streams-impl
:make-input-stream (fn [^Socket x opts] (.getInputStream x))

:output-stream (fn [^Socket x opts] (output-stream (.getOutputStream x) opt
+ :make-output-stream (fn [^Socket x opts] (.getOutputStream x))))</pre>

Possible Unit Tests

A simple test that demonstrates this bug would be to check that the result of clojure.java.io/writer returns a java.net.Writer. Preferably, there should be unit tests for all the classes IOFactory is extended to.

[CLJ-403] extend-protocol doc is out of date

Tue, 24 Aug 2010 18:21:00 -0500

The documentation to extend-protocol is out of date. It refers to extend-class, which was merged into extend-type, and it shows the old ::TypeName notation.

[CLJ-402] degenerate defrecords should act like empty maps

Tue, 24 Aug 2010 18:21:00 -0500

In the 1.2 beta1, empty defrecords misbehave, e.g.

(defrecord Foo [])
(seq (Foo.)) => () ; should be nil!

This also contributes to the broken .containsValue mentioned in http://groups.google.com/group/clojure/browse_thread/thread/5d27195d45b2d214/fb51906b2f960d4c

[CLJ-399] make sure shell out and err handled in parallel

Tue, 24 Aug 2010 00:16:00 -0500

make sure shell out and err handled in parallel

[CLJ-398] Stray println snuck into clojure.java.shell/sh

Tue, 24 Aug 2010 00:08:00 -0500

Line 94 in the current build; let's get rid of that, shall we? =)

[CLJ-393] Fix rename-keys to work with defrecords

Tue, 24 Aug 2010 10:59:00 -0500

clojure.set/rename-keys does not work with defrecords, because they do not implement IFn. rename-keys uses (map key) rather than (get map key).

[CLJ-392] Numerous fixes for clojure.java.shell

Tue, 24 Aug 2010 10:59:00 -0500

I have posted a patch to a number of issues discussed on the mailing list:

read stdout and stderr simultanously from separate threads to prevent stderr filling the buffer and hanging the process
removed stray println
use clojure.java.io to copy streams, avoiding byte-at-a-time copying
added :inenc option, specifying the input character set
renamed :out input option to :outenc
changed default encoding to 'platform default encoding'
write to stdin from a separate thread to prevent stdout blocking before we write anything

Additionally I have added support for a feature mentioned in the post, but not implemented in the original patch set:

added support for passing a byte array to :in

Original post:
http://groups.google.com/group/clojure-dev/browse_thread/thread/b4a4d6eca3fd9b9d/b12cc9e47f03f617#b12cc9e47f03f617

Followup discussion:
http://groups.google.com/group/clojure-dev/browse_thread/thread/dde2b5311212c072/3b24749935cfa4aa#3b24749935cfa4aa

[CLJ-391] namespaces should be able to replace their own vars

Tue, 24 Aug 2010 11:58:00 -0500

The code that prevents vars from replacing vars referred in from other namespaces is too strict: it also prevents the :reload-all scenario for a single namespace by spotting collisions between the old and new versions of the same var.

See http://groups.google.com/group/clojure-dev/msg/13978a18923e3356

[CLJ-388] fix gvec reflection warnings

Tue, 24 Aug 2010 10:52:00 -0500

quick fix while I investigate followon issues...

[CLJ-387] clojure.java.shell needs different byte coercion

Tue, 24 Aug 2010 10:50:00 -0500

Same as contrib issue https://www.assembla.com/spaces/clojure-contrib/tickets/83-clojure-contrib-shell-broken-with-clojure-commit-37d8f7a

[CLJ-386] spit works only on strings

Tue, 24 Aug 2010 10:47:00 -0500

As reported by Kevin Downey (http://groups.google.com/group/clojure-dev/browse_thread/thread/36e1fa06d91e1ee1):

user=> (spit "/tmp/foo" {:a 1})
java.lang.IllegalArgumentException: No matching method found: write
for class java.io.BufferedWriter
(NO_SOURCE_FILE:0)
user=>

This happens because the argument to spit isn't converted to a string before calling BufferedWriter.write

[CLJ-382] fix underive for multiple inheritance

Tue, 24 Aug 2010 11:43:00 -0500

Underive is incorrect: not all maps are updated, and the wrong key is used. See http://groups.google.com/group/clojure/browse_thread/thread/b70f8bf16de2216

[CLJ-381] pprint: logical-block macro uses private vars

Tue, 24 Aug 2010 11:43:00 -0500

Alex Ott noticed that the logical-block macro that's used to build dispatch tables uses two vars and a function that are private to clojure pprint. This makes it more difficult for programmers to build custom dispatch.

[CLJ-377] fix file/line reporting in ant builds

Tue, 24 Aug 2010 11:35:00 -0500

clojure.test figures out where a failing test is by manufacturing an exception instance during reporting, then crawling back up the stack a fixed distance to find the test.

[CLJ-376] make sure tests pass on Windows

Tue, 24 Aug 2010 11:35:00 -0500

make sure tests pass on Windows

[CLJ-367] defrecord doesn't work with certain field names

Tue, 24 Aug 2010 08:21:00 -0500

When a field name in a defrecord coincides with an argument name in one of the method implementations, strange things happen:

(defrecord Foo [o])
java.lang.RuntimeException: java.lang.IllegalArgumentException: Unable to resolve classname: Foo
at clojure.lang.Compiler.maybePrimitiveType(Compiler.java:1118)
at clojure.lang.Compiler$MethodExpr.emitTypedArgs(Compiler.java:1142)
...

(defrecord Foo [this that])
(let [a-foo (new Foo :a :b)]
(with-meta a-foo {:foo :bar}))
java.lang.StackOverflowError

The attached patch introduces gensyms for all the critical method argument names, such that clashes are no longer possible.

In Clojure-Dev: http://groups.google.com/group/clojure-dev/browse_thread/thread/9e4c1956d75588b8/8261a8d112f1eadf?lnk=gst&q=defrecord#8261a8d112f1eadf

[CLJ-365] better defn error messages

Tue, 24 Aug 2010 00:21:00 -0500

better error message for (defn foo (not an arg-vec)), plus a place to hang other arg validations for defn

[CLJ-362] incorrect result of reductions for empty input sequences

Tue, 24 Aug 2010 00:20:00 -0500

reductions does not return the correct result in case of an empty input sequence.

(reductions + nil)
actual: 0
expected: (0)

[CLJ-361] clojure.core/merge-with doesn't work with defrecord types

Tue, 24 Aug 2010 00:20:00 -0500

clojure.core/merge-with works only with map types that implement clojure.lang.IFn for element lookup. In particular, it doesn't work with types defined by defrecord.

http://groups.google.com/group/clojure-dev/browse_thread/thread/8e546e595a4c4cab/aa47af1770183c56?lnk=gst&q=merge-with#aa47af1770183c56

[CLJ-360] NullPointerException on disj

Tue, 24 Aug 2010 00:20:00 -0500

disj throws a NPE when given a nil set. Tested using a build of clojure master branch as of 2010-05-26.

Clojure 1.2.0-master-SNAPSHOT
user=> (disj nil :test)
java.lang.NullPointerException (NO_SOURCE_FILE:0)

Expected result to be nil.

http://groups.google.com/group/clojure/browse_thread/thread/3090d34f9ccf75ea

[CLJ-359] promote contrib string

Tue, 24 Aug 2010 00:18:00 -0500

Promote the following from clojure.contrib.string to a new namespace, clojure.string

join
chop
chomp
split (fix perf, return vector)
replace (subsumes all replace- variants)
replace-first (subsumes all replace-first- variants)
reverse
capitalize
upper-case
lower-case
trim

Where necessary, change fns to take the string as first arg

Issues raised on mailing list:

perf test, particularly join
design guidelines for when nil works
arg order / point-free style?
trim options
trim modifiers at end

[CLJ-358] pop! on a 33-item-long transient vector returns a persisted transient vector

Tue, 24 Aug 2010 00:17:00 -0500

user=> (-> (range 33) vec transient pop! count )
java.lang.IllegalAccessError: Transient used after persistent! call (NO_SOURCE_FILE:0)

Obviously persistent! is never called.

The applied patch doesn't solve a similar issue with transient vectors of size 33+32^n with n >= 2/

[CLJ-357] promote contrib javadoc

Tue, 24 Aug 2010 17:17:00 -0500

Aside: while at it, don't use the clojure.repl/*-fns at the repl

[CLJ-356] better error message on failed refer

Tue, 24 Aug 2010 17:14:00 -0500

error message "is not public" is misleading when usually the var doesn't even exist.

[CLJ-355] eliminate reflection from bit-shift slow path

Tue, 24 Aug 2010 17:14:00 -0500

"slow path" doesn't have to be "cruelly slow path"

[CLJ-353] Deftypes not re-evaluated when protocols change; causes errors

Tue, 24 Aug 2010 09:12:00 -0500

I keep having to restart my JVM to avoid a puzzling issue when changing protocol definitions.

Here's the simplest way to reproduce the problem:

Start with file protoproblem/proto.clj:

(ns protoproblem.proto)

   (defprotocol Steps
     (one [x])
     (two [x]))
</code></pre>

and file protoproblem/impl.clj:
<pre><code>   (ns protoproblem.impl
     (:require [protoproblem.proto :as proto]))

   (deftype ArraySteps [a]
     proto/Steps
     (one [x] (first a))
     (two [x] (second a)))
</code></pre>

In the repl, everything works as expected:
<pre><code>   user> (load "protoproblem/proto")
   nil
   user> (load "protoproblem/impl")
   nil
   user> (protoproblem.proto/one (protoproblem.impl.ArraySteps. [1 2]))
   1
</code></pre>

Now, we redefine the protocol by commenting out the 'one' method def in proto.clj, and then reloading it:
<pre><code>   user=> (load "protoproblem/proto")
   nil
   user=> (protoproblem.proto/one (protoproblem.impl.ArraySteps. [1 2]))
   java.lang.NullPointerException (NO_SOURCE_FILE:9)
   user=> (load "protoproblem/impl")
   java.lang.IllegalArgumentException: Can't define method not in interfaces: one (impl.clj:5)
</code></pre>

That's pretty much as expected. But what if we try to get back to how things were, by reversing the changes in proto.clj, and reloading? This is where the problems arise:
<pre><code>   user=> (load "protoproblem/proto")
   nil
   user=> (load "protoproblem/impl")
   nil
   user=> (protoproblem.proto/one (protoproblem.impl.ArraySteps. [1 2]))
   java.lang.IllegalArgumentException: No implementation of method: :one of protocol: #'protoproblem.proto/Steps found for class: protoproblem.impl.ArraySteps (NO_SOURCE_FILE:0)

Ouch!

At this point, I have to restart my JVM to proceed, which is making incremental development of protocols-based code unusually tedious.

I initially encountered the behaviour using "C-c C-k" under slime/swank-clojure, but the above transcript is from a plain command-line repl, using a Clojure 1.2 snapshot from today. (The same thing happened with a snapshot from 2 weeks ago.)

When using "reify" inside a factory function instead of the constructor for a "deftype", everything works as expected.

[CLJ-352] clear macro meta when binding root of a var

Tue, 24 Aug 2010 08:11:00 -0500

clear macro meta when binding root of a var

[CLJ-351] Disable transients after persistent! call

Tue, 24 Aug 2010 08:10:00 -0500

Disable transients after persistent! call

[CLJ-349] propagate metadata on protocol fns

Tue, 24 Aug 2010 08:05:00 -0500

return type hints, arg lists, docstrings

[CLJ-347] Extend pprint to correctly handle data inside ref types

Tue, 24 Aug 2010 08:01:00 -0500

It's important that pprint can "see inside" all ref types so that it can format Clojure data inside them, This patch adds that, some more tests, and a little change to conform to the coding guidelines.

This is the last planned update to pprint for 1.2. Only approved bugs will be fixed from here on out.

[CLJ-344] reduce docstring args don't match docs

Tue, 24 Aug 2010 08:58:00 -0500

The reduce docstring says lists the arg forms as ([f coll] [f start coll]) but the docstring uses "val" instead of "start" in the description. To fix, replace all "val" with "start".

clojure.core/reduce
([f coll] [f start coll])
f should be a function of 2 arguments. If val is not supplied,
returns the result of applying f to the first 2 items in coll, then
applying f to that result and the 3rd item, etc. If coll contains no
items, f must accept no arguments as well, and reduce returns the
result of calling f with no arguments. If coll has only 1 item, it
is returned and f is not called. If val is supplied, returns the
result of applying f to val and the first item in coll, then
applying f to that result and the 2nd item, etc. If coll contains no
items, returns val and f is not called.

[CLJ-343] duck type RT.err usage

Tue, 24 Aug 2010 08:56:00 -0500

Some tools rebind err to PrintWriters, others (e.g. swank) to merely Writers. Work with all of them.

[CLJ-341] Vec test bugs.

Tue, 24 Aug 2010 11:54:00 -0500

There are a couple of bugs in the tests defined in clojure.test-clojure.vectors:

1. There are two definitions of test-vec (from different patches), so the second overrides the first.

2. The rand-replace function for the compare tests should use (rand-int 99) instead of (rand-int 100). Currently, if 99 is chosen, it will append to the end of the sequence instead of replacing the last element.

[CLJ-340] reify doc string examples are missing the 'this' argument

Tue, 24 Aug 2010 09:53:00 -0500

The reify doc string correctly discusses the need for a "this" argument, but fails to put it in the example usage it gives.

IIUC, the defprotocol examples are also missing the "this" argument.

[CLJ-338] promote c.c.set subset? and superset? to c.set

Tue, 24 Aug 2010 10:23:00 -0500

promote c.c.set subset? and superset? to c.set

[CLJ-337] mark deprecations

Tue, 24 Aug 2010 10:52:00 -0500

Last chance to care: Should the values for the deprecated tag be:

'true' only?
a version number, just like the values of :added (Tom's original proposal)?
free text, allowing for suggestions and commentary

I (Stu) am ok with either of the first two.

[CLJ-335] integrate pprint docs into clojure.org (?)

Tue, 24 Aug 2010 08:52:00 -0500

integrate pprint docs into clojure.org (?)

[CLJ-334] double-colon with nonexistent alias throws NPE

Tue, 24 Aug 2010 08:52:00 -0500

double-colon with nonexistent alias throws NPE

Clojure 1.1.0
user=> ::a/dig
java.lang.NullPointerException

[CLJ-333] protocol redef NPE

Tue, 24 Aug 2010 08:52:00 -0500

outgrowth of Mike's tests in #239.

[CLJ-332] last var in wins

Tue, 24 Aug 2010 07:51:00 -0500

not sure we should import source-fn and dir-fn into repl

[CLJ-331] Fix PersistentQueue equality methods

Tue, 24 Aug 2010 10:51:00 -0500

(= (into clojure.lang.PersistentQueue/EMPTY (range 5)) (into clojure.lang.PersistentQueue/EMPTY (range 5)))
=> NPE

[CLJ-330] Declare wipes out existing var metadata

Tue, 24 Aug 2010 07:51:00 -0500

If you call declare on a var that has already been defined, it wipes out the metadata on that var. While this is possible to avoid in most cases, it makes like difficult for automatic code generators or scanners. In particular, autodoc loses information in some cases because of this if it scans a file that has already been pulled in because of a dependency.

Here's an example. This code:

(defn hello "I like documentation!" [] (println "works"))
(hello)
(println (:doc (meta #'hello)))

(println "I do declare!")
(declare hello)

(hello)
(println (:doc (meta #'hello)))

Prints:

works
I like documentation!
I do declare!
works
nil

The last line should be "I like documentation!" again.

[CLJ-329] Fix defrecord keyword lookup on fields matching no-arg methods

Tue, 24 Aug 2010 10:50:00 -0500

Current behaviour:

user=> (defrecord Foo [size seq])
user.Foo
user=> (:size (Foo. 42 :b))
2
user=> (:seq (Foo. 42 :b))
([:size 42]
 [:seq :b])

Patch attached, with a test. Thoughts:

1. filtering on base-fields would probably be cleaner, but it looks like (class ~'gtarget) isn't fully formed at that point?
2. this results in a .valAt lookup for any fields that collide with no-arg methods implemented by the defrecord's class. That's unavoidable AFAICT.
3. The best solution would be to disambiguate field access vs. no-arg method invocation, i.e. have a variant of (. obj field) that would be specifically for field access (which is very occasionally needed elsewhere as well). This would nullify # 2.

[CLJ-328] don't assume chunked seq will alway stay chunked

Tue, 24 Aug 2010 10:49:00 -0500

optimized reduce always needs to check if underlying type can change as it walks the seq

[CLJ-327] core fns cannot use literal Patterns

Tue, 24 Aug 2010 11:49:00 -0500

Discussed here: http://groups.google.com/group/clojure-dev/browse_thread/thread/6c6c5d8ced3c6bea

Before pr-on is ready to use, code using pattern literals fails on initialization because they weren't printed correctly by RT/print at compile-time.
This actually give such messages:

#"\." is a Character and cannot be cast to Pattern
#"[.]" is a PersistentVector and cannot be cast to Pattern

The current patch is simpler than print-method for Pattern because it only tries to print readably Patterns created by RegexReader and not any user-created Pattern.

For the record print-method for Pattern is buggy in trying to support \Q and \E:
Here is a pathological case:
user=> (re-pattern "\\Q\\\\E\"")
#"\Q\\E\E\"\Q" ; should be #"\Q\E\\\Q\E\""
There's an obvious quick fix for that but I doubt one can achieve proper printing of all Patterns

[CLJ-325] mark :added for 1.0 vars

Tue, 24 Aug 2010 07:49:00 -0500

Followup to #321 but for all the original 1.0 vars.

[CLJ-324] 1.2 release notes

Tue, 24 Aug 2010 07:49:00 -0500

The 1.1 release notes were very nice. We need someone to do the same for 1.2. Basically this means

walking through the git commit messages (and maybe some diffs)
walking through the assembla tickets

and producing a text file similar to the 1.1 change file.

[CLJ-323] get annotation tests working regardless of jdk 5/later

Tue, 24 Aug 2010 07:48:00 -0500

get annotation tests working regardless of jdk 5/later

[CLJ-321] mark when vars got added to clojure

Tue, 24 Aug 2010 10:17:00 -0500

Third patch is the one. Adds metadata after docstring, public doc-ed vars only.

[CLJ-318] add annotation support to gen-class

Tue, 24 Aug 2010 15:44:00 -0500

Just call the same code used by definterface/type/record

[CLJ-317] clojure.main/repl isn't quite defensive enough

Tue, 24 Aug 2010 07:43:00 -0500

Because of an unguarded call to =, bad objects that get printed by the repl can cause the repl to crash if their equals method throws an exception.

For example:
java -cp clojure.jar clojure.main
user> (proxy [Object] [] (equals [o] (.toString nil)))

[CLJ-316] tests for annotation support

Tue, 24 Aug 2010 07:43:00 -0500

the hard part is extracting data from the annotations. Some variant of annotation->map might be of general use.

One nitpick with the API: If you said something like {Deprecated false} weirdness would ensue. Maybe barf on false/nil?

[CLJ-314] Vec contains AbstractMethodError

Tue, 24 Aug 2010 07:42:00 -0500

user=> (def vi (into (vector-of :int) [1 2 3]))
#'user/vi
user=> (contains? vi 1)
java.lang.AbstractMethodError (NO_SOURCE_FILE:0)

Vec is does not implement the containsKey or entryAt methods of clojure.lang.Associative.

[CLJ-313] add java.util.Map support to defrecord

Tue, 24 Aug 2010 10:42:00 -0500

add java.util.Map support to defrecord

[CLJ-312] audit & promote some clojure.contrib.seq fns

Tue, 24 Aug 2010 07:41:00 -0500

Proposed for promotion to clojure.core:

flatten, separate, indexed, group-by, partition-by, frequencies, reductions, positions

[CLJ-311] clojure.java.io namespace

Tue, 24 Aug 2010 07:41:00 -0500

audit and promote clojure.contrib.io to clojure.java.io.

all fns must be tested
fns that require consumption before closing are not idiomatic and will not be brought in (e.g. read-lines)
minimal set of fns is spit, input-stream, output-stream, reader, writer. others are welcome too

changes from the contrib version:

no dynamic vars magic (default-encoding, buffer-size, append all gone)
dynamic vars replaced by pervasive options arg (rest arg in public API)
protocol is named IOFactory (Streams mean something else)
protocol is not the public interface
use specific Javaland exceptions as appropriate, e.g. IllegalArgumentException
will not promote functions that magically close after consuming (e.g. write-lines)
no earmuffs on byte-array-type et al (these should never be rebound)
combine as-file and as-url into Coercions protocol

[CLJ-310] clojure.repl namespace

Tue, 24 Aug 2010 07:41:00 -0500

promote the following fns

clojure.contrib.repl-utils/source to clojure.repl/source
clojure.contrib.repl-utils/apropos to clojure.repl/apropos
clojure.contrib.ns-utils/dir to clojure.repl/dir

Also:

promote any necessary helper fns
make sure helper fns are private
promote and or write unit tests
make the clojure repl use these four fns

[CLJ-309] Reloading a multimethod declaration strips off the metadata

Tue, 24 Aug 2010 15:40:00 -0500

With commit 1b8d50, running past a defmulti twice (e.g., reloading),
causes the metadata to get squashed:

(defmulti add
"Add item to the accumulator acc. The exact meaning of adding an
an item depends on the type of the accumulator.")

(doc add)

yields =>

-------------------------
user/add
nil
nil

To standard out. I assume that this is because of the def on line
1191:

`(let v# (def ~mm-name)
(when-not (and (.hasRoot v#) (instance? clojure.lang.MultiFn
(deref v#)))
(def ~(with-meta mm-name m)
(new clojure.lang.MultiFn ~(name mm-name) ~dispatch-fn
~default ~hierarchy)))))))

This is mucking up the autodoc stuff.

[CLJ-307] consistent deprecation strategy

Tue, 24 Aug 2010 10:39:00 -0500

I am filing this as a Clojure ticket even though the problem currently occurs mostly in Contrib. That's because I think there needs to be an official "Clojure way" – this ticket is a request for documentation, not implementation.

Absent official guidelines, function and namespace deprecation has been handled in an ad hoc fashion. Sometimes functions move from one namespace to another, leaving no trace. Other times, something is left in the old namespace: either an error or a warning redirecting to the new namespace.

I would like to see a recommendation that leaves some kind of visible trace (for at least one point release) both in the source code tree and in the api docs.

Fallback namespaces (http://groups.google.com/group/clojure-dev/browse_thread/thread/9fde83ce4c06ed77) are out of scope for this ticket, but might be a separate ticket that builds on whatever is decided here.

[CLJ-306] definterface does not convert - to _

Tue, 24 Aug 2010 10:38:00 -0500

Hit this issue while writing a test for #104. The test paths include "test-clojure", and my generated interfaces are going to "test-clojure" while everything else goes to "test_clojure".

This conversion ought to be encapsulated in a fn instead of having str replacement all over the place.

[CLJ-303] re-import changed deftype should work at REPL

Tue, 24 Aug 2010 10:38:00 -0500

Current behavior:

(require :reload 'foo)
-> nil

(import foo.Bar)
-> foo.Bar

; make a change to Bar!

(require :reload 'foo)
-> nil

(import foo.Bar)
-> java.lang.IllegalStateException: Bar already refers to: class foo.Bar in namespace: user (NO_SOURCE_FILE:9)

[CLJ-302] Prefer more derived interface in protocol

Tue, 24 Aug 2010 10:33:00 -0500

Currently if a protocol is extended to 2 interfaces with a derivation relationship, the more derived is not consistently chosen.

[CLJ-301] (pr x & more) makes a self-call, which can lead to stack overflow

Tue, 24 Aug 2010 15:33:00 -0500

When 'pr', or any related functions are called with a sequence of arguments, they can cause stack overflow.

Eg:

(apply pr (range 5000))

[CLJ-299] clojure.main -e disables stdin

Tue, 24 Aug 2010 16:33:00 -0500

As mentioned on the mailing list, using the -e argument of clojure.main rebinds in and makes stdin unavailable to other code:

http://groups.google.com/group/clojure/browse_thread/thread/fbb36bea2034bc7a/197fd51798de01e7?lnk=gst&q=stdin#197fd51798de01e7

[CLJ-298] correct accessors for VecNode fields

Tue, 24 Aug 2010 10:33:00 -0500

Accessors :arr and :edit should be .arr and .edit.

Holding InternalReduce (#289) due to test failures until this is approved

[CLJ-297] gvec comparison AbstractMethodError

Tue, 24 Aug 2010 17:33:00 -0500

(def iv (into (vector-of :int) [1 2 3]))
->#'user/iv

(= iv (range 1 4))
-> java.lang.AbstractMethodError (NO_SOURCE_FILE:0)

[CLJ-296] tests for case

Tue, 24 Aug 2010 10:32:00 -0500

we need tests for case

[CLJ-295] disambiguate protocol extended to two interfaces

Tue, 24 Aug 2010 10:32:00 -0500

If a protocol is extended to two interfaces, and an object implements both, it will get one of them at random. I'd be happy if this case was simply an error.

[CLJ-294] extend should check protocol is not implemented directly

Tue, 24 Aug 2010 10:32:00 -0500

The extend function should check and throw an exception if you try to extend a protocol into a class or type that already implements the protocol directly. The current behavior in this case is unintuitive: the extend succeeds, and a method is added to the :impls map, but you can never call it.

[CLJ-290] bigint truncates large ratios

Tue, 24 Aug 2010 10:16:00 -0500

Clojure 1.2.0-master-SNAPSHOT
user=> (bigint (/ 1000000000000000000000000000 3))
9223372036854775807
</code></pre>
The correct result should be:
<pre><code>333333333333333333333333333

This is caused by clojure.core/bigint first converting the ratio to a long:
<pre><code>(defn bigint
"Coerce to BigInteger"
{:tag BigInteger}
[x] (cond
(instance? BigInteger x) x
(decimal? x) (.toBigInteger #^BigDecimal x)
(number? x) (BigInteger/valueOf (long x))
:else (BigInteger. x)))</code</pre>

[CLJ-289] Add InternalReduce

Tue, 24 Aug 2010 17:14:00 -0500

InternalReduce protocol lets seqs take advantage of structural details, make reduce run faster.

[CLJ-281] Make more datastructures serializable

Tue, 24 Aug 2010 08:51:00 -0500

Make several more core data structures serializable via java-serialization.

The attached patch is a minimal patch- serialVersionUIDs are not created, nor are redundant "implements Serializable" statements added to classes who already implement Serializable from some other interface. This patch also does not make functions, lazy lists, atoms, references, etc. serializable, nor those data structures that depend upon any of the above (so tree maps and tree sets, which depend upon an ordering function, are not serializable, for example). Only those data structures which can be reasonably expected to be serialized by one program and deserialized by another are made serializable. Whether any of these extra functions should be made serializable is a different debate.

The classes made serializable are:
PersistentVector$Node (PersistentVector is already serializable)
PersistentQueue
PersistentList
PersistentList$EmptyList

[CLJ-275] division by negative ratio resulting in a whole number returns a ratio when an integer is expected

Tue, 24 Aug 2010 07:33:00 -0500

i.e.

user=> (/ 1 -1/2)
-2/1
user=> (/ 1 1/2)
2 
user=> (/ 0 -1/2)
0/1

This happens since the check for the denominator being 1 is done before the negative sign is switched from the denominator to the numerator.
i.e. http://github.com/richhickey/clojure/blob/master/src/jvm/clojure/lang/Numbers.java#L296

I have provided a patch which makes this work as I expected, of course this may not be the optimal way of solving this problem as i'm not familiar enough with the internals of clojure.

Thanks
.Tristan

[CLJ-271] Determine direct binding policy and controls

Tue, 24 Aug 2010 06:25:00 -0500

Direct binding is in place as an experiment. We need ways to control the application and granularity of direct binding.

[CLJ-269] Add List and Collection support to Vec

Tue, 24 Aug 2010 15:22:00 -0500

Add List and Collection support to Vec

[CLJ-268] Add Iterable support to Vec

Tue, 24 Aug 2010 07:22:00 -0500

Add Iterable support to Vec

[CLJ-267] Add Reversible support to Vec

Tue, 24 Aug 2010 15:22:00 -0500

Add Reversible support to Vec

[CLJ-266] Add Comparable support to Vec

Tue, 24 Aug 2010 06:22:00 -0500

Add Comparable support to Vec

[CLJ-265] Add IFn support to Vec

Tue, 24 Aug 2010 06:22:00 -0500

Add IFn support to Vec

[CLJ-264] Complete interface impls for generic vector

Tue, 24 Aug 2010 06:22:00 -0500

Vec needs several more interface implementations in order to be as complete as persistentvector.
See subtasks

[CLJ-262] (resolve 'FooBar.) should return nil

Tue, 24 Aug 2010 06:19:00 -0500

Currently, (resolve 'FooBar.) throws a ClassNotFoundException it should return nil.

[CLJ-260] Cannot load clojure classes from Jar files outside classpath using URLClassLoader

Tue, 24 Aug 2010 15:18:00 -0500

It is not possible to use URLClassLoader to load a class from a jar file if the jar is outside the classpath. It seems that RT.load tries to search for *.clj file within the classpath and doesn't respect the URLClassLoader and then fails with:

Exception in thread "main" java.lang.ExceptionInInitializerError
at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)
at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:39)
at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:27)
at java.lang.reflect.Constructor.newInstance(Constructor.java:513)
at java.lang.Class.newInstance0(Class.java:355)
at java.lang.Class.newInstance(Class.java:308)
at Loader.main(Loader.java:17)
Caused by: java.io.FileNotFoundException: Could not locate net/experimentalworks/Foo__init.class or net/experimentalworks/Foo.clj on classpath:
at clojure.lang.RT.load(RT.java:402)
at clojure.lang.RT.load(RT.java:371)
at clojure.core$load_6449$fn_6458.invoke(core.clj:4171)
at clojure.core$load__6449.doInvoke(core.clj:4170)
at clojure.lang.RestFn.invoke(RestFn.java:413)
at clojure.lang.Var.invoke(Var.java:359)
at net.experimentalworks.Foo.<clinit>(Unknown Source)
... 7 more

The attached .tar.gz contains a sample jar file and the clojure file that was used to generate the class files as well as the sample loader. Java oder Scala written classes can be loaded without any problems. For sure the I put the clojure runtime into the classpath but not the loadme-clojure.jar

[CLJ-258] can't recur from case special form

Tue, 24 Aug 2010 17:12:00 -0500

the 'case' special form doesnt't propagate tailcontext.
eg.

; SLIME 2010-02-01
user> (loop [foo "foo"]
        (case foo
          "foo" (recur 'bar)
          'bar (recur :baz)
          :done (do (println "hello"))
          (recur :done)))
; Evaluation aborted.

Can only recur from tail position
  [Thrown class java.lang.UnsupportedOperationException]

This can be easily fixed by making
Compiler.CaseExpr#emit() and Compiler.CaseExpr.Parser#parse() propagate their context instead stetting the context to C.EXPRESSION.

[CLJ-257] add fnil for wrapping functions to handle nil

Tue, 24 Aug 2010 17:07:00 -0500

As per discussion
http://groups.google.com/group/clojure/browse_thread/thread/99cc4a6bfe665a6e
fnil provides a convenient way to supply a default value to be used instead of nil for functions that expect values

[CLJ-256] get-in optional default value

Tue, 24 Aug 2010 15:07:00 -0500

As per discussion
http://groups.google.com/group/clojure/browse_thread/thread/99cc4a6bfe665a6e
adding optional default value to be like get

[CLJ-254] Expose c.l.Ratio's accessors

Tue, 24 Aug 2010 15:07:00 -0500

Expose accessor functions of c.l.Ratio in c.core.

See here:
http://groups.google.com/group/clojure/browse_thread/thread/775351b9335a561e

[CLJ-253] Improve errors for "use"/"require" in REPL

Tue, 24 Aug 2010 06:07:00 -0500

in the repl when i (use 'foo) vs. (use :foo) they both return nil, so it doesn't help me know that the former is right and the latter isn't. would it be sensible for use to return true or something more positive than nil if it successfully brings in a namespace? or print out an error message when symbols are used?

[CLJ-247] Remove clojure.lang.Stream/Streamable and all uses

Tue, 24 Aug 2010 08:59:00 -0500

Remove clojure.lang.Stream/Streamable and all uses

[CLJ-246] direct binding inhibits clojure.test functionality

Tue, 24 Aug 2010 15:53:00 -0500

clojure.test offers features that rely on binding functions from its namespace. With the way direct binding is implemented now, this is impossible, since any namespaces beginning with "clojure" are directly bound. This also has a (probably) unintended of direct binding namespaces that just start with "clojure", like "clojuresque" and "clojure-http-client."

I propose that direct binding use a whitelist, instead of direct binding anything that starts with "clojure."

An ever better solution, though, would be to use namespace metadata to turn off/on direct binding. That way user code could turn on direct binding in hotspots. Unfortunately, there's an open bug (#130) which causes AOT compilation to erase namespace metadata, so until that's closed I would prefer the whitelist option.

[CLJ-245] Unnamed fns should have no binding of name

Tue, 24 Aug 2010 15:52:00 -0500

Currently a binding is created for a generated name

[CLJ-243] Add bound? and thread-bound? predicates

Tue, 24 Aug 2010 07:51:00 -0500

Discussed in #clojure here: http://clojure-log.n01se.net/date/2010-01-12.html#i96

Wraps Var.isBound and the truthiness of Var.getThreadBinding; the latter particularly handy for knowing whether or not set! will fail or not.

[CLJ-242] Support implicit macro args for form and locals

Tue, 24 Aug 2010 07:51:00 -0500

Inside the body of a defmacro expander, &whole will refer to the original macro invocation form (thus allowing recovery and use of metadata on that form by the expander itself). &env will refer to a map of symbols to TBD, such symbols representing the locals in scope. This will break code that depends on defmacro creating a fn with exactly the same signature.

Possible alternate names: &locals, &form

[CLJ-241] Tests for reify

Tue, 24 Aug 2010 15:51:00 -0500

subsumed by #239 and #240

[CLJ-240] Tests for protocols and types, round 1

Tue, 24 Aug 2010 05:51:00 -0500

I have included a patch with my original tests (pre deftype/defrecord split) plus a final patch to get things passing with the latest greatest.

The final commit is worth looking at. The reduced dynamism of deftype is visible, in that I used to be able to redefine a type multiple times in a series of tests, and am now relying on per-test named types. I think this tradeoff is worth it for unification with the host type system. But a bigger issue occurs when working at the REPL: you can reload types, but not refer to the reloaded ones.

[CLJ-239] Tests for types and protocols, round 2

Tue, 24 Aug 2010 05:51:00 -0500

Tests for types and protocols, round 2

[CLJ-232] Locals cleared too aggressively on delay

Tue, 24 Aug 2010 05:43:00 -0500

Steve Gilardi and I noticed some strange behaviour with local clearing in conjunction with delay:

(defn do-something [x]
(throw (Exception.)))

(defn clear-locals [x]
(delay (try (do-something x)
(catch Exception e
(println "argument value:" x)))))

(force (clear-locals :argument))
;; => argument value: nil

It seems the locals are getting cleared a little too aggressively here. I did some digging to discover that delay is a thin wrapper around an fn, but using an fn in this context (rather than a delay) does not trigger the problem. Also I found that throwing an exception directly inside the try block (rather than in a function called from the try block) did not trigger it.

I was able to reproduce in the new branch, the master branch, and 1.0.0, so it's not related to the more recent locals-clearing changes.

[CLJ-231] deftype cons doesn't support maps

Tue, 24 Aug 2010 14:43:00 -0500

The deftype IPersistentMap implementation of cons does not have the same functionality as APersistentMap.cons. The current implementation doesn't support cons'ing a map.

See this thread for the original report:
http://groups.google.com/group/clojure/browse_thread/thread/d4dd1a0a1ffee6c5#

[CLJ-230] More precise docstring for #'delay

Tue, 24 Aug 2010 14:40:00 -0500

Change docstring for #'delay to precise deref forces delays too.
See http://groups.google.com/group/clojure-dev/browse_thread/thread/d835fd49d08e48ca

[CLJ-222] line-seq returns element only when successive line is read

Tue, 24 Aug 2010 05:34:00 -0500

The lazy-seq</code> in <code>line-seq</code> wraps the call to <code>cons</code> but not the <code>readLine(), which delays each line of the reader from being returned until the next one is read.

http://groups.google.com/group/clojure/browse_thread/thread/e0d6efd53187ea09

[CLJ-220] Adding more test cases for clojure set

Tue, 24 Aug 2010 08:31:00 -0500

Expanded the test cases to also test
difference
select
project
rename-keys
index
map-invert
join

Attached is the patch file.

[CLJ-204] clojure.test/thrown-with-msg? uses re-match, should use re-find

Tue, 24 Aug 2010 04:54:00 -0500

By the time exceptions percolate up to clojure.test, they are often decorated by additional levels of RuntimeException, thus the message being matched is usually "java.lang.RuntimeException: My error message". I end up always having to prefix my re with ".*". If I want to match the entire message, RE's allow that using ^.

[CLJ-198] attr-map argument for clojure.core/ns

Tue, 24 Aug 2010 04:51:00 -0500

Discussion: http://groups.google.com/group/clojure-dev/browse_thread/thread/97ffdb1d89c0f847

The clojure.core/ns macro should have an attr-map argument like defn and defmulti. For example:

(ns example
  {:author "John Doe"
   :email "jdoe@example.com"}
  (:use clojure.contrib.def))

[CLJ-162] gen-class fails to accept type hints for arrays of primitives

Tue, 24 Aug 2010 04:03:00 -0500

gen-class fails to accept type hints for arrays of primitives like "[B"

This is a bug in genclass.clj's the-class, which should have (some #{\. \[} strx) instead of (some #{\.} strx)

[CLJ-145] clojure.set/rename-keys doesn't do what it says it should

Tue, 24 Aug 2010 06:52:00 -0500

user=> (doc rename-keys)
-------------------------
clojure.set/rename-keys
([map kmap])
Returns the map with the keys in kmap renamed to the vals in kmap
nil

I expect the return value to be a map of the same size as the first input, with its keys being those in the first input other than those which are substituted according to the second. That is not what occurs:

user=> (rename-keys {:foo 1 :bar 2} {:foo :new-foo :zob :new-zob})
{:new-zob nil, :new-foo 1, :bar 2}

Tested against a1397390d8b3b63f2039359520629d87b152d717 (current at time of filing).

[CLJ-119] GC Issue 115: Document that #^chars works

Tue, 24 Aug 2010 03:45:00 -0500

Reported by hlship, May 05, 2009

In the Java Interop documentation, it identifies a number of special tags
that can be used to identify parameters of primitive arrays.

It does not list #^chars (equivalent to char[]) but this appears to (at
least partially) work.

On a related note, for this code:

(def #^StringBuilder *buffer*)

(defn- add-text
  "Adds text to the buffer and manages the *text-location* var."
  [#^chars ch start length]
  (.append *buffer* ch start length)  ;;;; line 98
  (when (nil? *text-location*)
        (set! *text-location* (current-location))))


I still get 

Reflection warning,
com/howardlewisship/cascade/internal/xmltokenizer.clj:98 - call to append
can't be resolved.

[CLJ-104] GC Issue 100: gen-class creating non-compliant field names

Tue, 24 Aug 2010 03:45:00 -0500

Reported by solkis, Apr 04, 2009

The defect was described in this clojure group post by Michael Reid: 

http://groups.google.com/group/clojure/browse_thread/thread/e64719d716c29ce0

It appears that IBM JVMs have more stringent field name checking than do
the Sun JVMs.

Classes generated by gen-class and executing on IBM JVMs used with the
WebSphere Application Server require the basic fix outlined in the patch on
the thread above or the following error will result:

java.lang.ClassFormatError: JVMCFRE114 field name is invalid;

[CLJ-96] GC Issue 92: Check for key identity in collections

Tue, 24 Aug 2010 06:45:00 -0500

Reported by chouser, Mar 09, 2009

Describe the feature/change.

Clojure code routes key lookups through
clojure.lang.Util.equiv()/equals(), which do an identity check.

Using the collections from outside Clojure may to more expensive
comparisons even on identical objects, so each collection should also
do an identity check.

Was this discussed on the group? If so, please provide a link to the
discussion:

http://groups.google.com/group/clojure/msg/6a2e491fcf564a68

[CLJ-94] GC Issue 90: Support metadata on fns

Tue, 24 Aug 2010 06:45:00 -0500

Reported by richhickey, Mar 04, 2009

Requires dealing with the with-meta copying issues for closures

[CLJ-30] GC Issue 26: agent error queue

Tue, 24 Aug 2010 03:44:00 -0500

Reported by richhickey, Jan 01, 2009
Route agent errors to a queue

Comment 1 by chouser, May 26, 2009
The attached patch allows agents to have an error-callback, which could be used to
route errors to one or more queues.

Example usage:

user=> (send-off (agent nil
                   :error-callback (fn [a es]
                                     (.printStackTrace (first es))
                                     (clear-agent-errors a)))
                 (fn f [_]
                   (Thread/sleep 1000)
                   (send-off *agent* f)))
#<Agent@5dd2b9b7: nil>
user=> (shutdown-agents)
nil
user=>
java.lang.RuntimeException: Error executing action user$eval__394$f__399@63d87b85
        at clojure.lang.Agent$Action.execute(Agent.java:63)
        at clojure.lang.Agent$Action.doRun(Agent.java:107)
        at clojure.lang.Agent$Action.run(Agent.java:118)
        at
java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:886)
        at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:908)
        at java.lang.Thread.run(Thread.java:619)
Caused by: java.util.concurrent.RejectedExecutionException
        at
java.util.concurrent.ThreadPoolExecutor$AbortPolicy.rejectedExecution(ThreadPoolExecutor.java:1760)
        at java.util.concurrent.ThreadPoolExecutor.reject(ThreadPoolExecutor.java:767)
        at java.util.concurrent.ThreadPoolExecutor.execute(ThreadPoolExecutor.java:658)
        at clojure.lang.Agent$Action.execute(Agent.java:57)
        ... 5 more

 error-callback.patch
3.9 KB Download

[CLJ-20] GC Issue 16: Pretty printing

Tue, 24 Aug 2010 06:44:00 -0500

Reported by richhickey, Dec 17, 2008
A rudimentary pprint
Some efforts have been posted to group, no consensus yet