Reported by m...@kotka.de, Dec 17, 2008
Currently import expects lists of a packages with classes, which should be imported
from the given package.

(import '(package ClassA ClassB ...) ...)

Sometimes however one just wants to import a single class from a package, but
one still has to specify the list form:

(import '(package ClassA))

use/require also allow a similar style of prefix list. However, they also allow a
shortcut in case one requires only one lib with a given prefix. The attached
patch modifies import to allow also this style of shortcut:

(import 'package.ClassA)

This change would be backward compatible.


 modified-import.patch
1.3 KB Download
Comment 1 by m...@kotka.de, Dec 17, 2008
I'm sorry for the dupe. The proxy gave me an error message, so I submitted again...
Also sorry for the wrong categorisation. Should be Enhancement not Defect...
Comment 2 by richhickey, Dec 17, 2008
(No comment was entered for this change.)

Comment 3 by richhickey, Dec 17, 2008
 Issue 4  has been merged into this issue.
Comment 4 by richhickey, Dec 17, 2008
(No comment was entered for this change.)
Owner: richhickey
Comment 5 by richhickey, Dec 17, 2008
I think the conditional here should test for symbol? rather than seq?, i.e. reverse
the condition. We should avoid hardwiring lists when possible.
Comment 6 by m...@kotka.de, Dec 19, 2008
Ok. Checks for symbol? now and if branches reversed. 
 modified-import.patch
1.3 KB Download
Comment 7 by richhickey, Dec 21, 2008
patch applied (svn 1179) - thanks!
Status: Done

Reported by richhickey, Dec 14, 2008
switch to keyword args for reference ctors:
(ref val & options)
(atom va & options)
(agent val & options)

Comment 1 by richhickey, Dec 17, 2008
(No comment was entered for this change.)
Labels: -Priority-Medium Priority-High
Comment 2 by richhickey, Dec 17, 2008
(No comment was entered for this change.)
Status: Accepted
Comment 3 by richhickey, Dec 17, 2008
(No comment was entered for this change.)
Owner: ---
Comment 4 by richhickey, Jan 05, 2009
svn 1193
Status: Fixed

Reported by richhickey, Dec 17, 2008
rather than positional arg

Comment 1 by richhickey, Dec 17, 2008
(No comment was entered for this change.)
Status: Accepted
Comment 2 by richhickey, Dec 17, 2008
(No comment was entered for this change.)
Owner: ---
Comment 3 by m...@kotka.de, Jan 05, 2009
Attached is a patch, which turns the default dispatch value from a positional argument
into a keyword argument. The syntax is:

(defmulti name dispatch-fn :default dispatch-value)

Uses of the old syntax throw an exception pointing to the new syntax.
 issue-8.diff
1.4 KB Download
Comment 4 by m...@kotka.de, Jan 11, 2009
Another update of the patch, also including docstring and metadata support, similar to defn.

This was brought up several times on the list. Eg. in the thread:

http://groups.google.com/group/clojure/tree/browse_frm/thread/21af6e0431ae2e63/f7751913e875953f?
rnum=11&_done=%2Fgroup%2Fclojure%2Fbrowse_frm%2Fthread%2F21af6e0431ae2e63%2Ff7751913e875953f%
3F#doc_b5a6f9049a2ff250
 issue-8.diff
2.5 KB Download
Comment 5 by richhickey, Jan 22, 2009
Patch applied - svn 1226 - thanks!
Status: Fixed

Reported by richhickey, Dec 17, 2008
Need to audit and fix

Comment 1 by richhickey, Dec 17, 2008
(No comment was entered for this change.)
Owner: ---
Comment 2 by mmcgrana, Jan 05, 2009
Possible audit method for public clojure.core *vars missing doc strings:

(prn
  (let [publics (ns-publics 'clojure.core)
        stared  (filter #(re-find #"^\*" (str (first %))) publics)
        undoced (remove #(:doc (meta (second %))) stared)]
    (map first undoced)))

=> (*compile-path* *ns* *allow-unresolved-vars* *macro-meta* ...
Comment 3 by chouser, Jan 09, 2009
For Vars interned via Java code, there's no original 'def' statement
where docstrings can be added.  Should a block of statments like
following be added to the end of clojure/core.clj ?

(alter-meta! #'*compile-path* assoc :doc
  "Specifies the directory where 'compile' will write out .class
  files.  This directory must be in the classpath for 'compile' to
  work.")

Comment 4 by richhickey, Jan 10, 2009
That seems right to me
Comment 5 by Chousuke, Jan 18, 2009
Here is a patch that adds at least *some* documentation for some of the variables.
It's missing a few that I did not know the purpose for, though.

--
Jarkko
 var-doc.patch
2.5 KB Download
Comment 6 by richhickey, Jan 22, 2009
Patch applied - svn 1225 - thanks!
Status: Fixed

Reported by richhickey, Dec 17, 2008
should turn into fully.qualified.Classname.

Comment 1 by m...@kotka.de, Jan 02, 2009
The attached patch fixes this issue. It checks for Symbol w/o namespace and a trailing dot.
Then the dot is removed, the Symbol resolved and the dot added again.

The second patch fixes a related issue with .methodName Symbols: those were also resolved
when in fact they should not have been. So the patch checks for Symbols w/o namespace
and a leading dot. Those Symbols are simply left alone and quoted as is.

The second patch should be applied after the first.
 issue-7.diff
785 bytes Download
 issue-7a.diff
611 bytes Download
Comment 2 by richhickey, Jan 07, 2009
Patches applied svn 1202 - thanks!
Status: Fixed

Reported by richhickey, Dec 14, 2008
right now it doesn't look at type of field, and always generates instance
methods

Comment 1 by richhickey, Dec 17, 2008
(No comment was entered for this change.)
Owner: ---
Comment 2 by richhickey, Dec 18, 2008
(No comment was entered for this change.)
Labels: Priority-High
Comment 3 by chouser, Mar 06, 2009
The attached patch was tested using this gen-class call:

  (ns clojure.contrib.test-clojure.gen-class
    (:gen-class
      :extends javax.swing.border.TitledBorder
      :exposes {TEXT_SPACING {:get getSpacing :set setSpacing}
                title {:get getString :set setString}}))

Then:

  (compile 'clojure.contrib.test-clojure.gen-class)
  (def x (clojure.contrib.test_clojure.gen_class. "hi"))

  (.getString x)  ==> "hi"
  (.setString x "yo")
  (.getString x)  ==> "yo"

  (.getSpacing x)
  ==> java.lang.IncompatibleClassChangeError: Expected non-static field
clojure.contrib.test_clojure.gen_class.TEXT_SPACING (NO_SOURCE_FILE:0)

With the patch applied, .getString and .setString work the same, but now:

  (clojure.contrib.test_clojure.gen_class/getSpacing)  ==> 2
  (clojure.contrib.test_clojure.gen_class/setSpacing 10)
  ==> java.lang.IllegalAccessError (NO_SOURCE_FILE:0)

This last error is because the static field is final.

--Chouser
 genclass-exposes-statics.patch
2.2 KB Download
Comment 4 by richhickey, Apr 12, 2009
Patch applied - r1345 - thanks!
Status: Fixed

Reported by m...@kotka.de, Dec 17, 2008
fcase from clojure.contrib is planned to be included in clojure.core
as condp. Currently there are two suggestions, on how to implement
this.

The attached patch proposes the following form:

(condp predicate expr
  test-expr result-expr
  ...)

For each pair test-expr and result-expr the predicate will
be called in the form (predicate test-expr result-expr). If
the predicate returns true, the result-expr is executed.

Using the vector binding form as known from let and friends
one might also assign the result of the call to predicate to
a local which may be used in the result-expr.

(condp [local predicate] expr
  test-expr result-expr
  ...)

If the number of test-expr result-expr pairs is odd, then
the last clause is evaluated, in case no previous clause was
chosen by the predicate. If the number of pairs is even,
an exception is thrown.

Furthermore the following changes to cond are proposed
to implement functionality similar to clojure.contrib.cond-let.

(cond
  [local test-expr] result-expr
  ...)

In case the test-expr in cond is a vector the first item is
used as the name of local which is bound to result of the
test-expr (second item in the vector) in result-expr.

This patch is backward-compatible.

Additionally another solution using templates was proposed
on the google group. More information on this proposal
may be found in the following thread:

http://groups.google.com/group/clojure/browse_thread/thread/19544014a89f24e9/dfdecc1bc
8330785?lnk=gst&q=stab+condp#dfdecc1bc8330785

 cond-condp.patch
2.9 KB Download
Comment 1 by richhickey, Dec 17, 2008
You meant: (predicate test-expr expr) right?

I think we should take any changes to cond off the table at present. 

It would be nicer if condp could do binding on a per-clause basis, as you did for
cond, but doing it the same way would rule out vectors as test exprs (with predicates
like nth, I guess).

The alternative is Scheme's:

test-expr => result-fn

where result-fn is a fn of one arg, passed the result of the predicate.

This can end up being more compact. I guess in Clojure we'd use a keyword:

test-expr :> result-fn

Comment 2 by m...@kotka.de, Dec 18, 2008
Modified the implementation. The syntax is now as follows:

(condp predicate expr
  test-expr result-expr
  test-expr :> result-expr
  ...)

In the first case result-expr is evaluated and its result is returned
if (predicate test-expr expr) returns non-nil/non-false.

In the second case, result-expr is expected to evaluate to a function
of one argument, which is passed the result of the predicate call.
Again: only if the result of the predicate call is non-nil/non-false.

In case there is a single clause at the end which does not fit the above
patterns it is taken as default in case no test-expr yields a success.
If no such default is given an exception is thrown.

The patterns may be freely mixed.

Cond is not modified.
 condp.diff
2.6 KB Download
Comment 3 by richhickey, Dec 21, 2008
In the end I found these patches too complex and wrote my own, thanks for the input!
svn 1180
Status: Done

Reported by richhickey, Dec 17, 2008
Need to audit and fix

Comment 1 by richhickey, Dec 17, 2008
(No comment was entered for this change.)
Owner: ---
Comment 2 by chouser, Jan 09, 2009
Currently:

user=> (java.util.HashMap. {:a 1})
#=(java.util.HashMap. {:a 1})

What's the desired output?  Perhaps the unreadable:

#<java.util.HashMap. {:a 1}>

Comment 3 by richhickey, Jan 10, 2009
Yes, the unreadable form is good
Comment 4 by chouser, Jan 10, 2009
All occurrences of #= are in RT.java and core_print.clj.

In core_print.clj, all occurrences of #= are in print-dup methods,
except for one occurrence in the print-ctor function.  print-ctor is
in turn used only in print-dup methods, except for 3 print-method's
for Java set, map, and collection types.  The attached patch removes
those print-method's, causing the default method to be used, which
generates informative output for humans but which is not readable by
LispReader.

In RT.java, #= is generated by the print() method when
PRINT_INITIALIZED is unbound or false, for Class and Var objects.
Changing these to print non-dup style output makes the Clojure compile
process throw exceptions, so I'm guess these need to stay the way they
are.

For the record, these outputs can be tested at the REPL like this:

(defmacro pre-init [x]
  `(binding [clojure.core/print-initialized false]
    (clojure.lang.RT/print ~x *out*)
    (println)))

user=> (pre-init Integer)
#=java.lang.Integer
nil
user=> (pre-init #'filter)
#=(var clojure.core/filter)
nil

--The janitor
 remove-hash-equals.patch
1.7 KB Download
Comment 5 by richhickey, Jan 14, 2009
Patch applied, svn 1209 - thanks!
Status: Fixed

Reported by richhickey, Dec 17, 2008
To enable applets and other situations not supporting custom classloaders
or dynamic classloading

Comment 1 by richhickey, Dec 17, 2008
(No comment was entered for this change.)
Labels: -Type-Defect Type-Enhancement
Comment 2 by richhickey, Jan 07, 2009
svn 1201
Status: Fixed

Reported by richhickey, Dec 17, 2008
Explain the behavior of contains? for vectors, i.e. associative

Comment 1 by richhickey, Dec 18, 2008
(No comment was entered for this change.)
Labels: Priority-High
Comment 2 by chouser, Jan 09, 2009
Expands both 'contains?' and 'some' docstrings.
 contains-docs.patch
1.3 KB Download
Comment 3 by richhickey, Jan 14, 2009
patch applied, svn 1214 - thanks!
Status: Fixed

Reported by tomfaulhaber, Dec 28, 2008
The clojure.core/bigdec function truncates rationals to ints:

(bigdec 1/2) => 0M

As opposed to:

(bigdec 0.5) => 0.5M

This happens because rationals fall into the number? case in bigdec which
converts the rational to a long before converting to BigDecimal.

I'm not sure that this behavior is completely wrong (since there's no way
to specify the behavior of 1/3 in the current API, for instance), but we
should probably at least document it.


Comment 1 by jbondeson, Jan 30, 2009
I discussed this along with another issue here:
http://groups.google.com/group/clojure/browse_thread/thread/1f8029b8556ce851/9b7538b192490097?lnk=gst&q=ratio#9b7538b192490097

(bigdec (/ 1 3)) should probably behave exactly like (/ 1M 3) which throws an
ArithmeticException unless you are inside a "with-precision" block.

I have attached a rather simple fix for this to have Ratio -> BigDecimal conversion
behave like this.
 bigdec-ratio.diff
521 bytes Download
Comment 2 by richhickey, Feb 09, 2009
fixed in svn 1257 - thanks!
Status: Fixed

Reported by richhickey, Dec 23, 2008
Agents in perpetual loops would like to send actions that don't wait for
current action completion

Comment 1 by richhickey, Dec 23, 2008
added in svn 1185
Status: Fixed

Reported by richhickey, Dec 28, 2008
Right now not all integers that can be represented as longs are, leading to
cases of different representations of integers of same value, and the
resulting problems for hashCode.

Comment 1 by christophe.grand, Dec 29, 2008
added LongOps 
uncommented the two lines in reduce(BigInteger)
 LongOps.patch
4.9 KB Download
Comment 2 by richhickey, Dec 29, 2008
Thanks! I'm having some problems applying the patch. I get a LongOps class with
several opsWith(LongOps) entries and other errors.

Would it be possible to produce a patch/diff with more context?
Comment 3 by christophe.grand, Dec 29, 2008
diff -u5 this time, hoping it works better for you than plain svn diff.
 LongOps.diff
5.6 KB Download
Comment 4 by richhickey, Dec 29, 2008
Patch applied - svn 1190 - thanks!
Status: Fixed

Reported by richhickey, Dec 29, 2008
On Android, class can be loadable without accessible .class file, so no
.class or .clj is not an error. Must try to load speculatively in that case.

Comment 1 by richhickey, Dec 29, 2008
svn 1188
Status: Fixed

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

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

Reported by olov.lassus, Jan 02, 2009
Negating Integer.MIN_VALUE overflows but should return a BigInteger.  
It also affects binary subtraction since Clojure implements it using  
negation and addition. The overflow occurs silently without throwing  
an exception.

These examples trigger the flaw:

user=> (- -2147483648)
-2147483648

user=> (- -1 -2147483648)
-2147483649


The attached patch apply cleanly against trunk r1195. Same examples after
applying it:

user=> (- -2147483648)
2147483648

user=> (- -1 -2147483648)
2147483647

 detect_negate_overflow.patch
507 bytes Download
Comment 1 by richhickey, Jan 07, 2009
fixed in svn 1203
Status: Fixed

Reported by va3pip, Dec 29, 2008
What (small set of) steps will reproduce the problem?
Compile and run a clojure based Android app under the emulator.  While
loading, it'll throw a clojure.core__init class not found exception. A good
sample clojure Android app is at: 
    http://github.com/Nafai77/helloandroid/tree/master

What is the expected output? What do you see instead?
It should be able to find clj based classes

What version are you using?
svn 1191

Please provide any additional information below.
I've attached a patch to RT.java that changes classForName() and
loadClassForName() to try the default classloader if DynamicClassLoader
fails to find the requested class.  I probably overlooked some corner case
with these 2 changes, so I'll expect you'll fix/change this patch :) - but
I'll send in a CA tomorrow anyways, just in case.  Note: it still won't run
the app as there's a VerifyError with StringBuilder.append() - for details,
see:   http://paste.lisp.org/display/72811

 dalvikvm.patch
791 bytes Download
Comment 1 by richhickey, Dec 30, 2008
svn 1192
Status: Fixed

Reported by richhickey, Dec 23, 2008
Currently yields an Unquote object
clojure/unquote would remain undefined, thus would still be error if
evaluated, but user macros could use for escaping

Comment 1 by richhickey, Dec 23, 2008
added svn 1184
Status: Fixed

Reported by dresweng...@dreish.org, Jan 06, 2009
Like any good Lisp, Clojure can compile and run code while reading, read and compile code while 
running, and read and run code while compiling.

Unfortunately that means it isn't safe for an application to use (read) to read something from an 
untrusted source. If the input stream contains, e.g., #=(eval (def core-app-function #(throw 
(Exception.)))), the application would blow up.

Feature request is for a version of the reader that has #= shut off, for reading untrusted data.

(Discussion was on #clojure; rhickey asked me to add this, so there's no Google Group discussion 
that I'm aware of.)
Comment 1 by christophe.grand, Jan 17, 2009
the aforementioned discussion: http://clojure-log.n01se.net/date/2009-01-06.html#18:41c
Comment 2 by the.stuart.sierra, Feb 18, 2009
Common Lisp has *read-eval*:
http://www.lispworks.com/documentation/HyperSpec/Body/v_rd_eva.htm

A similar flag might be an easy way to implement a "safe" reader.
Comment 3 by jhawk28, Mar 17, 2009
adds the *read-eval*
 Issue 34.patch.txt
2.1 KB Download
Comment 4 by richhickey, Apr 12, 2009
Patch applied - r1347 - thanks!
Status: Fixed

Reported by olov.lassus, Jan 02, 2009
The clojure.lang.Range type is implemented with integer type start and end
indices. The core.clj range function should detect when an index doesn't
fit inside an int and create an increasing sequence using iterate instead.
The current implementation detects too large positive, but not too large
negative ranges. The overflow occurs silently without throwing an exception.

Before:
user=> (take 1 (range -2147483649 0))
(2147483647)


After applying the patch which apply cleanly against trunk r1195:
user=> (take 1 (range -2147483649 0))
(-2147483649)


I also made the range function slightly less conservative < vs <= wise.
This shouldn't affect correctness but (range 0 2147483647) will now return
a Range (after patch) instead of a LazyCons (before).

 support_large_negative_range.patch
918 bytes Download
Comment 1 by richhickey, Jan 07, 2009
Patch applied - svn 1204 - thanks!
Status: Fixed

Reported by arohner, Jan 05, 2009
What (small set of) steps will reproduce the problem?

Create a java.sql.ResultSet containing two columns with the same name, and
then pass it to resultset-seq. 

What is the expected output? What do you see instead?
The expected output is an exception message that is helpful in diagnosing
the problem. resultset-seq instead throws "java.lang.RuntimeException:
java.lang.IllegalArgumentException: Too many arguments to struct constructor" 

What version are you using?
This was first reproduced in r1060. svn shows the function has not been
touched since 2008-07-29

Was this discussed on the group? If so, please provide a link to the
discussion:
http://groups.google.com/group/clojure/browse_frm/thread/e533e9cdf758f210?q=#43e5775c2d959dc7
http://groups.google.com/group/clojure/browse_thread/thread/58e4921de5c86c3c/7ca1de2daad839da?show_docid=7ca1de2daad839da

The attached patch modifies resultset-seq to throw an exception with the
message "
 resultset-seq.patch
1.5 KB Download
Comment 1 by arohner, Jan 05, 2009
The attached patch modifies resultset-seq to throw an exception with the message
"result set must have unique column names" when the column names are duplicated. It
also moves the definition of the set function above resultset-seq.
Comment 2 by richhickey, Feb 09, 2009
shouldn't this use distinct? instead of set+count
Comment 3 by arohner, Feb 09, 2009
You're right. I was not aware of distinct? Attaching v2 of the patch. This time,
rather than moving the definition of set up, I've moved the definition of
resultset-seq to below the definition of distinct?
 resultset-seq.v2.patch
2.2 KB Download
Comment 4 by richhickey, Apr 12, 2009
Fixed (also moved to using column labels) - r1346 - thanks for the report.
Status: Fixed

Reported by richhickey, Jan 04, 2009
Support
(ns
  "doc string goes here"
 ...)

Comment 1 by m...@kotka.de, Jan 04, 2009
Allow optional docstrings for a namespace as:

(ns foo.bar
  "docstring here"
  ...)

Attached patch uses the namespace symbol to transport the docstring to the namespace.

The Namespace class is modified to call its superclass constructor with the metadata
of the symbol passed in as name. If the first item of the references part of the ns macro
is a string, it is attached to the metadata of the given Symbol. From there it is again
transferred to the meta data of the Namespace.

Modified doc to check for the given symbol being a Namespace. If so its docstring is
retrieved and printed. If the symbol does not name a Namespace, it is interpreted as
a Var.

Added helper print-namespace-doc to complement print-doc.


 issue-30.diff
2.2 KB Download
Comment 2 by richhickey, Jan 22, 2009
Patch applied - svn 1224 - thanks!
Status: Fixed

Reported by richhickey, Jan 07, 2009
Marker interface with no methods. Would indicate O(1) count for
collections. Could be used to optimize vectr creation and some sequence ops

Comment 1 by chouser, Feb 10, 2009
Besides Java built-in classes, RT.count() works on
IPersistentCollection objects, so these are the ones that need to be
examined to see if they should implement Counted.

(defn getClassesInDir [basedir subdir]
  (let [regex (re-pattern (str basedir #"(.*)\.class"))]
    (map #(Class/forName (.replace (second (re-find regex (str %))) "/" "."))
         (.listFiles (java.io.File. (str basedir subdir))))))

(def clj-classes (getClassesInDir "/home/chouser/build/clojure/classes/" "clojure/lang"))

(defn has-count? [c]
  (try (.getMethod c "count" (into-array Class []))
    (catch NoSuchMethodException e)))

(count (filter has-count? clj-classes))  ==> 62 -- hopefully there are some useful
base classes.

(defn has-own-count? [c]
  (try (.getDeclaredMethod c "count" (into-array Class []))
    (catch NoSuchMethodException e)))

(count (filter has-own-count? clj-classes))  ==> 26 -- better

user=> (dorun (map println (sort (map str (filter has-own-count? clj-classes)))))
class clojure.lang.AMapEntry
class clojure.lang.APersistentSet
class clojure.lang.APersistentVector$RSeq
class clojure.lang.APersistentVector$Seq
class clojure.lang.APersistentVector$SubVector
class clojure.lang.ASeq
class clojure.lang.ArraySeq
class clojure.lang.ArraySeq$ArraySeq_double
class clojure.lang.ArraySeq$ArraySeq_float
class clojure.lang.ArraySeq$ArraySeq_int
class clojure.lang.ArraySeq$ArraySeq_long
class clojure.lang.Cons
class clojure.lang.LazilyPersistentVector
class clojure.lang.PersistentArrayMap
class clojure.lang.PersistentArrayMap$Seq
class clojure.lang.PersistentHashMap
class clojure.lang.PersistentList
class clojure.lang.PersistentList$EmptyList
class clojure.lang.PersistentQueue
class clojure.lang.PersistentQueue$Seq
class clojure.lang.PersistentStructMap
class clojure.lang.PersistentTreeMap
class clojure.lang.PersistentTreeMap$Seq
class clojure.lang.PersistentVector
class clojure.lang.Range
interface clojure.lang.IPersistentCollection

The promise of Counted is non-obvious for some of these.  For example,
APersistentSet and PersistentStructMap both have count() methods that
rely on IPersistentMap, which does not itself promise Counted.
However everything in clojure.lang that currently provides
IPersistentMap also promises Counted, so it seems appropriate for
APersistentSet and PersistentStructMap to promise Counted as well.

Similarly, APersistentVector$Seq uses count() from IPersistentVector
which does not itself promise Counted.  But since PersistentVector and
LazilyPersistentVector are both Counted, APersistentVector$Seq can
promise Counted.

Finally, PersistentQueue and PersistentQueue$Seq use ISeq.count(), but
the only ISeqs they use are PersistentList or seqs on vectors, so both
can also promise Counted.

If any of those conclusions are incorrect, or if the Counted marker
should be moved up to a parent interface in any of those cases, please
let me know.

The first patch attached marks almost all of the 'has-own-count?'
classes listed earlier as Counted.  The four non-Counted classes from
that list are:

(dorun (map println (filter #(and (has-own-count? %)
                                  (not (isa? % clojure.lang.Counted)))
                            clj-classes)))
clojure.lang.ASeq
clojure.lang.PersistentTreeMap$Seq
clojure.lang.Cons
clojure.lang.IPersistentCollection

There are total of 36 Counted classes, and 27 non-Counted
IPersistentCollection classes.

 add-counted.patch
10.2 KB Download
Comment 2 by richhickey, Feb 10, 2009
Thanks for doing the legwork on this, fixed in 1268
Status: Fixed

Reported by dresweng...@dreish.org, Jan 07, 2009
Last night on IRC, the GC Wiki was suggested as a way to organize a users'
wish-list so rhickey can conveniently ignore it and focus on real issues in
the issue tracker.  Unfortunately, it appears only rhickey is allowed to
edit, much less add pages to, the wiki.

Comment 1 by richhickey, Jan 07, 2009
No, the suggestion was to use the user wiki:

http://en.wikibooks.org/wiki/Clojure_Programming
Status: WontFix

Reported by achim.passen, Jan 10, 2009
There are some corner cases where "-"  overflows but does not throw an "integer overflow" 
exception:

user=> (- Integer/MAX_VALUE Integer/MIN_VALUE) 
-1 
user=> (- Long/MAX_VALUE Long/MIN_VALUE) 
-1 

The problem seems to be that negating MIN_VALUE yields MIN_VALUE   
again, so it slips through the overflow check. The patch adds extra checks for MIN_VALUE.

version: SVN r1205

http://groups.google.com/group/clojure/browse_thread/thread/3db6cb2b71008e73

 overflow.diff
693 bytes Download
Comment 1 by richhickey, Jan 14, 2009
patch applied - svn 1212 - thanks!
Status: Fixed

Reported by chouser, Jan 09, 2009
What (small set of) steps will reproduce the problem?

Originally reported by Tomasz Skutnik.

user=> (meta #^{:some :metadata} [])
nil
user=> (meta #^{:some :metadata} ())
nil
user=> (meta #^{:some :metadata} {})
nil

What is the expected output? What do you see instead?

I expect to see metadata {:some :metadata} as I do in these cases:

user=> (meta #^{:some :metadata} [1])
{:some :metadata}
user=> (meta '#^{:some :metadata} (1))
{:some :metadata}
user=> (meta #^{:some :metadata} {:a 1})
{:some :metadata}
user=> (meta (with-meta [] {:some :metadata}))
{:some :metadata}
user=> (meta (with-meta () {:some :metadata}))
{:some :metadata}
user=> (meta (with-meta {} {:some :metadata}))
{:some :metadata}

What version are you using?

SVN 1205

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

http://groups.google.com/group/clojure/msg/686e161b780565e0

Please provide any additional information below.

Comment 1 by richhickey, Jan 14, 2009
fixed in svn 1213 - thanks for the report
Status: Fixed

Reported by wolfe.a.jason, Jan 07, 2009
What (small set of) steps will reproduce the problem?

(list (.equals [1 2] (seq [1 2])) (.hashCode [1 2]) (.hashCode '(1 2)))


What is the expected output? What do you see instead?

I see (true 994 -1919631597)
I think I should see (true i i), or perhaps (false i j) for some ints i, j


What version are you using?

svn 1162


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

http://groups.google.com/group/clojure/browse_thread/thread/5d11bc0da25a7ecc/06d6cd888a516119?hl=en&lnk=gst&q=.hashcode#06d6cd888a516119

Nobody asked me to add this here so I hope this isn't out of line ... 


Please provide any additional information below.

The crux is, the Java API says:

"If two objects are equal according to the equals(Object)  method, then
calling the hashCode method on each of the two objects must produce the
same integer result. "

As you can see here:

user> (doseq [s ['(1 2) (seq '(1 2)) [1 2] (seq [1 2]) (ArrayList. [1
2])]]
  (print "\n" (.hashCode s))
  (doseq [t ['(1 2) (seq '(1 2)) [1 2] (seq [1 2]) (ArrayList. [1
2])]]
    (print "\t" (.equals s t))))

 -1919631597     true    true    true    true    false
 -1919631597     true    true    true    true    false
 994             true    true    true    true    true
 -1919631597     true    true    true    true    false
 994             false   false   true    false   true 

this contract is violated by Clojure collections.  In particular, 
(.equals [1 2] (seq [1 2]) but they have different .hashCode()s.  

Some issues this can cause
  - Clojure hash-map and array-map have different behaviors
  - (get (hash-map {(vector &args) true} (list &args)) will return true
    iff there is a hash collision between the vector and list (?) 
  - Unexpected pain in unknown places when using Clojure objects in Java
code that expects the contract to hold

-Jason

Comment 1 by wolfe.a.jason, Jan 07, 2009
P.S. I realized that the reason for the discrepancy might be that java.util.list
hashCodes (used by vectors) are defined in a front-to-back manner.  This makes it
perhaps non-obvious how to efficiently compute them recursively back-to-front, as one
might like to to for seqs (so that each rest-seq can cache its hash value).  This
code shows that this can in fact be done efficiently.

(defn forward-hash "How hashCode() is defined in java.util.List"  [s]
  (loop [s (seq s) h (int 1)]
    (if s 
        (recur (rest s) (int (unchecked-add (unchecked-multiply h 31) (hash (first s)))))
      h)))


(def int-pow 
     (memoize 
      (fn [x n]
 (let [x (int x) n (int n)]
   (cond (zero? n) 1
  (even? n) (let [h (int-pow x (/ n 2))] (unchecked-multiply h h))
  :else     (unchecked-multiply x (int-pow x (dec n))))))))

(defn backward-hash "Compute the same value as forward-hash, but back-to-front."  [s]
  (let [s (seq s) p (int-pow 31 (count (rest s)))]
    (if s 
        (unchecked-add (unchecked-multiply 30 p)
         (unchecked-add (int (backward-hash (rest s)))
          (unchecked-multiply (int (hash (first s))) p)))
 1)))
   

user> (map #(% (vec (range 1000))) [forward-hash backward-hash #(.hashCode %)])
(133786869 133786869 133786869)

HTH, Jason

P.P.S. I'll put my Contributor Agreement in the mail today.   
Comment 2 by wolfe.a.jason, Jan 07, 2009
Note on a small special case: Java null effectively has hash code 0 (I think) whereas
java.util.List hashCode() must be 1 for empty Lists.  Fortunately, the REPL says:

user> (.equals [1] (seq [1]))
true
user> (.equals [] (seq []))
false

so I guess the empty vector is not .equal to the empty seq anyway (as mandated by the
Java API).  So, giving non-empty seqs java.util.List hashCodes would in fact be the
right thing to do, it seems to me.  


Comment 3 by richhickey, Jan 15, 2009
Fixed - svn 1215 - thanks for the report

Status: Fixed

Reported by Chousuke, Jan 10, 2009
several macros only check for (even? (count bindings)) even though they
should check for exactly 2 bindings. 

Trivial fix attached.

--
Jarkko Oranen
 fix-check.patch
1.5 KB Download
Comment 1 by chouser, Jan 10, 2009
Whoops.  Thanks for catching that, Jarkko.
Comment 2 by richhickey, Jan 14, 2009
patch applied - svn 1211 - thanks!
Status: Fixed

Reported by m...@kotka.de, Jan 10, 2009
What (small set of) steps will reproduce the problem?

Try to compile the following namespace definition:

(ns foo.Bar
  (:gen-class
     :methods [[toString [Object] String]]))

What is the expected output? What do you see instead?

I expect this to compile normally, but the following exception is thrown:

1:1 user=> (binding [*compile-path* "tmp"] (compile 'foo.Bar))
java.lang.RuntimeException: java.lang.RuntimeException: java.lang.RuntimeException: 
java.lang.RuntimeException: java.lang.RuntimeException: java.lang.RuntimeException: 
java.lang.RuntimeException: java.lang.IllegalArgumentException: No matching field found: 
getSimpleName for class clojure.lang.Symbol (Bar.clj:1)

What version are you using?

SVN rev. 1195

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

The issue was brought up by baggins on #clojure:
http://clojure-log.n01se.net/date/2009-01-10.html#07:54

Please provide any additional information below.

Attached is a patch, which fixes the symptom. But I'm not sure, whether it fixes the issue.
 issue-42.diff
851 bytes Download
Comment 1 by richhickey, Jan 14, 2009
Path applied - svn 1210 - thanks!
Status: Fixed

Reported by mark.engelberg, Dec 28, 2008
mod (modulo) is similar to rem, but the result will always have the same
positive/negative sign as the second input:

(mod 2 5) -> 2
(mod 2 -5) -> -3
(mod -2 5) -> 3
(mod -2 -5) -> -2

Note that rem always has a result whose sign matches the first input:
(rem 2 5) -> 2
(rem 2 -5) -> 2
(rem -2 5) -> -2
(rem -2 -5) -> -2

So mod and rem yield the same answer if both inputs are positive, or both
are negative.  However, when the signs are mixed, they differ.

Sample code:
(defn mod [a b]
  (if (or (< a 0 b) (< b 0 a))
    (+ (rem a b) b)
    (rem a b)))
Comment 1 by achim.passen, Dec 29, 2008
A patch that adds mod and sgn
 mod-sgn.diff
6.9 KB Download
Comment 2 by richhickey, Jan 07, 2009
hmm... Mark's sample code looks much simpler - anything wrong with it?
Comment 3 by achim.passen, Jan 08, 2009
Not at all, it's functionally equivalent. I implemented it in Java mainly to be consistent with the other basic math 
op implementations. The Java version seems to be a tick faster (~500ms vs ~600ms on 100000 pairs of 
randomly typed random numbers here).

Attached is a simplified patch which adds LongOps support and gets rid of signum.


 r1205+mod.diff
4.9 KB Download
Comment 4 by richhickey, Jan 22, 2009
Added Mark's version, SVN 1227 - thanks!
Status: Fixed
Comment 6 by timothypratley, Feb 12, 2009
Yet another 'fixed' version:
(mod 9 -3) -> 0
(mod 4.5 2) -> 0.5
 mod-fix.patch
759 bytes Download
Comment 7 by richhickey, Feb 23, 2009
patch applied - svn 1302 - thanks!

Reported by chouser, Jan 15, 2009
Describe the feature/change.

It's currently impossible to create a class with gen-class that calls
its own (or its super-classes') methods when it is constructed.  This
is a problem if there is stateful initialization required by the
superclass for which it has only methods, not ctor args.

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

http://groups.google.com/group/clojure/browse_thread/thread/d1fbdb7450b46dee/b5539e4a21c411c6
http://clojure-log.n01se.net/date/2009-01-15.html#10:22-10:37a

Comment 1 by chouser, Mar 07, 2009
(No comment was entered for this change.)
 gen-class-post-init.patch
3.7 KB Download
Comment 2 by cjkent, Mar 24, 2009
I've applied the patch and it works, but only if I define a post-init function with
no args.  I was expecting at least a single "this" arg.  Or is there another way to
get a reference to the newly created object?
Comment 3 by chouser, Mar 24, 2009
The args given to your class's constructor are passed directly to your post-init
function.  But you have a good point -- the 'this' arg should probably be included. 
I've attached a new patch that allows this:

(ns clojure.contrib.test-clojure.gen-class
  (:gen-class
     :extends javax.swing.border.TitledBorder
     :constructors {[int int String] [String]}
     :init my-init
     :post-init after-ctor))

(defn -my-init [x y s]
  (println "Init ignoring" x "and" y ", going with" (str "foo+" s))
  [[(str "foo+" s)] nil])

(defn -after-ctor [this x y s]
  (print "Constructing" this "with ")
  (prn x y s))

user=> (compile 'clojure.contrib.test-clojure.gen-class)
clojure.contrib.test-clojure.gen-class

user=> (clojure.contrib.test_clojure.gen_class. 3 4 "hello")
Init ignoring 3 and 4 , going with foo+hello
Constructing #<gen_class clojure.contrib.test_clojure.gen_class@1ab2b55> with 3 4 "hello"
#<gen_class clojure.contrib.test_clojure.gen_class@1ab2b55>

 post-init.patch
3.8 KB Download
Comment 4 by cjkent, Mar 25, 2009
Great, that works perfectly.  Thanks.
Comment 5 by richhickey, Apr 12, 2009
Patch applied - r1348 - thanks!
Status: Fixed

Reported by m...@kotka.de, Jan 17, 2009
Describe the feature/change.

This is an issue from the TODO list.

Currently multimethods use the global-hierarchy. However sometimes it is useful, to define
a dedicated hierarchy for a given set of multimethods. The attached patch adds a new
:hierarchy argument to defmulti which specifies the desired hierarchy. The hierarchy may be
any IRef.

The patch depends on the patch of  issue #8 .

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

http://groups.google.com/group/clojure/browse_frm/thread/f8b1be403c927b03/438c21a8007
2a105?lnk=gst&q=per+defmulti+hierarchies

 issue-48.diff
3.4 KB Download
Comment 1 by richhickey, Feb 09, 2009
patch applied - svn 1262 - thanks!
Status: Fixed

Reported by chouser, Jan 05, 2009
What (small set of) steps will reproduce the problem?

The (< x y) form of the following expression is silently ignored:
(take 100 (for [x  (range 1000) y (range 1000) (< x y)][x y])) 

What version are you using?

SVN 1195

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

http://groups.google.com/group/clojure/browse_thread/thread/56793fb65903cece

Please provide any additional information below.

The attached patch is identical to the one in the discussion thread.
 binding-assertions.patch
8.1 KB Download
Comment 1 by richhickey, Jan 07, 2009
patch applied - svn 1205 - thanks!
Status: Fixed

Reported by s-google...@localtoast.org, Jan 14, 2009
What (small set of) steps will reproduce the problem?

Invoke a java method which throws an AssertionError (or any other subclass
of Error), through reflection.

What is the expected output? What do you see instead?

I'd expect to catch the AssertionError object itself (which is the case for
a type-annotated call). Instead, I end up catching an InvocationTargetException

What version are you using?

r1214

Comment 1 by richhickey, Jan 24, 2009
Could you please provide some Clojure code that demonstrates the problem?
Comment 2 by s-google...@localtoast.org, Jan 29, 2009
; method defined through clojure for brevity, any java method which throws a
non-Exception will suffice
user=> (defn throw-error [] (throw (new Error)))
#'user/throw-error

; expected behavior (type annotation creates a direct call)
user=> (let [#^clojure.lang.IFn o throw-error] (.invoke o))
java.lang.Error (NO_SOURCE_FILE:0)

; unexpected behavior, the reflect code has wrapped our Throwable
user=> (let [o throw-error] (.invoke o))
java.lang.reflect.InvocationTargetException (NO_SOURCE_FILE:0)


Comment 3 by richhickey, Feb 23, 2009
fixed - svn 1303 - thank for the report
Status: Fixed
Comment 4 by berlin.brown, Feb 24, 2009
I am getting this error also, is this fixed in the december release:


(defn simple-mkdirs-handler
  "Simple utility to make directories.  Throws error on invalid directory."
  [dirname]
  ;;;;;;;;;;;;;;;;;
  (if (empty? dirname) (throw (Error. "ERR: <mkdirs> Invalid Directory"))
      (let [file (new File dirname)]
        (if (not (.canWrite file))
          (throw (Error. "ERR: <mkdirs> Invalid Directory. Cannot write."))
          (try (.mkdirs file)
               (catch Exception e
                      (throw (Error. (str "ERR: <mkdirs> Invalid Directory. Cannot
mkdir => " 
                                          (.getMessage e))))))))))

Reported by Chousuke, Dec 18, 2008
in Clojure rev 1175
------------------------------
(import '(java.io StringReader))
(def stream1 (StringReader. "Test"))
(def stream2 (StringReader. "Test2"))

(with-open [a stream1 b stream2] (println "do-nothing"))

(.ready stream2) ; should throw an exception, but doesn't, since stream2 is
not closed

the reason is simple enough: in with-open, the bindings parameter is passed
to let, but in the finally clause, only a simple (.close ~(first bindings))
is emitted.

I have a patch to make with-open support multiple bindings cleanly, but I
do not have a CA so I can't submit it.

--
Jarkko Oranen




Comment 1 by m...@kotka.de, Dec 18, 2008
Reading the docstring I get the impression, that only one binding was intended at the beginning.
However allowing multiple bindings would a) not be a problem, b) justify the let style binding and
c) have pratical relevance.

I attached a simple patch.
 with-open.diff
887 bytes Download
Comment 2 by Chousuke, Dec 19, 2008
The patch is essentially identical to the solution I came up with, except that I used
let* instead of let for safety because destructuring doesn't make sense for with-open.

I'm going to have to go submit a CA so that next time I have an issue with a solution
I can actually submit the patch myself :)
Comment 3 by richhickey, Dec 19, 2008
That is a good point - destructuring is not supported. I don't think let* is the way
to achieve that - no one should be writing let* directly. The macro could just check
that all bindings are to simple symbols.

Also, the closes should happen in reverse order.

This can be achieved by solving the last, most important problem - you can't just do
this with one let, as each binding could potentially fail:

open a
open b
throw exception opening c

b and a never closed.

i.e., multi-binding with-open must use nested lets/trys, maybe could be defined in
terms of recursive calls to current macro.
Comment 4 by m...@kotka.de, Dec 19, 2008
Argh. I should really think, before writing.

Ok. Here the next idea:
with-open calls itself recursively. It checks that the next binding target (?) is a symbol.
Then it assigns just this one binding to a let and again calls itself wrapped in a try. This
gives in particular the reverse order for closing. And in case something bad happens
during opening ensures that previously opened streams are closed.

I had one problem though: Since we check for the bindings being a vector, one has
to take care, that one passes around the bindings again as a vector. Or one uses an
inner macro, where with-open does the check and then passes on to with-open*,
which also accepts a seq.

I took a different route and simply did a subvec instead of a take or drop. Then
we check for the vector several times, but I think this is the least intrusive version.
I had to move the subvec defn upwards, though.

 with-open.diff
2.3 KB Download
Comment 5 by richhickey, Dec 23, 2008
Patch applied (svn 1182) - thanks!
Status: Fixed

Reported by konrad.h...@laposte.net, Jan 20, 2009
Describe the feature/change.

set/union currently accepts two sets as arguments. The operation is well defined for any number of 
sets and similar operations (+, concat, ...) take any number of arguments, so I think set/union 
should do so as well.

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


 set-union.diff
575 bytes Download
Comment 1 by scgilardi, Jan 20, 2009
There are implementations here as well:

http://groups.google.com/group/clojure/msg/5bb548afe2cbbe02

At one point, this was on Rich's public Clojure "to-do" list.

Comment 2 by wolfe.a.jason, Jan 22, 2009
On a somewhat related note, let me plug my own post about efficient set operations:

http://groups.google.com/group/clojure/browse_thread/thread/29609929e94f279c?hl=en#

(I am in favor of multi-arg set functions as well; while my versions don't do this, I
believe this can be done simply by (reduce union #{} sets) or (reduce
intersection/difference (first sets) (rest sets))


Comment 3 by wolfe.a.jason, Feb 02, 2009
This may now be subsumed by this issue:

http://code.google.com/p/clojure/issues/detail?id=64&colspec=ID%20Type%20Status%20Priority%20Reporter%20Owner%20Summary

which includes a patch for multi-arg, more efficient
clojure.set/union/difference/intersection functions.
Comment 4 by richhickey, Feb 22, 2009
(No comment was entered for this change.)
Status: Fixed

Reported by christophe.grand, Jan 15, 2009
format, name and namespace are not hinted {:tag String}

(If you can't apply this patch, I promise to never use svn diff again!)
 name-namespace-format.patch
933 bytes Download
Comment 1 by christophe.grand, Jan 16, 2009
A broader patch that removes many reflection warnings from clojure.core files.
(I don't see how to prevent inlinable fns to raise warnings for their "default case".)
 clojure-core-hints.patch
14.6 KB Download
Comment 2 by christophe.grand, Jan 17, 2009
added hints to to-array and to-array-2d.
 clojure-core-hints2.patch
15.2 KB Download
Comment 3 by richhickey, Jan 24, 2009
This is great!

patch applied - svn 1229 - thanks!
Status: Fixed

Reported by timothypratley, Jan 28, 2009
"repeat" with multiple arities and remove "replicate"
http://groups.google.com/group/clojure/browse_thread/thread/bb2bd6ad6986fed
f


 repeat.patch
746 bytes Download
Comment 1 by timothypratley, Jan 28, 2009
opps didn't mean to submit this yet sorry, hang on.
Comment 2 by timothypratley, Jan 28, 2009
Please find attached the 'real' patch! Sorry, first time using this system.
 repeat.patch
3.1 KB Download
 contrib-repeat.patch
740 bytes Download
Comment 3 by richhickey, Feb 09, 2009
I don't have a CA from you, please see:

http://clojure.org/contributing
Comment 4 by richhickey, Apr 12, 2009
Repeat now supports an optional number-of-repetitions arg. Removing replicate is
another issue
Status: Fixed

Reported by chouser, Jan 15, 2009
Describe the feature/change.

This much works already:
(defstruct rect :width :height)
(struct rect 5 10)  ==> {:width 5, :height 10}

With the included patch you can also:
(rect 5 10)  ==> {:width 5, :height 10}

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

http://groups.google.com/group/clojure/browse_thread/thread/12a138ad58ff6c36/b20b68ef939fccf7

Comment 1 by chouser, Jan 15, 2009
Forgot the patch attachment.
 structmap-def-extends-restfn.patch
923 bytes Download

Reported by tuomas.lukka, Jan 29, 2009

What (small set of) steps will reproduce the problem?

Run the code `()   

What is the expected output? What do you see instead?

I'd expect it to return a quoted empty list. Instead an error results:

Exception in thread "main" java.lang.UnsupportedOperationException: Unknown
Collection type (test.clj:1)
 at clojure.lang.Compiler.load(Compiler.java:4476)
 at clojure.lang.Compiler.loadFile(Compiler.java:4437)
 at clojure.lang.Script.main(Script.java:65)
Caused by: java.lang.UnsupportedOperationException: Unknown Collection type
 at clojure.lang.LispReader$SyntaxQuoteReader.syntaxQuote(LispReader.java:711)
 at clojure.lang.LispReader$SyntaxQuoteReader.invoke(LispReader.java:655)
 at clojure.lang.LispReader.read(LispReader.java:143)
 at clojure.lang.Compiler.load(Compiler.java:4466)
 ... 2 more

What version are you using?

20081217

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

no

Please provide any additional information below.

Comment 1 by richhickey, Feb 09, 2009
fixed in svn 1255 - thanks for the report
Status: Fixed

Reported by tuomas.lukka, Jan 29, 2009
 
------------------------------

What (small set of) steps will reproduce the problem?

Run the following code:

(defmacro b [] (let [ x (fn [] []) ] x))
(def a (fn [] (b)))


What is the expected output? What do you see instead?

I'd expect it to either pass or give a proper error. What I get
is

Exception in thread "main" java.lang.ExceptionInInitializerError (test.clj:2)
 at clojure.lang.Compiler$DefExpr.eval(Compiler.java:308)
 at clojure.lang.Compiler.eval(Compiler.java:4147)
 at clojure.lang.Compiler.load(Compiler.java:4470)
 at clojure.lang.Compiler.loadFile(Compiler.java:4437)
 at clojure.lang.Script.main(Script.java:65)
Caused by: java.lang.ExceptionInInitializerError
 at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)
 at
sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:39)
 at
sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:27)
 at java.lang.reflect.Constructor.newInstance(Constructor.java:513)
 at java.lang.Class.newInstance0(Class.java:355)
 at java.lang.Class.newInstance(Class.java:308)
 at clojure.lang.Compiler$FnExpr.eval(Compiler.java:3218)
 at clojure.lang.Compiler$DefExpr.eval(Compiler.java:297)
 ... 4 more
Caused by: java.lang.RuntimeException: java.lang.ClassNotFoundException:
clojure.core$b__1$x__3
 at clojure.lang.RT.readString(RT.java:1192)
 at clojure.core$a__7.<clinit>(test.clj:2)
 ... 12 more
Caused by: java.lang.ClassNotFoundException: clojure.core$b__1$x__3
 at java.net.URLClassLoader$1.run(URLClassLoader.java:200)
 at java.security.AccessController.doPrivileged(Native Method)
 at java.net.URLClassLoader.findClass(URLClassLoader.java:188)
 at clojure.lang.DynamicClassLoader.findClass(DynamicClassLoader.java:52)
 at java.lang.ClassLoader.loadClass(ClassLoader.java:307)
 at java.lang.ClassLoader.loadClass(ClassLoader.java:252)
 at java.lang.ClassLoader.loadClassInternal(ClassLoader.java:320)
 at java.lang.Class.forName0(Native Method)
 at java.lang.Class.forName(Class.java:247)
 at clojure.lang.RT.classForName(RT.java:1506)
 at clojure.lang.LispReader$EvalReader.invoke(LispReader.java:908)
 at clojure.lang.LispReader$DispatchReader.invoke(LispReader.java:530)
 at clojure.lang.LispReader.read(LispReader.java:143)
 at clojure.lang.RT.readString(RT.java:1188)
 ... 13 more


What version are you using?

20081217

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

no

Reported by arohner, Jan 31, 2009

Describe the feature/change.

clojure.main should not exit if there was an exception in an --init file
and the user requested a repl. i.e.

java -cp clojure.jar clojure.main --init file-with-errors.clj -r 

should not exit. Having a repl is useful to read doc strings and experiment
with fixes for the problem.

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

http://groups.google.com/group/clojure/browse_frm/thread/f96109d71fa47200#

The attached patch adds a try/catch around the init method for the repl in
clojure.main. It does not change the behavior for the case where the user
specifies a file to run rather than a repl.

Comment 1  by richhickey, Feb 09, 2009
patch applied - svn 1258 - thanks!
Status: Fixed

Reported by chouser, Jan 28, 2009
Describe the feature/change.

Currently when you print an IRef, the contents are not printed.  With
the attached patch:

user=> (agent 99)
#<Agent@1dee400: 99>

user=> (atom #{:bomb})
#<Atom@e3849c: #{:bomb}>

user=> (ref "erence")
#<Ref@44cbbe: "erence">

Note that they're no more readable in the (read) sense.  Also note
that Vars are unchanged:

user=> #'nil?
#'clojure.core/nil?

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

http://groups.google.com/group/clojure/browse_thread/thread/e63c5e3e1f429f6e

 print-iref.patch
694 bytes Download
Comment 1 by richhickey, Feb 09, 2009
patch applied - svn 1254 - thanks!
Status: Fixed

Reported by jbondeson, Jan 30, 2009
Originally discussed here:
http://groups.google.com/group/clojure/browse_thread/thread/1f8029b8556ce851/9b7538b192490097?lnk=gst&q=ratio#9b7538b192490097

The current implementation of conversion of clojure.lang.Ratio class
to a fixed precision floating point value is not what one would expect. 
For example:

user> (double (/ (expt 2 1024) 3))
Infinity
when in fact it should return: 5.992310449541053E307  

Attached is the fix for the Ratio class to allow for the proper conversion
to a double or float by using a BigDecimal division with a fixed precision.

Additionally, I sent in my CA on Monday, so it should be there anytime now.
 ratio-conv.diff
1.0 KB Download
Comment 1 by richhickey, Feb 09, 2009
patch applied - svn 1256 - thanks!
Status: Fixed

Reported by chouser, Feb 02, 2009

What (small set of) steps will reproduce the problem?

user=> (doc refer-clojure)
-------------------------
clojure.core/refer-clojure
([& filters])
Macro
  Same as (refer 'clojure <filters>)
nil

What is the expected output? What do you see instead?

Docstring should be:
Same as (refer 'clojure.core <filters>)

What version are you using?

1241 trunk

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

http://groups.google.com/group/clojure/msg/598502ccfc42a5ce

Please provide any additional information below.

Comment 1 by richhickey, Feb 09, 2009
fixed in svn 1259 - thanks for the report
Status: Fixed

Reported by christophe.grand, Jan 19, 2009
defmacro forms return nil when evaled while defn and def forms return the
defined Var.

 defmacro-returns-var.patch
700 bytes Download
Comment 1 by richhickey, Jan 24, 2009
patch applied - svn 1230 - thanks!
Status: Fixed

Reported by m...@kotka.de, Jan 19, 2009
Describe the feature/change.

Currently only the basename of the file where a Var is defined is stored in its metadata.
Unfortunately this is ambiguous. So extracting the exact position in the source tree
through use of the metadata is not possible.

Attached is a patch which stores

- the classpath relative filename for files loaded through the classpath or
- an absolute filename for files loaded via load-file

in the metadata of the Var. This removes any ambiguity.

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

This was briefly asked on the #clojure channel.
http://clojure-log.n01se.net/date/2009-01-19.html#11:55

 issue-50.diff
921 bytes Download
Comment 1 by richhickey, Apr 12, 2009
Patch applied - r1349 - thanks!
Status: Fixed

Reported by chouser, Jan 05, 2009
What (small set of) steps will reproduce the problem?

N/A

What is the expected output? What do you see instead?

N/A

What version are you using?

SVN 1195

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

http://groups.google.com/group/clojure/browse_thread/thread/b2d73537eca894f0/d5a3b1e31f80c49f

Please provide any additional information below.

 comments.patch
5.6 KB Download
Comment 1 by richhickey, Jan 06, 2009
oops - gotta change that comment template in all my editors

patch applied svn 1197 - thanks!
Status: Fixed

Reported by cemer...@snowtide.com, Feb 06, 2009

Attached is a patch to add a warn-on-reflection option to  
clojure.lang.Compile.  In short, if the clojure.compile.warn-on-
reflection system property is set to "true", then *warn-on-reflection*  
is set to true for the duration of the compile process for the libs  
specified as command line args. 

This feature was discussed in #clojure on 2009.02.05:
http://clojure-log.n01se.net/date/2009-02-05.html#08:18

Comment 1  by richhickey, Feb 09, 2009

patch applied - svn 1260 - thanks!

Status: Fixed

Reported by hlship, Feb 08, 2009

In many cases, when you have a script with no extra dependencies outside of
Clojure, it is easiest to:

java -jar clojure.jar myscript.clj

This should be mentioned on:

http://clojure.org/repl_and_main

Comment 1 by richhickey, Feb 09, 2009

This was discussed and it was agreed that this causes more problems later on for new
users when try to use other jars and -cp doesn't work

Status: WontFix

Reported by pedro.dhb, Feb 07, 2009


What (small set of) steps will reproduce the problem?

(every? #{\space} "dam") #=> nil


What is the expected output? What do you see instead?
The expected output is false according to the API docs for every?. Currently nil is returned.

What version are you using?
SVN revision 1252

Was this discussed on the group? If so, please provide a link to the discussion:
It was mentioned on the IRC but not in the groups.

Please provide any additional information below.
_END _INFO



Comment 1 by slash2314, Feb 24, 2009

If need needs to return false in this case, I propose the following patch.

Reported by m...@kotka.de, Jan 18, 2009
Describe the feature/change.

Currently, ~ outside a syntax-quote is expanded to (clojure.core/unquote ..).
However this is not done for ~@. The attached patch also introduces the
expansion of ~@ to (clojure.core/unquote-splicing ...).

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

http://groups.google.com/group/clojure/browse_frm/thread/6d0c5f7e37909055


 issue-49.diff
2.2 KB Download
Comment 1 by richhickey, Feb 09, 2009
patch applied - svn 1263 - thanks!
Status: Fixed

Reported by christophe.grand, Jan 24, 2009
r1228:
user=> (tree-seq (constantly false) seq [1 2 3])
([1 2 3] 1 2 3) ; I expected ([1 2 3])

The actual behaviour of tree-seq is undefined when the root isn't a branch
since children is called upon root without any check.

The attached patch fixes this by always checking if a node is a branch
before calling children on it.

http://groups.google.com/group/clojure/browse_thread/thread/335707370295b654



 tree-seq.patch
1.6 KB Download
Comment 1 by richhickey, Jan 24, 2009
patch applied - svn 1231 - thanks!
Status: Fixed

Reported by jbondeson, Feb 09, 2009

This patch was originally discussed in IRC on 2009-02-09.

The proxy macro allows you to define methods of the same name within the
same block.  This results in a last-defined-wins scenario.

user> (def p (proxy [java.util.concurrent.Future] [] (get [] 0) (get
[timeout unit] 2)))
#'user/p
user> (.get p 30 java.util.concurrent.TimeUnit/MILLISECONDS)
2
user> (.get p)
java.lang.IllegalArgumentException: Wrong number of args passed to:
user$fn--1271$fn (NO_SOURCE_FILE:0)

If the definitions are reversed the results are reversed:

user> (def q (proxy [java.util.concurrent.Future] [] (get [timeout unit] 2)
(get [] 0)))
#'user/q
user> (.get q)
0
user> (.get q 30 java.util.concurrent.TimeUnit/MILLISECONDS)
java.lang.IllegalArgumentException: Wrong number of args passed to:
user$fn--1416$fn (NO_SOURCE_FILE:0)

The correct way to do multiple files of different -arity is:

user> (def r (proxy [java.util.concurrent.Future] [] (get ([] 0) ([timeout
unit] 2))))
#'user/r
user> (.get r)
0
user> (.get r 30 java.util.concurrent.TimeUnit/MILLISECONDS)
2

The attached patch will catch an incorrect method redefinition and throw an
IllegalArgumentException identifying the method redefined.

Comment 1  by richhickey, Feb 10, 2009

patch applied - svn 1257 - thanks!

Status: Fixed

Reported by scgilardi, Feb 09, 2009

What (small set of) steps will reproduce the problem?

    The defect is apparent by inspecting the source for clojure.main/repl.
Defect reported by Eric Thorsen.

What is the expected output? What do you see instead?

    N/A

What version are you using?

    svn 1263

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

    http://groups.google.com/group/clojure/browse_frm/thread/237632fbd6b3c01b

Please provide any additional information below.

    A patch is attached. I cleaned up the read manipulations into a new
repl-read hook. clojure.main/repl is more readable now. I also made
functions in clojure.main for repl-prompt and repl-caught, formerly inline
functions. Clarified some hook descriptions in the docs string for repl.
Avoided using a var (even private) for the eof sentinel in eval-opt.

Note: the number of arguments to the read hook has changed. When this patch
is applied, all previous read hooks will need an update. I'll handle the
one in clojure.contrib.repl-ln.

Comment 1  by richhickey, Feb 10, 2009

Thanks! While this patch says it is based on rev 1263, it doesn't seem to reflect the
changes of rev 1258:

http://code.google.com/p/clojure/source/diff?spec=svn1258&r=1258&format=side&path=/trunk/src/clj/clojure/main.clj

added for  issue 62 :

http://code.google.com/p/clojure/issues/detail?id=62

Is this just an oversight or was there a problem with that patch?


Comment 2 by scgilardi, Feb 10, 2009

It was just an oversight. I'm sorry about that. Thanks for catching it. The enclosed
patch: repl-read-2.patch includes the functionality from the fix to  issue 62 .

Comment 3  by richhickey, Feb 10, 2009

patch applied - svn 1269 - thanks!

Status: Fixed

Reported by wolfe.a.jason, Feb 02, 2009

Describe the feature/change.

Efficient set operations that use information about set sizes to iterate
through smaller sets.

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

http://groups.google.com/group/clojure/browse_thread/thread/29609929e94f279c?hl=en&tvc=2&q=jason+set

http://groups.google.com/group/clojure/browse_thread/thread/b35ca9035762e1a4/67786ac082d5f6f4?hl=en&lnk=gst&q=jason+set#67786ac082d5f6f4

In particular, the second link describes the algorithms implemented in the
patch.

Also, since you didn't respond to my question, I decided to go ahead and
merge with this issue:

http://code.google.com/p/clojure/issues/detail?id=52&colspec=ID%20Type%20Status%20Priority%20Reporter%20Owner%20Summary

and make multi-arg versions.  I can provide a two-arg patch instead if you
prefer.

Right now, there is no calling of "set" or "set?", and so very unexpected
behavior may occur when non-sets are passed:

user=> (union #{1 2} #{4 5} [5 6 7] #{7 8})
[5 6 7 7 8 1 2 4 5]

(Since the vector is largest, it is used as the seed, and calling "into" on
it uses vector rather than set semantics).

user=> (intersection #{0 1} [5 6 7])
#{0 1}

(Since the set is smallest, it is used as the seed.  Then, when we ask the
vector if it "contains?" 0 or 1, it uses vector contains-index semantics
rather than set contains-value semantics).

If calling "set" on the arguments is low-overhead for things that are
already sets, I would be in favor of that rather than allowing the above
behaviors.  But, I got the impression you would not be, so I left this out
for now.  

Please let me know if you have any questions.

Comment 1  by richhickey, Feb 09, 2009

I don't understand the homemade sorts and two-arg-xxx fns

Comment 2 by wolfe.a.jason, Feb 09, 2009

The "homemade sorts" are several times faster than "sort" or the equivalent, because
they only need to extract the largest or smallest set from the arglist.  I couldn't
find a more concise or clear way of writing them that didn't slow the entire
functions down significantly when passed a few small sets.

The two-arg fns are just helpers, to avoid repeating the same code several times.

Take intersection, for instance.  The basic algorithm is:
  - Find and extract the smallest set from the arg-list
  - Successively intersect this with each other set, using two-arg-intersection
    (which iterates through the first set, which will always be smaller).
This means that intersection is O((size of smallest set) * (number of sets)) rather
than O(sum of set sizes) as you would get with a simpler implementation.



Comment 3 by richhickey, Feb 10, 2009

You could probably use min-key + remove if you don't need a full sort:

(let [mv (apply min-key count vs) ovs (remove #(identical? mv %) vs)]...

If the hand bubble still has utility, you could make it its own function. The
duplication in union and intersection is not good.

As far as the external two-arg-fns, all I meant was that there's no reason not to
make these the two-arg arity overloads themselves, i.e. the variable arity
'difference' can reduce with the arity-2 'difference'. Lots of Clojure code does this.


Comment 4 by wolfe.a.jason, Feb 10, 2009

My main goal in writing these was efficiency; in particular, my objective was to
perform as close to the current versions (with reduce) when passed a bunch of small
(e.g., single-element sets).  This meant trying to reduce the amount of per-argument
overhead.

I originally had factored out an extract-min (i.e., hand-bubble) function, but for it
to work for both union and intersection it had to take a compare or key function as
argument, and this seemed to slow things down substantially over inlining (along with
the one or two extra conses needed).

Similarly, the problem with making the two-arg version of intersection an arity
overload is that (the only way I can see how to do it) this would triple-up on calls
to count.  I guess there's no reason not to do it for difference though; that's
probably an artifact of an earlier version.

Anyway, I'll do some more experiments, try to clean these up a bit more, and submit
an updated patch.  In general, what's your feeling about trading off clarity of code
for efficiency?  For instance, if getting rid of the inlined hand-bubble causes a 50%
performance hit on small sets, is this worth it?

Thanks!

Comment 5 by richhickey, Feb 10, 2009

> In general, what's your feeling about trading off clarity of code for efficiency?  

I'm generally against it.

> For instance, if getting rid of the inlined hand-bubble causes a 50% performance
hit on small sets, is this worth it? 

Yes. Really, I don't want micro-optimized code like this in Clojure. Things change
pretty fast in JVM optimization, and the tweaks you make now can become completely
irrelevant or even counterproductive in the future. Best to keep things simple and clear.


Comment 6 by wolfe.a.jason, Feb 11, 2009

OK, a second stab at a patch is attached.  I've tried to clean things up as much as I
could without hurting efficiency too much.

In particular,
 - I extracted a bubble-max-key function.  I must have done something wrong before,
since this didn't actually slow things down by much.
 - I got rid of the "two-arg" functions.  This had no effect for difference, but
slowed down intersection by up to 40% by tripling calls to "count".

Overall, intersection is up to 45% slower than the previous version, union is up to
25% slower, and difference is the same.  If you really don't like bubble-max-key, you
can replace it by

(defn bubble-max-key [k coll]
  (let [max (apply max-key k coll)]
    (cons max (remove #(identical? max %) coll))))

but this will make union and intersection run about twice as slow on inputs like
[#{1} #{2} #{3}].

Thanks!

Comment 7  by richhickey, Feb 11, 2009

patch applied - svn 1276 - thanks!

Your analysis of how much % slower is flawed though, a classic example of improving
the time of something that will be only a small part of the total runtime with
non-trivial sized sets, i.e. the algorithmic differences matter, the constant factors
per set (count calls etc) don't.

Status: Fixed

Reported by tuomas.lukka, Jan 27, 2009
------------------------------

What (small set of) steps will reproduce the problem?

(defn a []
    (proxy [Runnable] []
        (run [] (println "HEI"))))

(defn a []
    (proxy [Foo] []
        (run [] (println "HEI"))))


What is the expected output? What do you see instead?

I'd expect a reasonable compiler error. I get an internal
error instead.
java.lang.NullPointerException (test.clj:7)
 at clojure.lang.Compiler.analyzeSeq(Compiler.java:4113)
 at clojure.lang.Compiler.analyze(Compiler.java:3935)
 at clojure.lang.Compiler.analyze(Compiler.java:3908)
 at clojure.lang.Compiler$BodyExpr$Parser.parse(Compiler.java:3613)
 at clojure.lang.Compiler$FnMethod.parse(Compiler.java:3456)
 at clojure.lang.Compiler$FnMethod.access$1100(Compiler.java:3335)
 at clojure.lang.Compiler$FnExpr.parse(Compiler.java:2921)
 at clojure.lang.Compiler.analyzeSeq(Compiler.java:4104)
 at clojure.lang.Compiler.analyze(Compiler.java:3935)
 at clojure.lang.Compiler.analyzeSeq(Compiler.java:4094)
 at clojure.lang.Compiler.analyze(Compiler.java:3935)
 at clojure.lang.Compiler.access$100(Compiler.java:38)
 at clojure.lang.Compiler$DefExpr$Parser.parse(Compiler.java:366)
 at clojure.lang.Compiler.analyzeSeq(Compiler.java:4106)
 at clojure.lang.Compiler.analyze(Compiler.java:3935)
 at clojure.lang.Compiler.analyzeSeq(Compiler.java:4094)
 at clojure.lang.Compiler.analyze(Compiler.java:3935)
 at clojure.lang.Compiler.analyze(Compiler.java:3908)
 at clojure.lang.Compiler.eval(Compiler.java:4146)
 at clojure.lang.Compiler.load(Compiler.java:4470)
 at clojure.lang.Compiler.loadFile(Compiler.java:4437)
 at clojure.main$load_script__5177.invoke(main.clj:117)
 at clojure.main$script_opt__5209.invoke(main.clj:163)
 at clojure.main$_main__5222.doInvoke(main.clj:227)
 at clojure.lang.RestFn.applyTo(RestFn.java:142)
 at clojure.main.main(Unknown Source)
Caused by: java.lang.NullPointerException
 at clojure.lang.Reflector.invokeInstanceMethod(Reflector.java:26)
 at clojure.core$get_super_and_interfaces__4708.invoke(core_proxy.clj:195)
 at clojure.core$proxy__4729.doInvoke(core_proxy.clj:270)
 at clojure.lang.RestFn.invoke(RestFn.java:445)
 at clojure.lang.Var.invoke(Var.java:335)
 at clojure.lang.AFn.applyToHelper(AFn.java:180)
 at clojure.lang.Var.applyTo(Var.java:444)
 at clojure.lang.Compiler.macroexpand1(Compiler.java:4025)
 at clojure.lang.Compiler.analyzeSeq(Compiler.java:4092)
 ... 25 more



What version are you using?

20081217

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

Please provide any additional information below.

Comment 1 by tuomas.lukka, Jan 27, 2009
(and note that the first expression with Runnable works, the second with Foo is the
one that produces the error).
Comment 2 by richhickey, Apr 12, 2009
Fixed by r1342 - thanks for the report
Status: Fixed

Reported by rwvtveer, Feb 12, 2009

The inlining as provided by r1265 produces verifyerrors on dalvik (android):

W/dalvikvm(  214): VFY: unable to resolve virtual method 6650: Ljava/lang/StringBuilder;.append (C)Ljava/lang/AbstractStringBuilder;
W/dalvikvm(  214): VFY:  rejecting opcode 0x6e at 0x0043
W/dalvikvm(  214): VFY:  rejected Lclojure/core$slurp__4218;.invoke(Ljava/lang/Object;)Ljava/lang/Object;
W/dalvikvm(  214): Verifier rejected class Lclojure/core$slurp__4218;
W/dalvikvm(  214): Exception Ljava/lang/VerifyError; thrown during Lclojure/core__init;.<clinit>
W/dalvikvm(  214): Exception Ljava/lang/RuntimeException; thrown during Lclojure/lang/RT;.<clinit>

As discussed on ml: http://groups.google.com/group/clojure/browse_thread/thread/399090b419f72add/395d3b2b32660211

Comment 1 by r.p.levy, Mar 11, 2009

Are you working on getting Clojure to function on the Android phone?  How can I learn
more about the status of that effort? Thanks! Rob

Comment 2 by rwvtveer, Mar 17, 2009

Without this r1265 changeset and the beans function AOT compiled code works on
dalvik.  Checkout http://github.com/remvee/clojure/ and
http://github.com/remvee/clojurehelloandroid/ if you want to play with clojure on
Android.

Comment 3 by richhickey, Apr 12, 2009

Fixed r1344 - thanks for the report!

Status: Fixed

Reported by rwvtveer, Feb 12, 2009

As discussed on ml: http://groups.google.com/group/clojure/browse_thread/thread/c40b2973109db24/f7c51fddaaec0d3b

The attached patch provides custom constant serialization for:

* base types: String, Integer, Double, Character, Class, Symbol, Keyword and Var
* collections: IPersistentMap, IPersistentVector and IPersistentList/ISeq

If meta data is associated with on the emitted object it will be serialized and attached.  If the object is of a different type the previous method, using 
lispreader, is used.  The *print-initialized* hack is still in place because in a theoretical case it is still needed.

Comment 1  by richhickey, Mar 03, 2009

patch applied (w/some changes) - svn 1322 - thanks
I'd like top revisit this as a multimethod at some point

Status: Fixed