hoc-users — General discussion about HOC

Re: [hoc-users] Pointer arguments to methods

[Wolfgang T. <wol...@gm...>]

> I notice that ifgen doesn't handle out-parameters, such as the
> increasingly-common (NSError**) idiom.  (Examples include NSDocument's
> readFromURL:ofType:error: and writeToURL:ofType:error:.)  Is that just
> a limitation of ifgen, or does the HOC runtime also need to be
> extended to handle that case?  Either way, I'm happy to help if I can.

I think there should be a certain division of labour here between  
ifgen and libHOC; the latter should provide some more variants of  
declareSelector which support smarter argument marshalling, while  
ifgen has to recognise the relevant situations (maybe aided by the  
binding-script) and generate calls to the appropriate things.

So we first need to figure out what kind of situations we want to  
handle:

- (BOOL)writeToFile:(NSString *)path options:(unsigned) 
writeOptionsMask error:(NSError **)errorPtr;

What should be the Haskell type of this?

NSString a -> CUInt -> (Bool, NSError ()) ?

Next, we should decide what TH macros libHOC should provide to import  
selectors with those types, and finally we need to decide in what  
situations ifgen can recognise these automatically, and when it will  
need help from the binding-script.

My brain is all spent for today... ideas are welcome.

Cheers,

Wolfgang

[hoc-users] Pointer arguments to methods

[Matthew M. <mat...@gm...>]

Greetings all,

I notice that ifgen doesn't handle out-parameters, such as the
increasingly-common (NSError**) idiom.  (Examples include NSDocument's
readFromURL:ofType:error: and writeToURL:ofType:error:.)  Is that just
a limitation of ifgen, or does the HOC runtime also need to be
extended to handle that case?  Either way, I'm happy to help if I can.

- Matt
--
http://www.matthewmorgan.net/blog

Re: [hoc-users] NSView/AppKit functions

[David C. <dw...@dw...>]

> It is an area that needed development until a few days ago :-).
> Functions with a fixed number of arguments are now imported  
> automatically in the bindings (as long as all the argument types  
> are supported by hoc). Supporting variable-argument functions like  
> NSLog would be quite hard and I think NSLog alone is not worth it -  
> it's easier to just write somehting to stderr yourself.
>
> Cheers,
>
> Wolfgang

Cool, thanks.

Really, the features that I liked from NSLog() was the ability to  
inline an NSObject's description within the format string (ala @"% 
@"), but that could be done just as well as haskellString (obj #  
description).  Now, depending on whether NSRect is a supported  
argument type (I think it's just a packed struct of floats), it'll  
make life easier.

Thanks!

David

Re: [hoc-users] NSView/AppKit functions

[Wolfgang T. <wol...@gm...>]

> I'm trying to write a custom NSView in Haskell.  For academic  
> purposes, I'm just trying to provide a basic drawRect selector; one  
> which just draws a frame around the bounds of the view.  However, I  
> am not seeing how to access the NSFrameRect() functions and friends.
>
> This seems to be a similar problem to the one I was describing in  
> my last message in regards to NSLog().  Is there some way to access  
> these functions from within Haskell with the use of HOC, or do I  
> manually have to wrangle in the functions through ffi?  Is this an  
> area which just needs development, and will be supported in future  
> versions?

It is an area that needed development until a few days ago :-).
Functions with a fixed number of arguments are now imported  
automatically in the bindings (as long as all the argument types are  
supported by hoc). Supporting variable-argument functions like NSLog  
would be quite hard and I think NSLog alone is not worth it - it's  
easier to just write somehting to stderr yourself.

Cheers,

Wolfgang

[hoc-users] Success building HOC under ghc 6.4.1 and gcc 4.0

[Matthew M. <mat...@gm...>]

I tried building HOC from CVS using ghc 6.4.1 and gcc 4.0 (on Tiger),
but got a link error citing an undefined symbol, _fprintf$LDBLStub.=20
The following change to HOC_cbits/Makefile.in fixes the problem:

33c33
<       ld -r -x -o HOC_cbits.o $(ALL_LOAD) libHOC_cbits.a
---
>       ld -r -x -o HOC_cbits.o -lSystemStubs $(ALL_LOAD) libHOC_cbits.a

This probably isn't the Right Fix, but it's at least a quick way to
get HOC running.

- Matt
--
http://www.matthewmorgan.net/blog

Re: [hoc-users] NSView/AppKit functions

[Wolfgang T. <wol...@gm...>]

> I'm trying to write a custom NSView in Haskell.  For academic  
> purposes, I'm just trying to provide a basic drawRect selector; one  
> which just draws a frame around the bounds of the view.  However, I  
> am not seeing how to access the NSFrameRect() functions and friends.

> Is this an area which just needs development, and will be supported  
> in future versions?

Yes. Definitely needs development.
Doing it by hand using the FFI is hard, as the FFI does not support  
passing structures like NSRect.
For now, I recommend writing a small objective-c helper class that  
wraps those functions in static methods; this class can then be used  
more conveniently using declareClass, declareSelector and an instance  
declaration.

About NSLog... variable argument functions are a nightmare and will  
probably never be properly supported.

CHeers,

Wolfgang

[hoc-users] NSView/AppKit functions

[David C. <dw...@dw...>]

Hello folks,

I'm trying to write a custom NSView in Haskell.  For academic purposes, 
I'm just trying to provide a basic drawRect selector; one which just 
draws a frame around the bounds of the view.  However, I am not seeing 
how to access the NSFrameRect() functions and friends.

This seems to be a similar problem to the one I was describing in my 
last message in regards to NSLog().  Is there some way to access these 
functions from within Haskell with the use of HOC, or do I manually 
have to wrangle in the functions through ffi?  Is this an area which 
just needs development, and will be supported in future versions?

Any pointers would be appreciated.

Thanks much,

David Christensen

[hoc-users] Foundation functions/va_lists?

[David C. <dw...@dw...>]

Greetings,

I was wondering about accessing the NSLog() function from hoc, which 
got me wondering about functions with va_list signatures from C.  Is it 
possible to have such functions in Haskell?  I've read the example of 
printf which requires a TH approach; I don't know the status of other 
such functions, but it seems like a similar approach would be 
necessary.

Thanks,

David Christensen

[hoc-users] Re: Haskell-on-Cocoa & ghc-6.4

[Wolfgang T. <wol...@gm...>]

On 9-Sep-05, at 6:58 PM, David Christensen wrote:

> Wolfgang,
>
> Thanks for the info -- got it all built and playing with it now.  A  
> few quick questions, as it seems like the mailing lists are fairly  
> sparse at this point, and google did not turn up anything useful:

So lets start making the mailing lists more interesting, shall we?  
I'm Cc-ing the hoc-users list now.

> 1) I'm trying to write a Show instance for NSString, NSArray and  
> some other core types in order to be able to use the string  
> representation of the Cocoa datatype in my Haskell program.   
> Currently, I'm using toNSString and fromNSString to try and convert  
> the types to something which can be used in Haskell's IO system,  
> but I keep running into errors ... but I'm still not enough of a  
> Haskell guru to be able to debug them.
>
> What I've got now is:
>
> instance Show (NSString ()) where
>     show = fromNSString

The error is due to a limitation in Haskell 98 (which, I think, was  
put there to prevent the type class system from becoming to complex).  
If you use -fglasgow-exts, the above should be legal.

A note on the () and why instance Show (NSString a) is a Bad Thing:  
The () means exacly NSString, no subclasses, and the a means,  
subclasses allowed. NSMutableString is a subclass of NSString, and  
using pure functions on mutable objects is evil; therefore,  
fromNSString has type NSString () -> String rather than NSString a ->  
String.

> 2) CamelBones, a Perl-Cocoa bridge, supports "toll-free bridging"  
> between Perl types and Obj-C types NSArray, NSDictionary, NSString,  
> etc.  Is this an option with hoc for future development?

It's not that easy; for Perl, they probably get to do it based on  
some runtime type of the object (NSString or Perl String). That won't  
work for Haskell, of course.
To do it the statically typed way, we might introduce some type  
class, say "StringClass", and make both NSString () and String  
instances of that class. This would mean a lot of ambiguity, though;  
whenever you get a NSString object from one method and pass it to  
another, you'd have to specify a type annotation (:: NSString ()  
or :: String) to specify whether it should be converted to a Haskell  
String and back, or not. I don't think it's worth it, I'd rather  
stick with explicit to/fromNSString.

However, if, one day in the future, there is a standard library type  
class for strings (to support things like packed strings more  
natively), we could take advantage of that by making NSString () an  
instance of that class.
Don't hold your breath though, this is being discussed about once a  
year on various Haskell mailing lists, with no results so far.


Cheers,

Wolfgang

[hoc-users] Wealth of Experience Is Needed.

[Arouna <aro...@ho...>]

Hello Dear
I am contacting you believing you are a honest and trustworthy person.I =
believe you will not betray the confidence I am willing to repose in you, for =
me to have contacted you even though we've not met before.
My name is Arouna  Goudal=E9,the only son of late Mr.Hassan Goudal=E9  who =
was the chairman /President/CEO,Cacao-Cafe & Agro-Alimentary industries =
plc.SCAFECAO)"Bouake" and ex-owner of Polyclinic LA-MADONNE, of the Republic =
of Cote d'Ivoire,please if possible you can send phone contact for good =
cominucation.
You can find the story on this =
website:http://news.bbc.co.uk/2/hi/africa/2269238.stm
On 19th of September 2002,a group of so called rebels took over the city of =
Bouake to over-throw the Government of Mr.Laurent Gbagbo the President of =
Cote d'Ivoire. On the 20th day morning at about 4:18am some fractions of the =
same group of rebels break into our house and killed our security guard and =
pass-through the gate entrance to the main building and took my father away =
to an un-known destination.
After about four hours later we found him (my father) at the road side, by =
then he had been totally beaten and was bleeding to death lying in the pool =
of his own blood. From there, we rushed him to a near by hospital and hours =
later he sent for my mother. So he told her that he can not make it from the =
doctors diagnosis, he also told her that it was because he (my father) was =
not from the northern part of the country that he has been treated this way =
and also being a political activist. And again he said he is suspecting some =
of his business partners for conspiracy and this sort of brutality. He also =
told my mother that he deposited a huge amount of money in a Finance house in =
the Capital city of Senegal (Dakar) and also told her where he kept the whole =
documents concerning this deposit that is made for investment purpose in the =
western world.
That same day, he (my father) gave up the ghost. Then with the help of the =
French soldiers, we escaped to Senegal, where we are now. I am now contacting =
you believing you are honest and trustworthy to help us to receive the money =
in your account for investment, I and my mother has decided in compensating =
you 20%of the total amount for your assistance .
Please do take your time to go through the mail and get back to us, so that I =
will simply introduce to the Finance house where the fund are being kept now =
for onward transfer.alternative mail is  ar...@wa... 
I am waiting for your response.
  Thanks and God bless.
     Arouna Goudal=E9 .  

Re: [hoc-users] performSelectorWithObject problems

[Wolfgang T. <wol...@gm...>]

> There's just one problem left: When I demand the state of a button,  
> it returns an IO Int, the according constants are of type  
> NSCellStateValue. Is there a way to "cast" those values?

The InterfaceGenerator generates the following:

data NSCellStateValue = NSMixedState
                       | NSOffState
                       | NSOnState
instance ObjCArgument NSCellStateValue CInt where
     exportArgument NSMixedState = return (-1)
     exportArgument NSOffState = return 0
     exportArgument NSOnState = return 1
     importArgument (-1) = return NSMixedState
     importArgument 0 = return NSOffState
     importArgument 1 = return NSOnState
     objCTypeString _ = "i"

You can use those, but it is obviously necessary to design something  
"proper".
Haskell's Enum class would not be appropriate because the succ, pred,  
enumFromTo, etc. functions aren't appropriate for everything that is  
defined using C enum.
And while we're at it, there's the question of what to do with  
anonymous enums (I'm just ignoring them so far).

Cheers,

Wolfgang

Re: [hoc-users] performSelectorWithObject problems

[Christian H. <chr...@gm...>]

Thank you very much for your help! Everything works great now!

There's just one problem left: When I demand the state of a button,  
it returns an IO Int, the according constants are of type  
NSCellStateValue. Is there a way to "cast" those values?

Cheers,
Chris

Re: [hoc-users] performSelectorWithObject problems

[Wolfgang T. <wol...@gm...>]

> To make it more interesting,  it uses the  
> "performSelectorWithObject" method on itself. When I do this, my  
> application crashes.

I found the problem. performSelectorWithObject calls a method that  
takes an object parameter _and returns an object_.

You need to declare:
$(declareSelector "dummy:" [t| forall a. NSObject a -> IO (NSObject  
())|])
and return nil from the method.

When performSelectorWithObject calls a method without a return value,  
it passes on whatever (garbage) value it finds in the machine  
register R3 that would have been used for the return value. When HOC  
tries to convert that garbage value back to a Haskell object, it  
crashes. I'll try to add some additional sanity checks there so that  
we can at least have better error messages.

Cheers,

Wolfgang

Re: [hoc-users] performSelectorWithObject problems

[Christian H. <chr...@gm...>]

Am 21.07.2005 um 04:23 schrieb Wolfgang Thaller:

>> now that the installation finally succeeded, the real problems  
>> begin...
>>
>> I wrote a small application with a "Quit" button. Before the  
>> application displays that it quits, I want it to perform some  
>> fancy action: call the dummy method "dummy" on itself. (Ok, this  
>> is just a simplified version of what I really want...)
>>
>> To make it more interesting,  it uses the  
>> "performSelectorWithObject" method on itself. When I do this, my  
>> application crashes.
>>
>
> Please stand by while I try to reproduce the problem... does  
> running the program from gdb turn up anything at all?
>

I could locate the problem a bit better: The problem is, that I  
import a module where (what I forgot) a global MVar is declared, like

module Test where

import Control.Concurrent.MVar
import System.IO.Unsafe

testVar :: MVar Int
testVar = unsafePerformIO (newMVar 0)

The module has so far not been used at all in my application, but the  
import alone seems to interfere somehow with the Cocoa framework.  
This problem remains when compiling with the -threaded option.

Cheers,
Chris

[hoc-users] Re: HOC and threads (was: performSelectorWithObject problems)

[Wolfgang T. <wol...@gm...>]

Christian Hofer wrote:

> By the way, is there a way to get MVar concurrency working together  
> with HOC? Can I use MVars just like that (after telling Cocoa to  
> run thread-safe)? Do I have to use POSIX threads?

Compile your program using the -threaded option to ghc in order to  
enable the threaded run time system.
Then,

"forkIO" spawns an "unbound" (lightweight) thread. The RTS may run  
multiple lightweight threads in a single POSIX thread. Do not call  
Cocoa from threads created by forkIO.

"forkOS" spawns a POSIX thread.

Best of all, you can also use NSThread from HOC.

And YES, MVars will work between all those different threads.

Note that if you don't use -threaded, calls to forkIO will cause a  
run time error and calls to any Haskell code from other OS threads  
(NSThreads, pthreads, whatever) will most likely result in a crash.

Cheers,

Wolfgang

Re: [hoc-users] performSelectorWithObject problems

[Wolfgang T. <wol...@gm...>]

> now that the installation finally succeeded, the real problems  
> begin...
>
> I wrote a small application with a "Quit" button. Before the  
> application displays that it quits, I want it to perform some fancy  
> action: call the dummy method "dummy" on itself. (Ok, this is just  
> a simplified version of what I really want...)
>
> To make it more interesting,  it uses the  
> "performSelectorWithObject" method on itself. When I do this, my  
> application crashes.

Please stand by while I try to reproduce the problem... does running  
the program from gdb turn up anything at all?

Cheers,

Wolfgang

[hoc-users] performSelectorWithObject problems

[Christian H. <chr...@gm...>]

Hi again,

now that the installation finally succeeded, the real problems begin...

I wrote a small application with a "Quit" button. Before the  
application displays that it quits, I want it to perform some fancy  
action: call the dummy method "dummy" on itself. (Ok, this is just a  
simplified version of what I really want...)

To make it more interesting,  it uses the "performSelectorWithObject"  
method on itself. When I do this, my application crashes.

In Objective-C it looks like this (and it works):

- (IBAction)quit:(id)sender
{
     [self performSelector: @selector(dummy:) withObject: self];
     [actionLabel setStringValue: [NSString stringWithCString:  
"Quit!" encoding: NSMacOSRomanStringEncoding]];
}
- (void)dummy:(NSObject*)ignored
{
}

In Haskell, I started like this:
I declared a selector for my dummy in my Selectors.hs:
$(declareSelector "dummy:" [t| forall a. NSObject a -> IO ()|])

Then I added an instance method:
$(exportClass [...] InstanceMethod Selectors.info_dummy])

inet_dummy _ self = return ()

inet_quit _ self =
     self # performSelectorWithObject (getSelectorForName "dummy:") self
     -- self # inet_dummy self   <-- that works, but is not what I want!
     self #. _actionLabel >>= setStringValue (toNSString ("Quit!"))  
-- replacing this line with just a "return ()" works as well!!!

Now, the selector is found ((self # respondsToSelector  
(getSelectorForName "dummy:")) returns true), but when I try to  
perform the selector, and change the label text afterwards, the  
application crashes. Am I doing something wrong?

(Remark: I did not use the -O compiler option.)

By the way, is there a way to get MVar concurrency working together  
with HOC? Can I use MVars just like that (after telling Cocoa to run  
thread-safe)? Do I have to use POSIX threads?

Regards,
Chris

Re: [hoc-users] Build problem with CVS version on Tiger with gcc3.3 and ghc6.4

[Wolfgang T. <wol...@gm...>]

> Indeed, there is no HOC.conf, only the files:
>
> $ find . -name 'HOC.conf*'
> ./HOC/HOC.conf-inplace
> ./HOC/HOC.conf.in

Oops, a missing dependency in the Makefile. Fixed now in CVS, a quick  
workaround:

cd HOC
make HOC.conf


Thanks for the report.

Cheers,

Wolfgang

Re: [hoc-users] Build problem with CVS version on Tiger with gcc3.3 and ghc6.4

[Christian H. <chr...@gm...>]

Am 18.07.2005 um 18:27 schrieb Wolfgang Thaller:
> Grmpf. Code built by 3.3 "should not" refer to fprintf$LDBLStub.
> a) Make sure you've completely rebuilt the HOC_cbits directory  
> after switching to gcc 3.3. ("make clean" should work, but don't  
> trust my Makefiles to work correctly; make sure that all the .o  
> files are really gone).

Ok. Indeed, "make clean" did not remove the libffi object files. I  
deleted the whole thing, refetched it from CVS, and now the  
compilation works fine. Thank you for your help!

But now, "make install" fails with the following error:

---

ghc-pkg update HOC.conf
Reading package info from "HOC.conf" ghc-pkg.bin: HOC.conf: openFile:  
does not exist (No such file or directory)

make[1]: *** [install] Error 1

---

Indeed, there is no HOC.conf, only the files:

$ find . -name 'HOC.conf*'
./HOC/HOC.conf-inplace
./HOC/HOC.conf.in

Regards,
Chris

Re: [hoc-users] Build problem with CVS version on Tiger with gcc3.3 and ghc6.4

[Wolfgang T. <wol...@gm...>]

> Unfortunately, yes. (In order to be sure, I tried it again, but the  
> problem remains.)
>
> Chris
>
> $ gcc --version
> gcc (GCC) 3.3 20030304 (Apple Computer, Inc. build 1809)
> Copyright (C) 2002 Free Software Foundation, Inc.

Grmpf. Code built by 3.3 "should not" refer to fprintf$LDBLStub.
a) Make sure you've completely rebuilt the HOC_cbits directory after  
switching to gcc 3.3. ("make clean" should work, but don't trust my  
Makefiles to work correctly; make sure that all the .o files are  
really gone).

b) Write a "Hello, world" program in C (#include <stdio.h> and use  
printf, other popular "Hello, world" variants don't do the trick  
here...). Look at or send me the output of gcc -S and gcc -E.
If the output mentions $LDBLStub, find out the values of the  
__LDBL_MANT_DIG__ and __DBL_MANT_DIG__ macros. (See sys/cdefs.h for  
info on how Apple's headers decide whether to use "foo$LDBLStub"  
instead of plain "foo" for some library functions).


Cheers,

Wolfgang

Re: [hoc-users] Build problem with CVS version on Tiger with gcc3.3 and ghc6.4

[Christian H. <chr...@gm...>]

Am 18.07.2005 um 07:54 schrieb Wolfgang Thaller:

>> ../HOC_cbits/HOC_cbits.o: unknown symbol `_fprintf$LDBLStub'
>>
>
> Are you sure that you are using gcc 3.3 and not gcc 4.0? That's a  
> known incompatibility between ghc 6.4 and gcc 4.0 (use "sudo  
> gcc_select 3.3")


Unfortunately, yes. (In order to be sure, I tried it again, but the  
problem remains.)

Chris

$ gcc --version
gcc (GCC) 3.3 20030304 (Apple Computer, Inc. build 1809)
Copyright (C) 2002 Free Software Foundation, Inc.

Re: [hoc-users] Build problem with CVS version on Tiger with gcc3.3 and ghc6.4

[Wolfgang T. <wol...@gm...>]

> ../HOC_cbits/HOC_cbits.o: unknown symbol `_fprintf$LDBLStub'

Are you sure that you are using gcc 3.3 and not gcc 4.0? That's a  
known incompatibility between ghc 6.4 and gcc 4.0 (use "sudo  
gcc_select 3.3")

Cheers,

Wolfgang

[hoc-users] Build problem with CVS version on Tiger with gcc3.3 and ghc6.4

[Christian H. <chr...@gm...>]

Hi,

I am unable to build HOC under the specified conditions.

make outputs the following:

ghc-6.4:
../HOC_cbits/HOC_cbits.o: unknown symbol `_fprintf$LDBLStub'
final link ... ghc-6.4: linking extra libraries/objects failed
make[1]: *** [ghcmake.build-stamp] Error 1
In directory InterfaceGenerator
mkdir -p build
/usr/local/bin/ghc  --make Main.hs      \
         -odir build -hidir build \
         -package-conf ../inplace.conf \
         -o ifgen
../inplace.conf: openBinaryFile: does not exist (No such file or  
directory)

... (and lots of further error messages)

Unfortunately, I am quite illiterate w.r.t. the make-procedure. So I  
have no idea what that means.

Can anybody give me a hint?

Cheers,
Chris

[hoc-users] Compiling HOC on Tiger

[Wolfgang T. <wol...@gm...>]

I just found out about a problem with GHC 6.4 and gcc 4.0 on Tiger  
that affects the HOC CVS version (thanks for Petter Egesund for  
making me aware of the problem).

You have to switch to gcc 3.3 for compiling (the libffi component of)  
HOC; if you compile it using gcc 4.0, there will be strange errors  
later during the build.

Enjoy,

Wolfgang

Re: [hoc-users] Problems with build for ghc 6.4

[Wolfgang T. <wol...@gm...>]

On 12-Jun-05, at 9:03 AM, Petter Egesund wrote:

> Thanks for answering!
>
> It does not seem like inplace.conf exists?!
>
> A dump from make is enclosed!

Ah, I see, make in directory HOC failed for another reason:

Compiling HOC.StdArgumentTypes ( ./HOC/StdArgumentTypes.hs, build/ 
objects/HOC/StdArgumentTypes.o )
ghc-6.4:
../HOC_cbits/HOC_cbits.o: unknown symbol `_fprintf$LDBLStub'

There seem to be lots of problems involving _fprintf$LDBLStub on the  
net, but I didn't quite manage to figure things out yet, and I'm  
unable to reproduce the problem.

Which gcc and OS version are you using?
If you are using gcc 4 and tiger: have you installed XCode 2.1 (the  
new "universal binaries" release)?

Cheers,

Wolfgang