richhickey/clojure-clr

A port of Clojure to the CLR, part of the Clojure project

clojure-clr's Issues

try/catch behavior different from Clojure JVM

In the Clojure JVM REPL:
user=> (try (throw (Exception. "Got it.")) (catch Exception ex (println ex)))
#
nil

In the Clojure CLR REPL:
user=> (try (throw (Exception. "Got it.")) (catch Exception ex (println ex)))
.... throws an exception, snipped stack trace because of easy reproduction ...
user=> (try (do (throw (Exception. "Got it.")) 1) (catch Exception ex (println ex)))
#<Exception System.Exception: Got it.
at lambda_method(Closure )>
nil

[EDIT: changed the first line in the CLR REPL so it's identical to the line in the JVM REPL.]

(= -some-System.Collections.Hashtable { .... }) is false when it should be true

With the arguments in the other order, it works okay, i.e. APersistentMap.Equals is okay. The problem is with S.C.Hashtable.Equals.

Compiler: Compile lambdas to a specific delegate type

Allows us to use Invoke instead of DynamicInvoke. A very small improvement, mostly likely.

Implement letfn*

Implement BigDecimal

We are using java.math.BigDecimal from vjslib.dll, part of the Visual J# Redistributable package. Need to get rid of this to run on Mono. Also, it is based on an older point release of the Java library and so is missing some needed functionality.

clojure.lang.Compiler/LOCAL_ENV not public

clojure.lang.Compiler/LOCAL_ENV is public in Clojure, but internal in ClojureCLR. This means that tricks like debug-repl https://gist.github.com/256129/981be4991d4d47fe856476b4c00d046c31823bd6
are not possible using ClojureCLR.

Maybe LOCAL_ENV & Co should be public in ClojureCLR too?

(they could in theory be accesed using explicit reflection and BindingFlags.NonPublic, but that seems to be overkill).

core.clj: Implement await, await1, await-for

The Java version uses java.util.concurrent.CountDownLatch. Either use the new concurrency library for the CLR or implement the equivalent directly. The concurrency lib is at http://msdn.microsoft.com/en-us/concurrency/default.aspx. Using this would require distribution with this project. I haven’t thought about how hard it would be to implement the equivalent.

transactions eat line numbers

If an exception is thrown from inside a dosync transaction, the stacktrace stops at the transaction call. I confirmed this with an exception thrown from the dosync form and from a function called from the dosync form. In the stracktrace below, irc.clj line 74 contains just "(dosync".

  ;; expecting more frames here
  at clojure.lang.LockingTransaction.Run(IFn fn) in C:\Users\Shawn\dev\clojure-clr\Clojure\Clojure\Lib\LockingTransaction.cs:line 645
  at clojure.lang.LockingTransaction.runInTransaction(IFn fn) in C:\Users\Shawn\dev\clojure-clr\Clojure\Clojure\Lib\LockingTransaction.cs:line 524
  at irc$fn__109$user_nick__111.__invokeHelper_2(irc$fn__109$user_nick__111_base this, Object user, Object args) in irc.clj:line 74
  at irc$fn__109$user_nick__111.invoke(Object , Object )
  at clojure.lang.Var.invoke(Object arg1, Object arg2) in C:\Users\Shawn\dev\clojure-clr\Clojure\Clojure\Lib\Var.cs:line 628
  at irc$fn__231$handle__233.__invokeHelper_3(irc$fn__231$handle__233_base this, Object user, Object transitions, Object line) in irc.clj:line 171
  at irc$fn__231$handle__233.invoke(Object , Object , Object )
  at irc$fn__239$user_loop__241.__invokeHelper_2(irc$fn__239$user_loop__241_base this, Object user, Object socket) in irc.clj:line 180
  at irc$fn__239$user_loop__241.invoke(Object , Object )
  at irc$fn__248$spawn__250$fn__251.__invokeHelper_0(irc$fn__248$spawn__250$fn__251_base this) in irc.clj:line 194

Implement gen-interface

`Compiler/analyze` is missing in ClojureCLR?

... or there is an alternative method I should use?

(I tried converting clojure.contrib.repl-utils/expression-info (which uses Compiler/analyze) for use on the CLR... and failed :-) ).

Implement printf

The format function in core.clj depends on Java’s String.format, a printf-style formatter. There is no BCL equivalent, so we will have to implement it.

Automate the build process

There are several manual steps in the build process. Need to automate them. These include:
(1) moving *.clj files to appropriate places when changed. (2) Code a step to run Clojure.Compile on the bootstrap files to create new DLLs, which will then need to be distributed. (3) Set up a test environment for the CLJ tests.

`rotations` from `clojure.contrib.seq-utils` doesn't work on ClojureCLR

Apparently, this is because of a weird interaction between concat, drop and take.

Steps to reproduce:
user=> (concat (drop 1 [1 2]) (take 1 [1 2]))
(1 2 1)

Which is obviously incorrect.

A few small installing/building issues and a fix

Hi,

Overall, I managed to install and build ClojureCLR without much trouble. Cool.

1. My main suggestion is to give a default layout of the project in the install page, e.g. something like "download the latest dlr to DLR_Main under the folder where you installed ClojureCLR. Build the DLR solution. Now build the Clojure solution". The problem I ran into was that after linking the DLR projects to where I had installed them, I needed to remove and add some references in Visual Studio, and also it seems that some of the DLR output folders were in unexpected places if II didn't put the DLR_Main folder where the ClojureCLR solution expects it.

2. The bootstrap compiler crashed because it could not found junit.clj. The fix is simple: the properties of this file in Clojure.Source are not set to "Copy If Newer".

Hope this helps.

Kurt

Improve use of threads in Agent code

Right now, just starting our own threads and forgetting it. We need to manage a thread pool. Also need to implement 'shutdown'.

core.clj: Implement pmap, pcalls, pvalues

Most likely these could be done fairly easily at this point. I just haven’t gotten to them yet.

proxy-super broken

proxy-super is broken (calling the base method from the proxy code calls the proxy method for virtual methods).

Cause: replacement of gen.Emit(OpCodes.Call... with ILGen.EmitCall, which emits callvirt if the method is a virtual one.

Proposed fix:
diff --git a/Clojure/Clojure/CljCompiler/GenProxy.cs b/Clojure/Clojure/CljCompiler/GenProxy.cs
index fab6352..7f4b0d8 100644
--- a/Clojure/Clojure/CljCompiler/GenProxy.cs
+++ b/Clojure/Clojure/CljCompiler/GenProxy.cs
@@ -320,7 +320,7 @@ namespace clojure.lang
gen.EmitLoadArg(0); // gen.Emit(OpCodes.Ldarg_0);
for (int i = 0; i < parmCount; i++)
gen.EmitLoadArg(i + 1); // gen.Emit(OpCodes.Ldarg, i + 1);
- gen.EmitCall(m); // gen.Emit(OpCodes.Call, m);
+ gen.Emit(OpCodes.Call, m);
}
else
{

(line 323 in GenProxy.cs)

Maybe there are other related issues - using ILGen.EmitCall isn't the same as Emit(OpCodes.Call...).

core.clj: Implement file-seq

Needs translation into CLR System.IO calls.

Premature casting: We cast local variables to tagged types too early.

ClojureJVM has no problem with this:

(defn f [#^String x] x)
(f 12) => 12

At present, ClojureCLR will complain about not being able to case 12 to a String. This is due to local variables being typed in the DLR expressions. This affects other binding forms such as 'let'.

Duplicate :name key in gen-class

When called from ns:

(ns foo (:gen-class :name "Foo"))

Implement proxies

Part of the standard bootstrap environment, in clojure/core_proxy.clj. Heavily tied to Java reflection and the ASM 3.0 bytecode library. Significant rewrite effort.

Might want to wait until RH finishes the new-new project.

import of nested classes/RT.classForName

(import ... ) of nested classes, such as Compiler+CompilerException does not work. This is actually a problem with RT.classForName that will bite us in other places eventually.

(set! var true) fails

(set! var true) fails because of not boxing the bool value.

There is no special site for Clojure CLR and no documentation

If a project is to become popular it has to offer at least a:

• Getting started
• My first CLR app
• What is CLR anyway?
  etc

I would help but unfortunately am only a newbie in Clojure, but if you need me in this capacity I am available.

Ratios convert to int differently in CLR clojure

In CLR Clojure (1.2.0):
user=> (int (/ 4 5))
1

In Java Clojure (1.2.0):
user=> (int (/ 4 5))
0

Create an installer for the casual user

We need an installer that installs a directory with all the DLLs (include the DLR build) + CLJ sources.

Implement genclass

Part of the standard bootstrap environment, in clojure/genclass.clj. Heavily tied to Java reflection and the ASM 3.0 bytecode library. Significant rewrite effort.

Might want to wait until RH finishes the new-new project.

core_proxy.clj: Implement bean

Whoops. Left out one of the pieces of core_proxy. Simple (compared to the rest of the file) translation problem.

Fix numeric conversions

I have used three versions of numeric conversion to implement int, long, float, etc.
(a) Convert.ToXXX
(b) switch on TypeCode
(c) if/elseif on Type

Results are mixed. More analysis and benchmarking needed.

Compiler: Get rid of extran (fn* [] form) wrapper in GenerateAST.

This introduces an extra layer to analyze/compile, and messes up the class names generated for fns. The JVM version fakes this by putting a dummy FnExpr on the dynamic bindings before compiling a form.

core.clj: Implement future-call

Depends on clojure.lang.Agent/soloExecutor (need to implement), but more worryingly on java.util.concurrent.Future. The CTP concurrency runtime has this, or we implement.

Cultures and formatting

Hi,

While following the Clojure wikibook, I found:

user=> 0.1
1

This is because my culture is set differently. For parsing program text culture should not be taken into account, and I managed to achieve this by changing LispReader.cs as follows:

656: radix = Int32.Parse( m.Groups[6].Value, System.Globalization.CultureInfo.InvariantCulture );
679: return (object)Double.Parse(s,System.Globalization.CultureInfo.InvariantCulture);

That works, i.e.

user=> 0.1
0,1

which looks weird but seems correct - problem is now that the output is still formatted according to culture conventions. Again, a REPL should not take culture into account when printing I guess (especially since comma is reserved for tuple syntax so i just looks weird), but I haven't been able to find where to fix this.

Kurt

Change RT.T and RT.F to Object.

Used in ClojureJVM to provide standardized values for boxed true and false. Equivalent to Boolean.TRUE and Boolean.FALSE. I had misinterpreted this at the very beginning, so I had made RT.T/F to be bool values, meaning they were worse than useless. Changing them to type Object will provide standardized boxed values for true and false. Is the static field lookup faster than the boxing? One wonders.

Translate clojure/xml.clj

Part of the standard environment bootstrap. The JVM implementation relies extensively on SAX parsers as defined in org.xml.sax and javax.xml.parsers. This will require non-trivial rewriting to make it rely on BCL equivalents.

Extend gen-class and gen-interface to handle properties.

Create a .config file with full assembly names for standard .Net assemblies

This will simplify loading dependent assemblies.

We'll need different versions of this for .Net 4

core.clj: Implement seque

Depends on java.util.concurrent.BlockingQueue and .LinkedBlockingQueue.

core.clj: Implement make-array, to-array-2d

We need consideration of how to handle multi-dimensional arrays versus ragged arrays. JVM only has ragged arrays.

`proxy` is broken

(read (proxy [clojure.lang.PushbackTextReader] [(System.IO.StringReader. "100")]))

ends up in:
System.MissingMethodException: Method 'System.String.replace' not found.
at System.RuntimeType.InvokeMember(String name, BindingFlags bindingFlags, Binder binder, Object target, Object[] providedArgs, ParameterModifier[] modifiers, CultureInfo culture, String[] namedParams)
at System.Type.InvokeMember(String name, BindingFlags invokeAttr, Binder binder, Object target, Object[] args)
at clojure.lang.Reflector.CallInstanceMethod(String methodName, Object target, Object[] args) in C:\Temp\clojure-clr\Clojure\Clojure\Runtime\Reflector.cs:line 157
at clojure/core$fn__8346$proxy_name__8348.__invokeHelper_2(clojure/core$fn__8346$proxy_name__8348_base this, Object super, Object interfaces)
at clojure/core$fn__8370$get_proxy_class__8372.__invokeHelper_0v(clojure/core$fn__8370$get_proxy_class__8372_base this, Object bases)
at clojure.lang.RestFn.applyTo(ISeq args) in C:\Temp\clojure-clr\Clojure\Clojure\Lib\RestFn.cs:line 177
at clojure/core$fn__4990$apply__4992.__invokeHelper_2(clojure/core$fn__4990$apply__4992_base this, Object f, Object args)
at clojure/core$fn__8403$proxy__8405.__invokeHelper_2v(clojure/core$fn__8403$proxy__8405_base this, Object class_and_interfaces, Object args, Object fs)
at clojure.lang.RestFn.invoke(Object arg1, Object arg2) in C:\Temp\clojure-clr\Clojure\Clojure\Lib\RestFn.cs:line 469
at clojure.lang.Var.invoke(Object arg1, Object arg2) in C:\Temp\clojure-clr\Clojure\Clojure\Lib\Var.cs:line 607
at clojure.lang.AFn.ApplyToHelper(IFn ifn, ISeq arglist) in C:\Temp\clojure-clr\Clojure\Clojure\Lib\AFn.cs:line 223
at clojure.lang.Var.applyTo(ISeq arglist) in C:\Temp\clojure-clr\Clojure\Clojure\Lib\Var.cs:line 735
at clojure.lang.Compiler.MacroexpandSeq1(ISeq form) in C:\Temp\clojure-clr\Clojure\Clojure\CljCompiler\Compiler.cs:line 497
at clojure.lang.Compiler.AnalyzeSeq(ISeq form, String name, Boolean isRecurContext) in C:\Temp\clojure-clr\Clojure\Clojure\CljCompiler\Compiler.cs:line 432 (NO_SOURCE_FILE:230)

(on the JVM, (read (proxy [java.io.PushbackReader] [(java.io.StringReader. "100")])) works fine).

This actually happened when trying to convert clojure.contrib.repl-utils/get-source to use the CLR, so the proxy had redefined methods, unlike the simplified example here.

with-open Close->Dispose ?

Just a suggestion:

Maybe the .Close call used in with-open should be replaced with .Dispose.

Since we also have IDispose, maybe a type-hint could be used to eliminate reflection.

Since this breaks the Clojure meaning of with-open (use .close() and reflection), maybe another macro with a similar name should be added to clojure-clr? using? :-p

Remove streams

They lost to chunked seqs. Vestiges to be removed.

Compiled assemblies should be generated in working directory

Check the initialization/usage of the environment variable for this.

Notate need to load System.Windows.Forms before running celsius example

RT.load fails when Assembly.GetEntryAssembly() is null.

When using NUnit and Excel Assembly.GetEntryAssembly() is not present.

RT.GetFindFilePaths() should do a null check before calling Location and yielding.

(keys -some-System.Collections.Hashtable) fails

Problem with IDictionary entries versus IMapEntry. This will also affect (vals ...) and some others.

:line metadata missing (and :source-span not always available)

ClojureCLR should have :line available for defns and defmacros to be compatible with Clojure.

Also, :source-span is missing on defns/defmacros (apparently metadata is lost on macro expansion?).

TypeInitializationException on startup

I've run Clojure for CLR successfully before on a different machine.
Just downloaded this distribution - https://github.com/downloads/richhickey/clojure-clr/clojure-clr-1.3.0-alpha1-debug-4.0.zip

Running Clojure.Main.exe I get this:

Unhandled Exception: System.TypeInitializationException: The type initializer for 'Clojure.CljMain' threw an exception. ---> System.TypeInitializationException:
The type initializer for 'clojure.lang.RT' threw an exception. ---> System.NotSupportedException: An attempt was made to load an assembly from a network location which would have caused the assembly to be sandboxed in previous versions of the .NET Framework. This release of the .NET Framework does not enable CAS policy by default, so this load may be dangerous. If this load is not intended to sandbox the assembly, please enable the loadFromRemoteSources switch. See http://go.microsoft.com/fwlink/?LinkId=155569 for more information.
at System.Reflection.RuntimeAssembly.nLoadFile(String path, Evidence evidence
)
at System.Reflection.Assembly.LoadFile(String path)
at clojure.lang.Compiler.LoadAssembly(FileInfo assyInfo) in C:\work\clojure-c
lr\Clojure\Clojure\CljCompiler\Compiler.cs:line 1261
at clojure.lang.RT.load(String relativePath, Boolean failIfNotFound) in C:\wo
rk\clojure-clr\Clojure\Clojure\Lib\RT.cs:line 1980
at clojure.lang.RT.load(String relativePath) in C:\work\clojure-clr\Clojure\C
lojure\Lib\RT.cs:line 1960
at clojure.lang.RT.DoInit() in C:\work\clojure-clr\Clojure\Clojure\Lib\RT.cs:
line 543
at clojure.lang.RT..cctor() in C:\work\clojure-clr\Clojure\Clojure\Lib\RT.cs:
line 538
--- End of inner exception stack trace ---
at clojure.lang.RT.var(String ns, String name)
at Clojure.CljMain..cctor() in C:\work\clojure-clr\Clojure\Clojure.Main\Main.
cs:line 26
--- End of inner exception stack trace ---
at Clojure.CljMain.Main(String[] args)

Following directions on the Microsoft page, I added a Clojure.Main.exe.config file with the following, which makes it run OK:

Add DebugInfoExpression nodes to the compiler's DLR code generation.

Not catching some non-tail-recursive calls to #recur

We need to mimic more closely the JVM approach, I'm afraid.

CLR version says the following is okay (but it is not):

(defn f [x](loop [y 12]
%28if %28< y 20%29
%28do %28do %28recur %28inc y%29%29%29 y%29
y%29))

The inner do is flagging tail recursion on the recur call.

Translate clojure/inspector.clj

This is part of the standard bootstrap environment. This is a Swing app. This will need to be rewritten completely to make it work with WinForms or WPF.