<< Back to previous view

[CLJ-1543] Type tags on argument vector appear to help avoid reflection when used with defn, but not with def foo (fn ...) Created: 30/Sep/14  Updated: 02/Oct/14

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.6
Fix Version/s: None

Type: Defect Priority: Minor
Reporter: Andy Fingerhut Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: interop, typehints


 Description   

I would have expected that both of the Java interop calls below would avoid reflection, but only the first involving f1 does.

Clojure 1.6.0
user=> (set! *warn-on-reflection* true)
true
user=> (defn f1 ^java.util.LinkedList [coll] (java.util.LinkedList. coll))
#'user/f1
user=> (def f2 (fn ^java.util.LinkedList [coll] (java.util.LinkedList. coll)))
#'user/f2
user=> (.size (f1 [2 3 4]))
3
user=> (.size (f2 [2 3 4]))
Reflection warning, NO_SOURCE_PATH:5:1 - reference to field size can't be resolved.
3

Not sure if this has anything to do with CLJ-1232, but was discovered when testing variants of that issue.



 Comments   
Comment by Andy Fingerhut [ 30/Sep/14 9:08 PM ]

What a nice number for a ticket, 1543. The year Copernicus's most celebrated book was published: http://en.wikipedia.org/wiki/Nicolaus_Copernicus

Comment by Jozef Wagner [ 01/Oct/14 4:05 AM ]

Isn't type hinting of arg vector meant only for primitive type hints? AFAIK non-primitive type hints should be on a function name, everything else is non idiomatic.

Comment by Nicola Mometto [ 01/Oct/14 7:05 AM ]

This isn't an issue of arg vector hinting vs function name hinting.
The issue here is that return type hinting cannot be put on anonymous functions but only on defns as the :arglists will be added by defn on the Var's metadata.

This is one of the reasons why I'd like to have that information as a field on the fn rather than as metadata on the Var

Comment by Andy Fingerhut [ 01/Oct/14 10:55 AM ]

Jozef, you may be correct that non-primitive type hints on the argument vector are non idiomatic. Do you have any source for that I could read?

Comment by Tassilo Horn [ 02/Oct/14 12:19 AM ]

Only the version with hints on the argument vectors is documented at http://clojure.org/java_interop#Java Interop-Type Hints. However, in the case you have just one arity (or all arities return a value of the same type) the hint on the var name also works. But the two versions seem to have different semantics. Have a look at CLJ-1232.

Comment by Jozef Wagner [ 02/Oct/14 5:48 AM ]

Type hinting is a very intricate part of Clojure but you can almost always apply a 'place hint on a symbol' idiom. Type hinting on an arg vector must be done only in two cases:

  • primitive hints
  • different return classes for different arities

In the first case, compiler needs type hints when compiling fn* (see [1]), not later, thus you must specify them on arg vector.

Second case, which is the issue discussed here, must be used only when defining with defn. Compiler first looks for the tag in the metadata of a var, and if it does not find one, it has a special case in which it looks for a return class inside :arglist metadata. This is clearly a very special case [2] to handle situations where you have different return classes for different arities. Obviously, using def instead of defn won't create an :arglist metadata for you thus you see a reflection warning. Example:

user=> (def f2 (fn ^java.util.LinkedList [coll] (java.util.LinkedList. coll)))
#'user/f2
user=> (.size (f2 [2 3 4]))
Reflection warning, /tmp/form-init.clj:1:1 - reference to field size can't be resolved.
3
user=> (alter-meta! #'f2 assoc :arglists '(^java.util.LinkedList [coll]))
{:ns #<Namespace user>, :name f2, :file "/tmp/form-init.clj", :column 1, :line 1, :arglists ([coll])}
user=> (.size (f2 [2 3 4]))
3

BTW CLJ-1491 has a discussion slightly relevant to this topic.

[1] https://github.com/clojure/clojure/blob/03cd9d159a2c49a21d464102bb6d6061488b4ea2/src/jvm/clojure/lang/Compiler.java#L5134
[2] https://github.com/clojure/clojure/blob/03cd9d159a2c49a21d464102bb6d6061488b4ea2/src/jvm/clojure/lang/Compiler.java#L3572

Comment by Jozef Wagner [ 02/Oct/14 7:15 AM ]

Andy, I've found sources that speak against my recommendations See CLJ-811 and [1].

[1] https://groups.google.com/d/msg/clojure/b005zQCPxOQ/6G0AlWKKKa0J





[CLJ-1533] Oddity in type tag usage for primInvoke Created: 24/Sep/14  Updated: 03/Oct/14

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.6
Fix Version/s: None

Type: Defect Priority: Minor
Reporter: Andy Fingerhut Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: ft, typehints

Attachments: Text File 0001-CLJ-1533-inject-original-var-form-meta-in-constructe.patch    
Patch: Code and Test
Approval: Triaged

 Description   

Some odd behavior demonstrated in Clojure 1.6.0 REPL below. Why does the (Math/abs (f2 -3)) call issue a reflection warning? It seems like perhaps it should not, given the other examples.

user=> (clojure-version)
"1.6.0"
user=> (set! *warn-on-reflection* true)
true
user=> (defn ^{:tag 'long} f1 [x] (inc x))
#'user/f1
user=> (Math/abs (f1 -3))
2
user=> (defn ^{:tag 'long} f2 [^long x] (inc x))
#'user/f2
user=> (Math/abs (f2 -3))
Reflection warning, NO_SOURCE_PATH:6:1 - call to static method abs on java.lang.Math can't be resolved (argument types: java.lang.Object).
2
user=> (defn ^{:tag 'long} f3 ^long [^long x] (inc x))
#'user/f3
user=> (Math/abs (f3 -3))
2

Cause: invokePrim path does not take into account var or form meta

Approach: apply var and form meta to invokePrim expression

Patch: 0001-CLJ-1533-inject-original-var-form-meta-in-constructe.patch

Screened by: Alex Miller



 Comments   
Comment by Nicola Mometto [ 25/Sep/14 9:47 AM ]

The issue is similar to http://dev.clojure.org/jira/browse/CLJ-1491

Comment by Nicola Mometto [ 25/Sep/14 9:58 AM ]

The root cause was also almost the same, the proposed patch is a superset of the one proposed for CLJ-1491

Comment by Alex Miller [ 25/Sep/14 10:09 AM ]

Can we include 1491 cases in this ticket and mark 1491 a duplicate?

Comment by Alex Miller [ 25/Sep/14 10:09 AM ]

Also needs tests in the patch.

Comment by Nicola Mometto [ 25/Sep/14 10:23 AM ]

Updated the patch with testcases for both issues, I agree that CLJ-1491 should be closed as duplicate





[CLJ-1514] Use qualified class names for return type hints of standard Clojure functions Created: 28/Aug/14  Updated: 28/Aug/14

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.6
Fix Version/s: None

Type: Enhancement Priority: Minor
Reporter: Tassilo Horn Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: enhancement, interop, patch, typehints

Attachments: Text File 0001-Use-fully-qualified-class-names-for-return-type-hint.patch    
Patch: Code

 Description   

The attached patch converts all function return type hints to spell out the class name fully qualified. There are two reasons for doing this:

1. Simple names in return type hints cause the issue described in http://dev.clojure.org/jira/browse/CLJ-1232. That's usually not a problem with return type hints referring to java.lang-classes because those are always imported. However, using `ns-unmap` you can remove them. For example, after `(ns-unmap ns 'String)` in my namespace, `(.length (format "foo = %s") 1)` throws an IllegalArgumentException: Unable to resolve classname: String. By using fully-qualified class names, that problem goes away.

2. tools.analyzer (used by the Clojure lint tool Eastwood) crashes when encountering such a simple-named return type hint. So currently, I cannot lint parts of my project because there's code that calls `clojure.core/format`.



 Comments   
Comment by Alex Miller [ 28/Aug/14 9:34 AM ]

1. that seems like a pretty weird thing to do
2. sounds like an issue with tools.analyzer, not with Clojure?

Comment by Nicola Mometto [ 28/Aug/14 10:46 AM ]

Just to clarify, tools.analyzer(.jvm) can analyze just fine forms in the form (defn x ^Class []) as long as Class is resolvable, whereas it will throw an exception if that function is then used in a namespace where that class is no longer resolvable, which is similar to what Clojure already does, except tools.analyzer.jvm will throw an exception even if the type hint is not used.

Since version 0.5.1 there's an handler that can be provided to change that behaviour, see https://github.com/clojure/tools.analyzer.jvm/blob/master/src/main/clojure/clojure/tools/analyzer/passes/jvm/validate.clj#L232

Comment by Nicola Mometto [ 28/Aug/14 11:02 AM ]

Now a comment regarding this ticket: the patch in this ticket is just a work-around for the issue exposed in http://dev.clojure.org/jira/browse/CLJ-1232, IMHO the correct move would be to actually recognize that issue as a bug rather than as an accepted "limitation" as Rich's comment seems to suggest so that a fix might be commited.

Comment by Tassilo Horn [ 28/Aug/14 1:29 PM ]

@Alex: 1. is not as weird as it sounds at first. For example, consider you have macros that generate complete APIs for something into some new namespace. Then it can make sense to use a real vanilla namespace, i.e., without referring clojure.core and importing java.lang. With 2. I side with Nicola and consider CLJ-1232 a bug.

@Nicola: Today I've used Eastwood (0.1.4) to lint my project. It crashed when it encountered this definition:

(defmacro error
  "Throws an exception with the given message and cause."
  ([msg]
     `(error ~msg nil))
  ([msg cause]
     `(throw (java.lang.Exception. ~msg ~cause))))

(defmacro errorf
  "Throws an exception with the given `msg` and `objs` passed to `format`.
  `msg` is a format string."
  [msg & objs]
  `(error (format ~msg ~@objs)))  ;; This is line 112 where the crash occurs

The message was:

Exception thrown during phase :analyze+eval of linting namespace funnyqt.tg-test
A function, macro, protocol method, var, etc. named clojure.core/format has been used here:
{:file "funnyqt/utils.clj",
 :end-column 19,
 :column 12,
 :line 112,
 :end-line 112}
Wherever it is defined, or where it is called, it has a type of String
This appears to be a Java class name with no package path.
Library tools.analyzer, on which Eastwood relies, cannot analyze such files.
If this definition is easy for you to change, we recommend you prepend it with
a full package path name, e.g. java.net.URI
Otherwise import the class by adding a line like this to your ns statement:
    (:import (java.net URI))

An exception was thrown while analyzing namespace funnyqt.tg-test 
Lint results may be incomplete.  If there are compilation errors in
your code, try fixing those.  If not, check above for info on the
exception.

So it seems it crashes because `format` has a `^String` return type hint. The namespace containing the `errorf` macro above has no modified ns-imports, i.e., all java.lang classes are imported there, too.

Comment by Nicola Mometto [ 28/Aug/14 1:46 PM ]

Tassilo, since `errorf` is a macro, that error is probably caused at the expansion point of that macro in a namespace that unmaps 'String.
If that's not the case, please open a ticket in the eastwood repo

Comment by Tassilo Horn [ 28/Aug/14 2:16 PM ]

Nicola, you are correct. As I've explained above to Alex, I generate APIs in fresh namespaces that don't refer clojure.core and also ns-unmap all java.lang classes, and the generated code also contains `errorf`-forms.

Well, since `ns-unmap` is there, I think it's legit to use it. So that makes CLJ-1232 even more important. But until that gets fixed which requires a common agreement that it is indeed a bug, I'd be very happy if this patch could be accepted. I mean, when it cannot do any harm and doesn't obscure anything but helps at least one person, then why not do it?





[CLJ-1491] External type hint inconsistency between regular functions and primitive functions Created: 05/Aug/14  Updated: 25/Sep/14  Resolved: 25/Sep/14

Status: Closed
Project: Clojure
Component/s: None
Affects Version/s: Release 1.6
Fix Version/s: None

Type: Defect Priority: Minor
Reporter: Gunnar Völkel Assignee: Unassigned
Resolution: Duplicate Votes: 0
Labels: compiler, typehints

Attachments: Text File 0001-preserve-fn-meta-on-invokePrim.patch    
Patch: Code
Approval: Triaged

 Description   

Consider the following example.

(set! *warn-on-reflection* true)

(defn f [n] (java.util.ArrayList. (int n)))

(let [al ^java.util.ArrayList (f 10)]
  (.add al 23))

As expected this does not warn about reflection. The following example shows the same scenario for a primitive function.

(set! *warn-on-reflection* true)

(defn g [^long n] (java.util.ArrayList. n))

(let [al ^java.util.ArrayList (g 10)]
  (.add al 23))
; Reflection warning, NO_SOURCE_PATH:2:3 - call to method add on java.lang.Object can't be resolved (no such method).

So the behavior of external type hints is inconsistent for regular functions and primitive functions.
Most likely, the external type hint information is somehow ignored for primitive functions since the case where they return no primitive value is not treated separately.



 Comments   
Comment by Nicola Mometto [ 05/Aug/14 4:32 AM ]

The following patch preserves the original metadata of the invoke form on the transformed .invokePrim expression

Comment by Alex Miller [ 05/Aug/14 7:40 AM ]

Not challenging the premise at all but workaround:

(let [^java.util.ArrayList al (g 10)]
  (.add al 23))
Comment by Gunnar Völkel [ 05/Aug/14 8:09 AM ]

Well, the example above was already changed such that you can also place the type hint on the binding to check whether that works.
The actual problem arose when using the return value of the function exactly once without an additional binding.

Comment by Jozef Wagner [ 05/Aug/14 10:48 AM ]

Responding to Alex's comment, is there a consensus on which variant is (more) idiomatic? IMHO latter variant seems to be more reliable (as this issue shows, and for primitive hits too), and is consistent with 'place hint on a symbol' idiom which is applied when type hinting vars or fn args.

(let [symbol ^typehint expr] body)
(let [^typehint symbol expr] body)
Comment by Alex Miller [ 05/Aug/14 4:59 PM ]

They have different meanings. Generally the latter covers some cases that the former does not so it's probably the better one. I believe one of the cases is that if expr is a macro, the typehint is lost in the former.

Comment by Nicola Mometto [ 25/Sep/14 9:59 AM ]

The patch for http://dev.clojure.org/jira/browse/CLJ-1533?jwupdated=61127&focusedCommentId=35814 fixes this issue and more and should be preferred over this

Comment by Alex Miller [ 25/Sep/14 10:31 AM ]

Dupe of CLJ-1533





[CLJ-1433] proxy-super calls generally use reflection Created: 28/May/14  Updated: 28/May/14

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.6
Fix Version/s: None

Type: Defect Priority: Minor
Reporter: Greg Chapman Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: typehints


 Description   

For example:

user=> (proxy [java.util.BitSet] []
  (flip [bitIndex]
    (proxy-super flip bitIndex)))
Reflection warning, NO_SOURCE_PATH:73:5 - call to method flip can't be resolved (target class is unknown).

I believe this issue might be fixed by simply adding type-hint metadata to the 'this symbol emitted by the proxy macro. I have not tried this change, but this macro seems to indicate it should work:

(defmacro proxy-super-cls [cls meth & args]
  (let [thissym (with-meta (gensym) {:tag cls})]
    `(let [~thissym ~'this]
      (proxy-call-with-super (fn [] (. ~thissym ~meth ~@args)) ~thissym ~(name meth))
    )))
;;;;;;;;;;;;;;;;;;;;;;
user=> (proxy [java.util.BitSet] []
  (flip [bitIndex]
    (proxy-super-cls java.util.BitSet flip bitIndex)))
#<BitSet$ff19274a {}>





[CLJ-1432] NullPointerException on function with primitive result declaration Created: 26/May/14  Updated: 30/May/14

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.6
Fix Version/s: None

Type: Enhancement Priority: Minor
Reporter: Gunnar Völkel Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: primitives, typehints


 Description   

The following minimal example shows the error:

(defn f ^double [])
(f)
=> NullPointerException

When decompiling the function `f` I found the following return expression:

return null.doubleValue();

This happened in a Java interop scenario where the called Java method had no return value but was in the return position of the primitive Clojure function.
The compiler should check for `null` on compilation.

Another example - calling a method with void return as the last expression fails in a similar way:

(defn f ^double [^SomeClassToAvoidRuntimeReflection obj, x, y]
  (.someMethod obj, x, y))
(f obj, x, y)
=> NullPointerException


 Comments   
Comment by Alex Miller [ 26/May/14 11:19 PM ]

What do you expect to happen in this case? You declared a function as returning a double but didn't return one.

Comment by Gunnar Völkel [ 27/May/14 8:48 AM ]

Since this is only the minimal example the error is relatively easy to spot.
Consider the following small example with Java interop:

(defn f ^double [^SomeClassToAvoidRuntimeReflection obj, x, y]
  (.someMethod obj, x, y))
(f obj, x, y)
=> NullPointerException

In this example it is much harder to find the reason for the NPE because you'd first suspect `obj` to be `null`.

I expect a check in the compiler at the point where "return null.doubleValue();" is emitted, followed by an error message, e.g. "Primitive return value of type 'double' expected, but no value is returned.".

Comment by Jozef Wagner [ 28/May/14 2:15 AM ]

Your second example seems perfectly OK to me, compiler should not report any error and NPE check must be at runtime.

Comment by Gunnar Völkel [ 28/May/14 2:46 AM ]

@Jozef: No, you are wrong. The compiler infers via reflection at compile time that the called method does not return a value and emits "return null.doubleValue()". So this can and should be reported as explicit error at compile time. I added a typehint to make it clear that there is no runtime reflection involved.
You would be right, if the compiler emitted something like "return somevar.doubleValue();" because then at compile time there is no knowledge about a possible "null" value.

Comment by Andy Fingerhut [ 28/May/14 10:00 AM ]

Gunnar, in your example, is the method 'someMethod' declared to return void, or something else? Adding that info to your example might help clarify it.

Comment by Jozef Wagner [ 29/May/14 2:26 AM ]

Gunnar, the second example was ambiguous and strayed away the discussion. Anyway, whether returning wrong type is through the native method or not, it is a user error in the first place. Right now it is reported at runtime. For me this ticket should be a minor enhancement instead of defect.

Comment by Gunnar Völkel [ 30/May/14 4:40 AM ]

Yes, the reason is a user error. But one that is harder to debug than necessary.
Also, it is clearly a defect since emitting 'null.doubleValue()' can not be considered as a valid compilation.

Andy, yes 'someMethod' is declared to return void. I'd edit the original ticket text to add the example and the java method return value information, but it seems jira does not let me.

Comment by Alex Miller [ 30/May/14 8:35 AM ]

I added the second example (with clarifying void comment) to the description.





[CLJ-1422] Recur around try boxes primitives Created: 14/May/14  Updated: 28/Jul/14

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.5, Release 1.6
Fix Version/s: None

Type: Enhancement Priority: Minor
Reporter: Kyle Kingsbury Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: compiler, performance, typehints


 Description   

Primitive function and recur variables can't pass through a (try) cleanly; they're boxed to Object instead. This causes reflection warnings for fns or loops that use primitive types.

user=> (set! *warn-on-reflection* true)
true
 
user=> (fn [] (loop [t 0] (recur t)))
#<user$eval676$fn__677 user$eval676$fn__677@3d80023a>
 
user=> (fn [] (loop [t 0] (recur (try t))))
NO_SOURCE_FILE:1 recur arg for primitive local: t is not matching primitive, had: Object, needed: long
Auto-boxing loop arg: t
#<user$eval680$fn__681 user$eval680$fn__681@5419323a>

user=> (fn [^long x] (recur (try x)))
NO_SOURCE_FILE:1 recur arg for primitive local: x is not matching primitive, had: Object, needed: long

CompilerException java.lang.IllegalArgumentException:  recur arg for primitive local: x is not matching primitive, had: Object, needed: long, compiling:(NO_SOURCE_PATH:1:1)


 Comments   
Comment by David James [ 15/Jun/14 10:27 PM ]

Without commenting on the most desirable behavior, the following code does not cause reflection warnings:

user=> (set! *warn-on-reflection* true)
true
user=> (fn [] (loop [t 0] (recur (long (try t)))))
#<user$eval673$fn__674 user$eval673$fn__674@4e56c411>
Comment by Nicola Mometto [ 16/Jun/14 6:33 AM ]

Similar ticket http://dev.clojure.org/jira/browse/CLJ-701

Comment by Kevin Downey [ 21/Jul/14 6:59 PM ]

try/catch in the compiler only implements Expr, not MaybePrimitiveExpr, looking at extending TryExpr with MaybePrimitiveExpr it seems simple enough, but it turns out recur analyzes it's arguments in the statement context, which causes (try ...) to essentially wrap itself in a function like ((fn [] (try ...))), at which point it is an invokeexpr which is much harder to add maybeprimitiveexpr too and it reduces to the same case as CLJ-701

Comment by Kevin Downey [ 22/Jul/14 9:27 PM ]

http://dev.clojure.org/jira/browse/CLJ-701 has a patch that I think solves this

Comment by Alex Miller [ 28/Jul/14 1:56 PM ]

Should I dupe this to CLJ-701?

Comment by Kevin Downey [ 28/Jul/14 5:22 PM ]

if you want the fixes for try out of the return context to be part of CLJ-701 then yes it is a dupe, if you are unsure or would prefer 701 to stay more focused (my patch may not be acceptable, or may be too large and doing too much) then no it wouldn't be a dupe. I sort of took it on myself to solve both in the patch on CLJ-701 because I came to CLJ-701 via Nicola's comment here, and the same compiler machinery can be used for both.

I think the status is pending on the status of CLJ-701.





[CLJ-1378] Hints don't work with #() form of function Created: 11/Mar/14  Updated: 29/Aug/14  Resolved: 29/Aug/14

Status: Closed
Project: Clojure
Component/s: None
Affects Version/s: None
Fix Version/s: Release 1.7

Type: Enhancement Priority: Major
Reporter: Roy Varghese Assignee: Unassigned
Resolution: Completed Votes: 1
Labels: interop, typehints

Attachments: File clj-1378.diff     File clj-1378-v2.diff    
Patch: Code and Test
Approval: Ok

 Description   

Example showing how a local fn can be hinted but an anonymous function cannot:

;; OK
user> (let [ex (java.util.concurrent.Executors/newFixedThreadPool 1)
            f (fn [])]
        (.submit ex ^Runnable f))
nil
;; ERROR - this should work the same as the previous
user> (let [ex (java.util.concurrent.Executors/newFixedThreadPool 1)]
        (.submit ex #()))
CompilerException java.lang.IllegalArgumentException: More than one matching method found: submit, compiling:(/private/var/folders/7r/_1fj0f517rgcxwx79mn79mfc0000gn/T/form-init7901279404687292754.clj:3:9)

Cause: Functions have metadata, but Compiler does not look in them for type hints. Var expressions and local bindings use :tag metadata to override return of getJavaClass(). Compiler parses #() into a FnExpr, which always return AFunction as its class.

Proposed: Change FnExpr.getJavaClass() to return tag as type if it is available.

Patch: clj-1378-v2.diff

Screened by: Alex Miller



 Comments   
Comment by Jozef Wagner [ 12/Mar/14 4:03 AM ]

Functions do have metadata, but Compiler does not look in them for type hints.

user=> (with-meta #() {:foo :bar})
#<clojure.lang.AFunction$1@779325ee>

When compiler is determining which native method to use, it matches method signature with classes of given args. There is a getJavaClass() method in Compiler.java which returns a class for given expression. Vars expressions and local bindings use :tag metadata to override this class, but most other expressions don't. Compiler parses #() into a FnExpr, which always return AFunction as its class.

Most of time this approach is OK, as AFunction implements Runnable and Callable so there is no need for type hint. However, in this particular case, there are overrides for both Runnable and Callable, and as AFunction can be either of them, the expression is ambiguous.

Comment by Jozef Wagner [ 12/Mar/14 4:17 AM ]

Patch added, following expression will now run without error

(.submit (java.util.concurrent.Executors/newCachedThreadPool) ^Runnable #())
Comment by Alex Miller [ 12/Mar/14 9:34 AM ]

Could you add a test to the patch?

Comment by Jozef Wagner [ 12/Mar/14 2:53 PM ]

Attached patch clj-1378-v2.diff which contains both fix and test.





[CLJ-1308] extend-type doesn't type-hint correctly as promised by the doc when the class is determined at run-time Created: 15/Dec/13  Updated: 05/Feb/14

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: None
Fix Version/s: None

Type: Defect Priority: Minor
Reporter: Nicola Mometto Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: typehints


 Description   

extend-type works with non-constant expressions as its type:

(extend-type (class 1) proto (protof [this]))

However, in this case this will get tagged with `(class 1)`, which is clearly wrong; the doc explicitely states that the args will be proberly type-hinted: "[..] Propagates the class as a type hint on the first argument of all fns."

I don't know if extend-type is not supposed to work with non-constant Classes, in which case it should be stated in the doc or if the current behaviour is wrong.






[CLJ-1256] Support type-hinted overrides of function parameters Created: 06/Sep/13  Updated: 09/Sep/13

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: None
Fix Version/s: None

Type: Enhancement Priority: Major
Reporter: Mike Anderson Assignee: Unassigned
Resolution: Unresolved Votes: 1
Labels: compiler, interop, typehints


 Description   

Problem: in many cases, the Clojure compiler has enough information about the type of a function argument to statically emit maximally efficient code on the JVM (i.e. without instance? checks, type casts or other forms of dynamic polymorphic dispatch). We are currently unable to do so in Clojure, which pushes developers with strong performance requirements to use some unidiomatic or convoluted workarounds.

Proposal is simply to allow functions to take type-hinted overloads of function arguments, e.g.

(defn foo
([^double x] (Math/floor x))
([^float x] (Math/floor (double x)))
([^String s] (count s)))

An "Object" version of the code with the correct arity will always be emitted, which will maintain compatibility with the IFn interface and ensure that the function can still be used in dynamic / interactive contexts. If the "Object" version is not explicitly provided, then it will be generated to use instance? checks that subsequently delegate to the appropriate typed version of the function (or throw an InvalidArgumentException if no match is found).

Matching rules would be the same as Java.

This will be backwards compatible with all existing uses of defn. In particular, it should extend / enhance / supercede the existing handling of primitive functions.

In the future, this technique might be used alongside core.typed to ensure that the most efficient function version is chosen based on type inference.






[CLJ-1232] Functions with non-qualified return type hints force import of hinted classes when called from other namespace Created: 18/Jul/13  Updated: 02/Oct/14

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.5, Release 1.6, Release 1.7
Fix Version/s: None

Type: Defect Priority: Major
Reporter: Tassilo Horn Assignee: Unassigned
Resolution: Unresolved Votes: 4
Labels: compiler, typehints

Attachments: Text File 0001-auto-qualify-arglists-class-names.patch     Text File 0001-throw-on-non-qualified-class-names-that-are-not-auto.patch    
Patch: Code
Approval: Triaged

 Description   

You can add a type hint to function arglists to indicate the return type of a function like so.

user> (import '(java.util List))
java.util.List
user> (defn linkedlist ^List [] (java.util.LinkedList.))
#'user/linkedlist
user> (.size (linkedlist))
0

The problem is that now when I call `linkedlist` exactly as above from another namespace, I'll get an exception because java.util.List is not imported in there.

user> (in-ns 'user2)
#<Namespace user2>
user2> (refer 'user)
nil
user2> (.size (linkedlist))
CompilerException java.lang.IllegalArgumentException: Unable to resolve classname: List, compiling:(NO_SOURCE_PATH:1:1)
user2> (import '(java.util List)) ;; Too bad, need to import List here, too.
java.util.List
user2> (.size (linkedlist))
0

There are two workarounds: You can import the hinted type also in the calling namespace, or you always use fully qualified class names for return type hints. Clearly, the latter is preferable.

But clearly, that's a bug that should be fixed. It's not in analogy to type hints on function parameters which may be simple-named without having any consequences for callers from other namespaces.



 Comments   
Comment by Andy Fingerhut [ 16/Apr/14 3:47 PM ]

To make sure I understand, Nicola, in this ticket you are asking that the Clojure compiler change behavior so that the sample code works correctly with no exceptions, the same way as it would work correctly without exceptions if one of the workarounds were used?

Comment by Tassilo Horn [ 17/Apr/14 12:18 AM ]

Hi Andy. Tassilo here, not Nicola. But yes, the example should work as-is. When I'm allowed to use type hints with simple imported class names for arguments, then doing so for return values should work, too.

Comment by Rich Hickey [ 10/Jun/14 10:41 AM ]

Type hints on function params are only consumed by the function definition, i.e. in the same module as the import/alias. Type hints on returns are just metadata, they don't get 'compiled' and if the metadata is not useful to consumers in other namespaces, it's not a useful hint. So, if it's not a type in the auto-imported set (java.lang), it should be fully qualified.

Comment by Alex Miller [ 10/Jun/14 11:55 AM ]

Based on Rich's comment, this ticket should probably morph into an enhancement request on documentation, probably on http://clojure.org/java_interop#Java Interop-Type Hints.

Comment by Andy Fingerhut [ 10/Jun/14 3:13 PM ]

I would suggest something like the following for a documentation change, after this part of the text on the page Alex links in the previous comment:

For function return values, the type hint can be placed before the arguments vector:

(defn hinted
(^String [])
(^Integer [a])
(^java.util.List [a & args]))

-> #user/hinted

If the return value type hint is for a class that is outside of java.lang, which is the part auto-imported by Clojure, then it must be a fully qualified class name, e.g. java.util.List, not List.

Comment by Nicola Mometto [ 10/Jun/14 4:02 PM ]

I don't understand why we should enforce this complexity to the user.
Why can't we just make the Compiler (or even defn itself) update all the arglists tags with properly resolved ones? (that's what I'm doing in tools.analyzer.jvm)

Comment by Alexander Kiel [ 19/Jul/14 10:02 AM ]

I'm with Nicola here. I also think that defn should resolve the type hint according the imports of the namespace defn is used in.

Comment by Max Penet [ 22/Jul/14 7:06 AM ]

Same here, I was bit by this in the past. The current behavior is clearly counterintuitive.

Comment by Nicola Mometto [ 28/Aug/14 12:58 PM ]

Attached two patches implementing two different solutions:

  • 0001-auto-qualify-arglists-class-names.patch makes the compiler automatically qualify all the tags in the :arglists
  • 0001-throw-on-non-qualified-class-names-that-are-not-auto.patch makes the compiler throw an exception for all public defs whose return tag is a symbol representing a non-qualified class that is not in the auto-import list (approach proposed in IRC by Alex Miller)
Comment by Tassilo Horn [ 29/Aug/14 1:49 AM ]

For what it's worth, I'd prefer the first patch because the second doesn't help in situations where the caller lives in a namespace where the called function's return type hinted class is `ns-unmap`-ed. And there a good reasons for doing that. For example, Process is a java.lang class and Process is a pretty generic name. So in some namespace, I want to define my own Process deftype or defrecord. Without unmapping 'Process first to get rid of the java.lang.Process auto-import, I'd get an exception:

user> (deftype Process [])
IllegalStateException Process already refers to: class java.lang.Process in namespace: user  clojure.lang.Namespace.referenceClass (Namespace.java:140)

Now when I call some function from some library that has a `^Process` return type hint (meaning java.lang.Process there), I get the same exception as in my original report.

I can even get into troubles when only using standard Clojure functions because those have `^String` and `^Class` type hints. IMO, Class is also a pretty generic name I should be able to name my custom deftype/defrecord. And I might also want to have a custom String type/record in my astrophysics system.

Comment by Andy Fingerhut [ 30/Sep/14 4:39 PM ]

Not sure whether the root cause of this behavior is the same as the example in the description or not, but seems a little weird that even for fully qualified Java class names hinting the arg vector, it makes a difference whether it is done with defn or def:

Clojure 1.6.0
user=> (set! *warn-on-reflection* true)
true
user=> (defn f1 ^java.util.LinkedList [coll] (java.util.LinkedList. coll))
#'user/f1
user=> (def f2 (fn ^java.util.LinkedList [coll] (java.util.LinkedList. coll)))
#'user/f2
user=> (.size (f1 [2 3 4]))
3
user=> (.size (f2 [2 3 4]))
Reflection warning, NO_SOURCE_PATH:5:1 - reference to field size can't be resolved.
3
Comment by Alex Miller [ 30/Sep/14 6:21 PM ]

Andy, can you file that as a separate ticket?

Comment by Andy Fingerhut [ 30/Sep/14 9:08 PM ]

Created ticket CLJ-1543 for the issue raised in my comment earlier on 30 Sep 2014.

Comment by Andy Fingerhut [ 01/Oct/14 12:38 PM ]

Tassilo (or anyone), is there a reason to prefer putting the tag on the argument vector in your example? It seems that putting it on the Var name instead avoids this issue:

user=> (clojure-version)
"1.6.0"
user=> (set! *warn-on-reflection* true)
true
user=> (import '(java.util List))
java.util.List
user=> (defn ^List linkedlist [] (java.util.LinkedList.))
#'user/linkedlist
user=> (.size (linkedlist))
0
user=> (ns user2)
nil
user2=> (refer 'user)
nil
user2=> (.size (linkedlist))
0

I suppose that only allows a single type tag, rather than an independent one for each arity.

Comment by Tassilo Horn [ 02/Oct/14 3:16 AM ]

I wasn't aware of the fact that you can put it on the var's name. That's not documented at http://clojure.org/java_interop#Java Interop-Type Hints. But IMHO the documented version with putting the tag on the argument vector is more general since it supports different return type hints for the different arity version. In any case, if both forms are permitted then they should be equivalent in the case the function has only one arity.





[CLJ-1231] fn and letfn don't support hinting the function's return type Created: 17/Jul/13  Updated: 03/Sep/13

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.5
Fix Version/s: None

Type: Defect Priority: Minor
Reporter: Tassilo Horn Assignee: Unassigned
Resolution: Unresolved Votes: 1
Labels: typehints


 Description   

You can add type hints to the argument vector(s) of `defn` to declare the return type of a function like so (with warn-on-reflection being true):

user> (defn foo ^String [s] s)
#'user/foo
user> (.substring (foo "hallo") 1 2)
"a"

But sadly, the same doesn't work with `fn` and `letfn`:

user> (.substring ((fn ^String [s] s) "hallo") 1 2)
Reflection warning, NO_SOURCE_PATH:1:1 - call to substring can't be resolved.
"a"
user> (letfn [(foo ^String [s] s)]
	(.substring (foo "hallo") 1 2))
Reflection warning, NO_SOURCE_PATH:2:7 - call to substring can't be resolved.
"a"

I don't see why this feature is available to `defn` but not to `fn` and `letfn`. I even consider it a kind of defect, because anything else including :pre/:post are also supported by the latter two, so the support for hinting the return type should be there simply because of analogy.






[CLJ-1212] Silent truncation/downcasting of primitive type on reflection call to overloaded method (Math/abs) Created: 28/May/13  Updated: 05/Feb/14

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.5
Fix Version/s: None

Type: Defect Priority: Major
Reporter: Matthew Willson Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: primitives, typehints
Environment:

Clojure 1.5.1
OpenJDK Runtime Environment (IcedTea6 1.12.5) (6b27-1.12.5-0ubuntu0.12.04.1)



 Description   

I realise relying on reflection when calling these kinds of methods isn't a great idea performance-wise, but it shouldn't lead to incorrect or dangerous behaviour.

Here it seems to trigger a silent downcast of the input longs, giving a truncated integer output:

user> (defn f [a b] (Math/abs (- a b)))
Reflection warning, NO_SOURCE_PATH:1:15 - call to abs can't be resolved.
#'user/f
user> (f 1000000000000 2000000000000)
727379968
user> (class (f 1000000000000 2000000000000))
java.lang.Integer
user> (defn f [^long a ^long b] (Math/abs (- a b)))
#'user/f
user> (f 1000000000000 2000000000000)
1000000000000
user> (class (f 1000000000000 2000000000000))
java.lang.Long



 Comments   
Comment by Matthew Willson [ 28/May/13 12:50 PM ]

For an even simpler way to replicate the issue:

user> (#(Math/abs %) 1000000000000)
Reflection warning, NO_SOURCE_PATH:1:3 - call to abs can't be resolved.
727379968

Comment by Andy Fingerhut [ 28/May/13 1:36 PM ]

I was able to reproduce the behavior you see with these Java 6 JVMs on Ubuntu 12.04.2:

java version "1.6.0_27"
OpenJDK Runtime Environment (IcedTea6 1.12.5) (6b27-1.12.5-0ubuntu0.12.04.1)
OpenJDK 64-Bit Server VM (build 20.0-b12, mixed mode)

java version "1.6.0_45"
Java(TM) SE Runtime Environment (build 1.6.0_45-b06)
Java HotSpot(TM) 64-Bit Server VM (build 20.45-b01, mixed mode)

However, I tried two Java 7 JVMs, and it gave the following behavior which looks closer to what you would hope for. I do not know what is the precise difference between Java 6 and Java 7 that leads to this behavior difference, but this is some evidence that this has something to do with Java 6 vs. Java 7.

user=> (set! warn-on-reflection true)
true
user=> (defn f [a b] (Math/abs (- a b)))
Reflection warning, NO_SOURCE_PATH:1:15 - call to abs can't be resolved.
#'user/f
user=> (f 1000000000000 2000000000000)
1000000000000
user=> (class (f 1000000000000 2000000000000))
java.lang.Long

Above behavior observed with Clojure 1.5.1 on these JVMs:

Ubuntu 12.04.2 plus this JVM:
java version "1.7.0_21"
Java(TM) SE Runtime Environment (build 1.7.0_21-b11)
Java HotSpot(TM) 64-Bit Server VM (build 23.21-b01, mixed mode)

Mac OS X 10.8.3 plus this JVM:
java version "1.7.0_15"
Java(TM) SE Runtime Environment (build 1.7.0_15-b03)
Java HotSpot(TM) 64-Bit Server VM (build 23.7-b01, mixed mode)

Comment by Matthew Willson [ 29/May/13 5:17 AM ]

Ah, interesting.
Maybe it's a difference in the way the reflection API works in java 7?

Here's the bytecode generated incase anyone's curious:

public java.lang.Object invoke(java.lang.Object);
Code:
0: ldc #14; //String java.lang.Math
2: invokestatic #20; //Method java/lang/Class.forName:(Ljava/lang/String;)Ljava/lang/Class;
5: ldc #22; //String abs
7: iconst_1
8: anewarray #24; //class java/lang/Object
11: dup
12: iconst_0
13: aload_1
14: aconst_null
15: astore_1
16: aastore
17: invokestatic #30; //Method clojure/lang/Reflector.invokeStaticMethod:(Ljava/lang/Class;Ljava/lang/String;[Ljava/lang/Object;)Ljava/lang/Object;
20: areturn

Comment by Matthew Willson [ 29/May/13 5:20 AM ]

Just an idea (and maybe this is what's happening under java 7?) but given it's a static method and all available overloaded variants are presumably known at compile time, perhaps it could generate code along the lines of:

(cond
(instance? Long x) (Math/abs (long x))
(instance? Integer x) (Math/abs (int x))
;; ...
)

Comment by Andy Fingerhut [ 29/May/13 3:19 PM ]

In Reflector.java method invokeStaticMethod(Class c, String methodName, Object[] args) there is a call to getMethods() followed by a call to invokeMatchingMethod(). getMethods() returns the 4 java.lang.Math/abs methods in different orders on Java 6 and 7, causing invokeMatchingMethod() to pick a different one on the two JVMs:

java version "1.6.0_39"
Java(TM) SE Runtime Environment (build 1.6.0_39-b04)
Java HotSpot(TM) 64-Bit Server VM (build 20.14-b01, mixed mode)

user=> (pprint (seq (clojure.lang.Reflector/getMethods java.lang.Math 1 "abs" true)))
(#<Method public static int java.lang.Math.abs(int)>
#<Method public static long java.lang.Math.abs(long)>
#<Method public static float java.lang.Math.abs(float)>
#<Method public static double java.lang.Math.abs(double)>)
nil

java version "1.7.0_21"
Java(TM) SE Runtime Environment (build 1.7.0_21-b11)
Java HotSpot(TM) 64-Bit Server VM (build 23.21-b01, mixed mode)

user=> (pprint (seq (clojure.lang.Reflector/getMethods java.lang.Math 1 "abs" true)))
(#<Method public static double java.lang.Math.abs(double)>
#<Method public static float java.lang.Math.abs(float)>
#<Method public static long java.lang.Math.abs(long)>
#<Method public static int java.lang.Math.abs(int)>)
nil

That might be a sign of undesirable behavior in invokeMatchingMethod() that is too dependent upon the order of methods given to it.

As you mention, type hinting is good for avoiding the significant performance hit of reflection.





[CLJ-1136] Type hinting for array classes does not work in binding forms Created: 20/Dec/12  Updated: 03/Sep/13

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.4, Release 1.5
Fix Version/s: None

Type: Defect Priority: Major
Reporter: Luke VanderHart Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: interop, typehints
Environment:

replicated on OpenJDK 7u9 on Ubuntu 12.04, and Hotspot 1.6.0_37 on OSX Lion



 Description   

Type hints don't work as expected in binding forms.

The following form results in a reflection warning:

(let [^{:tag (Class/forName "[Ljava.lang.Object;")} a (make-array Object 2)]
(aget a 0))

However, hinting does appear to work correctly on vars:

(def ^{:tag (Class/forName "[Ljava.lang.Object;")} a (make-array Object 2))
(aget a 0) ;; no reflection warning



 Comments   
Comment by Ghadi Shayban [ 20/Dec/12 10:51 PM ]

It's a little more insidious than type hinting: the compiler doesn't evaluate metadata in the binding vec.

This doesn't throw the necessary exception...

(let [^{:foo (Class/forName "not real")} bar 42]
bar)

neither this...

(let [^{gyorgy ligeti} a 42]
a)

Gyorgy Ligeti never resolves.

These two equivalent examples don't reflect:
(let [^objects a (make-array Object 2)]
(aget a 0))

(let [a ^objects (make-array Object 2)]
(aget a 0))

Comment by Ghadi Shayban [ 21/Dec/12 11:09 AM ]

On only the left-hand side of a local binding, metadata on a symbol is not analyzed or evaluated.





[CLJ-1080] Eliminate many uses of reflection in Clojure code Created: 30/Sep/12  Updated: 10/Oct/13

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.4, Release 1.5
Fix Version/s: None

Type: Enhancement Priority: Minor
Reporter: Andy Fingerhut Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: performance, typehints

Attachments: Text File clj-1080-v5.txt    
Patch: Code

 Description   

There are dozens of uses of reflection in Clojure code that can be eliminated by adding suitable type hints. This patch adds the necessary type hints for most of those.



 Comments   
Comment by Andy Fingerhut [ 30/Sep/12 11:39 AM ]

Patch clj-1080-eliminate-many-reflection-warnings-patch-v1.txt dated Sep 30 2012 adds type hints to eliminate many uses of reflection in Clojure core code.

Comment by Andy Fingerhut [ 14/Nov/12 1:26 PM ]

clj-1080-eliminate-many-reflection-warnings-patch-v2.txt dated Nov 14 2012 is identical to the previous patch (to be removed soon), except it applies cleanly to latest master.

Comment by Andy Fingerhut [ 07/Feb/13 8:54 AM ]

clj-1080-eliminate-many-reflection-warnings-patch-v3.txt dated Feb 7 2013 is identical to the previous patch (to be removed soon), except it applies cleanly to latest master. One type hint in the patch was added due to a different change, and was no longer needed in the patch.

Comment by Andy Fingerhut [ 09/Sep/13 11:37 PM ]

Patch clj-1080-v4.txt eliminates many, but not all, uses of reflection. To avoid overlap with CLJ-1259, it does not touch pprint or any of the files loaded from pprint.clj. See CLJ-1259 for those.

Comment by Andy Fingerhut [ 10/Oct/13 12:22 AM ]

Patch clj-1080-v5.txt eliminates many, but not all, uses of reflection. It does not touch pprint or any of the files loaded from pprint.clj – see CLJ-1259 for those. Similarly see CLJ-1277 for elimination of reflection in instant.clj





[CLJ-957] Typehints for variadic methods in deftype fail to compile Created: 22/Mar/12  Updated: 03/Sep/13

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.3, Release 1.4
Fix Version/s: None

Type: Defect Priority: Minor
Reporter: Chris Gray Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: deftype, typehints

Attachments: Text File 0001-Allow-for-typehinting-of-method-signatures-in-deftyp.patch     Text File clj-957-allow-typehinting-of-method-signatures-in-deftype-patch2.txt    
Patch: Code and Test

 Description   

When a method defined by a protocol has multiple typehinted signatures, it is impossible to implement them using deftype because deftype throws away the typehints. The compiler then looks for the proper signatures (i.e. with typehints) and throws an exception when it can't find them.



 Comments   
Comment by Chris Gray [ 22/Mar/12 5:41 PM ]

Clojure-dev discussion started here: http://groups.google.com/group/clojure-dev/browse_thread/thread/1f106a21ec1ce3de

Comment by Andy Fingerhut [ 19/Aug/12 4:31 AM ]

Patch clj-957-allow-typehinting-of-method-signatures-in-deftype-patch2.txt dated Aug 19 2012 is identical to Chris Gray's patch 0001-Allow-for-typehinting-of-method-signatures-in-deftyp.patch dated Mar 22, 2012, except it has some updated context lines so that it applies cleanly to latest master.





[CLJ-938] Output of clojure.reflect is not suitable for type hints Created: 23/Feb/12  Updated: 03/Sep/13

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.3
Fix Version/s: None

Type: Defect Priority: Minor
Reporter: Brandon Bloom Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: reflection, typehints


 Description   

I'm trying to write a macro which generate reify forms using some reflection.

clojure.reflect produces type symbols similar to 'java.util.Iterator<>

Unfortunately, the compiler doesn't accept the <> syntax in type hints:

(reify Iterable (^{:tag java.util.Iterator} iterator [this] nil)) ; works

(reify Iterable (^{:tag java.util.Iterator<>} iterator [this] nil)) ;; fails
CompilerException java.lang.RuntimeException: java.lang.ClassNotFoundException: java.util.Iterator<>, compiling:(NO_SOURCE_PATH:1325)

It seems like the compiler should understand the <> just enough to strip it, rather than reject it. This would make it much easier to write correct macros involving type hinting and reflection.

The workaround I have been using is:

(defn hint
"clojure.reflect demarks generics with angle brackets, but
the compiler does not support generics in type hints"
[obj tag]
(let [tag (-> tag .toString (.replace "<>" "") symbol)]
(with-meta obj {:tag tag}))



 Comments   
Comment by Brandon Bloom [ 24/Feb/12 1:37 AM ]

I'm sorry, I got this wrong earlier. The problem is real, but it's arrays, not generics.

My workaround is useless... :-/





[CLJ-919] cannot create anonymous primitive functions Created: 27/Jan/12  Updated: 03/Sep/13

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.3
Fix Version/s: None

Type: Defect Priority: Major
Reporter: Ben Mabey Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: typehints


 Description   

Primitive functions only work (e.g. return primitive types) when defined with `defn`. An equivalent function created with `fn` does not behave the same way as when created with `defn`. For example:

(definterface IPrimitiveTester
(getType [^int x])
(getType [^long x])
(getType [^float x])
(getType [^double x])
(getType [^Object x]))

(deftype PrimitiveTester []
IPrimitiveTester
(getType [this ^int x] :int)
(getType [this ^long x] :long)
(getType [this ^float x] :float)
(getType [this ^double x] :double)
(getType [this ^Object x] :object))

(defmacro pt [x]
`(.getType (PrimitiveTester.) ~x))

(defn with-defn ^double [^double x]
(+ x 0.5))

(pt (with-defn 1.0)) ; => :double

(let [a (fn ^double [^double x] (+ x 0.5))]
(pt (a 0.1))) ; => :object

Please see the discussion on the mailing list for more details and thoughts on what is happening:
http://groups.google.com/group/clojure/browse_thread/thread/d83c8643a7c7d595?hl=en






[CLJ-140] Single :tag for type hints conflates value's type with type of return value from an invoke Created: 01/Jul/09  Updated: 03/Sep/13

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: None
Fix Version/s: None

Type: Enhancement Priority: Major
Reporter: Anonymous Assignee: Unassigned
Resolution: Unresolved Votes: 0
Labels: typehints


 Description   

The value of a Var can be operated on directly, or, if it is a fn, it can be invoked and the resulting value operated on. :tag metadata on a Var is used to provide a type hint to the compiler to avoid reflection. Having a single metadata key for this two distinct uses makes it possible (even easy, if unlikely) to create a situation where type-hinting the value causes a ClassCastException on an operation on the invocation return value, or the reverse. The only obvious solution is two use different keys for the two uses.



 Comments   
Comment by Assembla Importer [ 24/Aug/10 3:51 AM ]

Converted from http://www.assembla.com/spaces/clojure/tickets/140





Generated at Wed Oct 22 16:57:03 CDT 2014 using JIRA 4.4#649-r158309.