Nabble - Haskell - Nhc-bugs

NHC Compilation Failure

2008-01-09T10:26:41Z

Hi, I downloaded the source of nhc98-1.20 and tried to compile it, but when it is linking greencard-nhc98, I got multiple errors like:

.../nhc98-1.20/lib/x86_64-Linux/Prelude.a(RealFloat_Float.o)(.text+0x49): In function `FR_Prelude_46cFIsNeg':
: undefined reference to `signbit'
.../nhc98-1.20/lib/x86_64-Linux/Prelude.a(RealFloat_Float.o)(.text+0xc9): In function `FR_Prelude_46cFIsNormal':
: undefined reference to `isnormal'
.../nhc98-1.20/lib/x86_64-Linux/Prelude.a(RealFloat_Float.o)(.text+0x149): In function `FR_Prelude_46cFIsFinite':
: undefined reference to `isfinite'

Then, when it is linking lib/x86_64-Linux/MkProg, there is another error:

.../nhc98-1.20/lib/x86_64-Linux/Prelude.a(RemoveFile.o)(.data+0xc): undefined reference to `FR_Directory_46unlink_35'

and soon after this the whole make fails because there is no lib/x86_64-Linux/MkConfig.

uname -a shows:

Linux ... 2.6.9-55.ELsmp #1 SMP Fri Apr 20 16:36:54 EDT 2007 x86_64 x86_64 x86_64 GNU/Linux

and targets/x86_64-Linux/config.cache shows:

INSTALLDIR=...
DOCDIR=...
BUILDWITH=gcc
BUILDOPTS=""
BUILDDIR=.../nhc98-1.20/targets
RTSFLAG=yes
CC=gcc
COPTS=""
ENDIAN=-DLOW_BYTE_FIRST
HEAP=100000
LIB=yes
BIN=yes
INC=yes
MAN=yes
DOCS=no
EXE=
CURSES="-ltermcap -lcurses"
READLINE=""
INSTALLVER="v1.20 (2007-11-22)"
INSTALLINFO="config: x86_64-Linux/ by ... on Wed Jan 9 12:00:05 CST 2008"
FW=""

Can anybody help me on how to get this build to work? Thanks!

Di, Yu
1.9

Looking for last minute shopping deals? Find them fast with Yahoo! Search.
_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Re: cpphs and hugs

2006-11-30T05:33:12Z

On Wed, Nov 29, 2006 at 04:49:36PM +0000, Malcolm Wallace wrote:

> Ian Lynagh <igloo@...> wrote:
>
> > $ /tmp/cpphs/inst/bin/cpphs /dev/null
> >
> > Program error: pattern match failure: macroProcess ((instEq_v4051
> > `elem` CpphsOption_CpphsLayout) []) Bool_True (preDefine
> > ((instEq_v4051 `elem` CpphsOption_CpphsAnsi) []) Bool_True
> > (runCpphs_v4067 [])) (tokenise_v3906 ((instEq_v4051 `elem`
> > CpphsOption_CpphsAnsi) []) ((instEq_v4051 `elem`
> > CpphsOption_CpphsStrip) []) SubMode_Any) $
>
> I tried this with hugs-Sept2006 and could not reproduce the error.
> Mystified.

Oops, looked like I did that in the wrong window, using an ancient,
buggy hugs. Sorry about that.

> > I tried upgrading to 1.2 to see if it is fixed, and trying to haddock
> > missingh then gave an almighty pattern match failure.
>
> Is there a more accessible mirror of MissingH anywhere?

Sorry, I wasn't expecting you to need it as /dev/null looked like a good
first test case.

Here's a smaller testcase for the original problem, with hugs September
2006 and cpphs 1.2:

$ runhugs -98 Setup.hs --hugs configure --prefix=/tmp/cpphs/inst
$ runhugs -98 Setup.hs build
$ runhugs -98 Setup.hs install
$ export LC_ALL=POSIX
$ printf "Bj\xf6rn Bringert" > foo
$ /tmp/cpphs/inst/bin/cpphs foo

Program error: <handle>: IO.getContents: protocol error (input contains
non-character data - use binary I/O for binary data)
$

Thanks
Ian

_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Re: cpphs and hugs

2006-11-29T08:49:36Z

Ian Lynagh <igloo@...> wrote:

> With cpphs 0.7 "compiled" with hugs I get the following when trying to
> haddock missingh:
>
> Program error: <handle>: IO.getContents: protocol error (invalid
> character encoding) missingh-0.16.2$
> --------------
>
> I believe this is because there is a mismatch between my locale and
> the encoding of the file, in particular with the \"o in Bjorn
> Bringert's name. I think opening the file in binary mode would fix
> this, but I couldn't swear to it.

Hmm. cpphs doesn't do anything special with character encodings, so I
would guess this is a Hugs issue maybe?

I'm afraid our work firewall blocks all outgoing traffic to port 70, so
I can't download MissingH to try to reproduce the error.

Also, as you noticed, cpphs-0.7 is kind of old now.

> I tried upgrading to 1.2 to see if it is fixed, and trying to haddock
> missingh then gave an almighty pattern match failure.

Is there a more accessible mirror of MissingH anywhere?

> A somewhat
> smaller failure occurs when trying to process /dev/null:
>
> --------------
> $ /tmp/cpphs/inst/bin/cpphs /dev/null
>
> Program error: pattern match failure: macroProcess ((instEq_v4051
> `elem` CpphsOption_CpphsLayout) []) Bool_True (preDefine
> ((instEq_v4051 `elem` CpphsOption_CpphsAnsi) []) Bool_True
> (runCpphs_v4067 [])) (tokenise_v3906 ((instEq_v4051 `elem`
> CpphsOption_CpphsAnsi) []) ((instEq_v4051 `elem`
> CpphsOption_CpphsStrip) []) SubMode_Any) $

I tried this with hugs-Sept2006 and could not reproduce the error.
Mystified.

Regards,
Malcolm
_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

cpphs and hugs

2006-11-28T08:50:22Z

Hi,

With cpphs 0.7 "compiled" with hugs I get the following when trying to
haddock missingh:

--------------
missingh-0.16.2$ /usr/bin/cpphs -Obar.hs MissingH/Time/ParseDate.hs --noline --strip -Dlinux_BUILD_OS -Dlinux_HOST_OS -Di386_BUILD_ARCH -Di386_HOST_ARCH -D__GLASGOW_HASKELL__=606 -D__HADDOCK__

Program error: <handle>: IO.getContents: protocol error (invalid character encoding)
missingh-0.16.2$
--------------

I believe this is because there is a mismatch between my locale and the
encoding of the file, in particular with the \"o in Bjorn Bringert's
name. I think opening the file in binary mode would fix this, but I
couldn't swear to it.

I tried upgrading to 1.2 to see if it is fixed, and trying to haddock
missingh then gave an almighty pattern match failure. A somewhat smaller
failure occurs when trying to process /dev/null:

--------------
$ /tmp/cpphs/inst/bin/cpphs /dev/null

Program error: pattern match failure: macroProcess ((instEq_v4051 `elem` CpphsOption_CpphsLayout) []) Bool_True (preDefine ((instEq_v4051 `elem` CpphsOption_CpphsAnsi) []) Bool_True (runCpphs_v4067 [])) (tokenise_v3906 ((instEq_v4051 `elem` CpphsOption_CpphsAnsi) []) ((instEq_v4051 `elem` CpphsOption_CpphsStrip) []) SubMode_Any)
$
--------------

cpphs compiled with GHC seems fine.

Thanks
Ian

_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Re: nhc98 bug

2006-08-21T08:59:29Z

GaMeR uWe <gamer_uwe@...> wrote:

> GHCSYM=
>
> 604
> TRUE=/bin/true

This problem has been known and fixed in CVS/darcs for a while now. At
last, I have created and published a patch to the stable release:

ftp://ftp.cs.york.ac.uk/pub/haskell/nhc98/patch-1.18-confhc

also noted on the official download page.

Regards,
Malcolm
_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

nhc98 bug

2006-07-25T13:06:31Z

I found something in the nhc98's configure script that leads me to believe that there is a bug somewhere, I don't see the point in searching for it on my own as it is nothing big, just slightly annoying.

When ./configure is run, the file nhc98-1.18/lib/ix86-linux/config is created, however make cannot be run because of the following message it will output, "*** missing separator.". I then checked what the file generated by ./configure contained and I found the problem. It contained:

"...
GHCSYM=

604
TRUE=/bin/true"

Where ... represents other variables set before GHCSYM (none of those above had a problem). As you can see, there is a gap between GHCSYM= and the value of 604 which seems to belong to GHCSYM. So I changed it to GHCSYM=604 and then everything compiled and installed. nhc98 does execute from the terminal, I havn't tried compiling anything yet but I am sure it will compile things without a problem, I just thought I'd let you know about this problem with that file.

Thanks.

_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Re: Trouble getting nhc98 to work

2006-07-21T08:01:09Z

Karl Crary <crary@...> wrote:

> I was experimenting with nhc98 because I was led to believe that using
> it, I might be able to use the Haskell FFI to link with libraries
> built using Cygwin.

Should be possible, yes.

> First I tried the binary distribution. The configure and install
> process seemed to go smoothly, but in my first attempt to use the
> compiler I get:
>
> /usr/local/bin/nhc98: line 131:
> C:/cygwin/usr/local/lib/nhc98/ix86-CYGWIN_NT-5.1/hmake-PRAGMA:
> No such file or directory

The executable hmake-PRAGMA.exe is certainly included in the tar bundle.
As far as I can see, it should therefore have been installed in the
quoted directory, given the standard installation prefix. Is it there?
If so, then maybe the driver script for nhc98 just needs to know about
the .exe suffix (but that would seem unlikely in Cygwin).

> I then tried to build from source, but with no better luck. Configure
> seems to go smoothly, but the build breaks down quickly:
>
> make[1]: Entering directory `/c/nhc98-1.18/src/runtime'
> Makefile:27: *** target pattern contains no `%'. Stop.

Did you use a Windows-style pathname at configure-time?
(e.g. --prefix="C:\blah") Replacing that with a Cygwin unix-style
pathname might fix it (e.g. --prefix="/c/blah").

Regards,
Malcolm
_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Trouble getting nhc98 to work

2006-07-17T16:26:57Z

Dear NHC98 maintainers,

I was experimenting with nhc98 because I was led to believe that using
it, I might be able to use the Haskell FFI to link with libraries built
using Cygwin. Unfortunately, I never got far enough to try that; it's
failing right out of the box.

First I tried the binary distribution. The configure and install
process seemed to go smoothly, but in my first attempt to use the
compiler I get:

/usr/local/bin/nhc98: line 131:
C:/cygwin/usr/local/lib/nhc98/ix86-CYGWIN_NT-5.1/hmake-PRAGMA: No such
file or directory
/usr/local/bin/nhc98: line 131:
C:/cygwin/usr/local/lib/nhc98/ix86-CYGWIN_NT-5.1/hmake-PRAGMA: No such
file or directory
/usr/local/bin/nhc98: line 131:
C:/cygwin/usr/local/lib/nhc98/ix86-CYGWIN_NT-5.1/hmake-PRAGMA: No such
file or directory
/usr/local/bin/nhc98: line 547:
C:/cygwin/usr/local/lib/nhc98/ix86-CYGWIN_NT-5.1/nhc98comp: No such file
or directory

I then tried to build from source, but with no better luck. Configure
seems to go smoothly, but the build breaks down quickly:

bash-3.1$ make
ghc -cpp -o lib/ix86-CYGWIN_NT-5.1/hmake-PRAGMA script/hmake-PRAGMA.hs
cd src/runtime; make all nhc98heap.exe
make[1]: Entering directory `/c/nhc98-1.18/src/runtime'
Makefile:27: *** target pattern contains no `%'. Stop.
make[1]: Leaving directory `/c/nhc98-1.18/src/runtime'
make: *** [targets/ix86-CYGWIN_NT-5.1/runtime] Error 2
bash-3.1$ gmake
bash: gmake: command not found
bash-3.1$ which make
/usr/bin/make
bash-3.1$ make
cd src/runtime; make all nhc98heap.exe
make[1]: Entering directory `/c/nhc98-1.18/src/runtime'
Makefile:27: *** target pattern contains no `%'. Stop.
make[1]: Leaving directory `/c/nhc98-1.18/src/runtime'
make: *** [targets/ix86-CYGWIN_NT-5.1/runtime] Error 2

Any idea what I might be doing wrong?

-- Karl Crary

_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Re: Bus Error and Segmentation Fault when Heap Profiling is On

2006-06-02T07:06:57Z

Shin-Cheng Mu <scm@...> wrote:

> scm$ cat > temp.xml
> <a><b><item>123</item></b><c></c></a>
> scm$ ./hxpc temp.xml
> Bus error
> scm$ ./hxpc temp.xml +RTS -p
> Segmentation fault
>
> If I compile the program without the -p option, it runs fine.
>
> Since Malcolm Wallace has already profiled the program
> on his machine, it seems to be a platform specific problem

When I profiled your code previously, it worked fine on either a
Linux/x86 or Solaris/sparc machine (I don't remember which), but not on
MacOS/powerpc, so yes it does look like a platform-specific problem.
Strangely, although the profile was generated OK on the other platform,
the hp2graph tool for converting it to a graph seg faulted there, but
worked correctly on MacOS/powerpc.

Regards,
Malcolm
_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Re: ld: Undefined symbols on OS X when compiling with -t

2006-05-22T07:53:03Z

Shin-Cheng Mu <scm@...> wrote:

> scm$ hmake -nhc98 -t text
> nhc98 -t -c -o text.z.o text.hs
> nhc98 -t -o text text.z.o
> /usr/bin/ld: Undefined symbols:
> _FN_NHC_46Internal_46_95apply1
> _FN_NHC_46Internal_46_95apply2
> _TMSUB_NHC_46DErrNo
> _TM_NHC_46DErrNo
> _TMSUB_NHC_46PackedString
> _TM_NHC_46PackedString
>
> Is it a bug, or is something missing in Darwin Ports?

This is a bug. Unfortunately, the time-profiling code has not been
properly maintained or tested since it was contributed to the main
compiler branch. (The reason for this particular breakage is that
time-profiling relies on a post-processor for object files after they
have been built, and this has some some hand-coded hacks for internal
Prelude-like definitions. When the runtime internals changed, these
hacks were not kept in sync.)

Regards,
Malcolm
_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Bus Error and Segmentation Fault when Heap Profiling is On

2006-05-22T07:39:42Z

Hi,

This is another potential bug. Unfortunately I currently
cannot reproduce it on smaller programs. I tried to do
heap profiling on the following code:

http://www.psdlab.org/~scm/hx.tar.gz

The main program is hxpc.hs. I tried to compile it with

hmake -nhc98 -p hxpc

and run it on a small sample input

scm$ cat > temp.xml
<a><b><item>123</item></b><c></c></a>
scm$ ./hxpc temp.xml
Bus error
scm$ ./hxpc temp.xml +RTS -p
Segmentation fault

Increasing the heap size does not seem to help.

If I compile the program without the -p option, it runs fine.

Since Malcolm Wallace has already profiled the program
on his machine, it seems to be a platform specific problem
I installed NHC98 1.18_4 using Darwin Ports, on a PowerPC
G4 based Mac, running OS X 10.4 and GCC 4.0.

sincerely,
Shin-Cheng Mu
_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

ld: Undefined symbols on OS X when compiling with -t

2006-05-22T07:28:31Z

Hi,

I just installed NHC98 using Darwin Ports on a PowerPC
G4 based Mac, running OS X 10.4 with GCC 4.0. NHC works
fine without profiling options. When I turned on time
profiling, however, NHC98 appears to have problems linking:

scm$ cat text.hs
main = print (foldr (+) 0 [1..10000])

scm$ hmake -nhc98 -t text
nhc98 -t -c -o text.z.o text.hs
nhc98 -t -o text text.z.o
/tmp/tprofusr3456.c: In function 'tprofTMInitTreeUsr':
/tmp/tprofusr3456.c:21: warning: incompatible implicit declaration
of built-in function 'strdup'
/usr/bin/ld: Undefined symbols:
_FN_NHC_46Internal_46_95apply1
_FN_NHC_46Internal_46_95apply2
_TMSUB_NHC_46DErrNo
_TM_NHC_46DErrNo
_TMSUB_NHC_46PackedString
_TM_NHC_46PackedString
collect2: ld returned 1 exit status

Is it a bug, or is something missing in Darwin Ports?

sincerely,
Shin-Cheng Mu

_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Re: What's the correct procedure for upgrading?

2006-04-18T06:10:34Z

Charles,

> Do I just install 1.18 and forget about 1.16? Or should I delete all
> of the 1.16 files first?

You can do either. There should be no conflicts either way. All of the
1.16 files should be overwritten by the 1.18 versions in any case.

> I didn't see a Makefile target "uninstall",
> although there are some "*clean*" options there.

The 'clean' options just tidy up the build tree, but alternatively, it
is safe simply to delete the whole build tree after installing.

You are right that there is no 'uninstall' option, and perhaps there
should be. If you want to delete an older version of nhc98, here are
the places that installed files live:

prefix=/usr/local (usually)
$prefix/bin/
$prefix/lib/nhc98
$prefix/lib/hmake
$prefix/include/nhc98
$prefix/man/man1
$prefix/doc/nhc98

> Perhaps you could add this topic to your website?

A good idea.

Regards,
Malcolm
_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

What's the correct procedure for upgrading?

2006-04-14T09:58:34Z

I hope nhc-bugs is the correct venue for this question.

I installed NHC 1.16 on Cygwin (Windows XP), had some problems with it,
and then realized I had grabbed an old package. I'm going to upgrade to
1.18 and try again.

However, I'm a little bit concerned about the right way to do it. I
haven't found any instructions for upgrading properly.

Do I just install 1.18 and forget about 1.16? Or should I delete all of
the 1.16 files first? I didn't see a Makefile target "uninstall",
although there are some "*clean*" options there.

Perhaps you could add this topic to your website?

Thanks.

Charles Hethcoat
_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Re: Linux install troubles

2006-04-12T05:18:35Z

[ original message delayed through spam filtering ]

Doug McIlroy <doug@...> wrote:

> On Fedora Linux, I fetched nhc98-1.18.tar.gz (Linux executable).
>
> Got some relatively innocuous diagnostics from configure and make
> install, but one stood out:
>
> ./configure: line 594: src/runtime/nhc98heap.c: No such file or
> directory

In the binary tarball, the directory src/runtime does not exist, so
writing a file into it fails. However, line 594 of the configure script
is protected by a test on the existence of directory src/, so I'm a
little unsure why it even got there.

> Bulling through anyway, I did an installation. and compiled a hello
> world program. The result appears to be related to the configure
> trouble:

It is not related, no.

> OS allocated a heap in high memory (>0x80000000)
> which breaks this program's run-time system.
> hpStart=0xb7d96008, hpEnd=0xb7e0b308

This is a known bug, triggered by more recent versions of Linux. I
attach a source patch which is claimed to work around the problem
(tested on a Fedora system). It isn't a full fix, but should avoid
high-memory allocation in many circumstances.

Regards,
Malcolm

Index: src/runtime/Kernel/collector.c
===================================================================
RCS file: /home/cvs/root/nhc98/src/runtime/Kernel/collector.c,v
retrieving revision 1.24
retrieving revision 1.26
diff -u -r1.24 -r1.26
--- src/runtime/Kernel/collector.c 7 Mar 2005 16:53:38 -0000 1.24
+++ src/runtime/Kernel/collector.c 28 Mar 2006 09:21:11 -0000 1.26
@@ -9,6 +9,7 @@

/*#define HEAPSIZE 100000 -- defined in top-level Makefile at config-time */
#define GCSTACKSIZE 20000
+#define SBRK 1 /* Use sbrk(2) instead of malloc(3) to allocate the heap */

WHEN_DYNAMIC(int ractive = 0;)

@@ -46,7 +47,13 @@
Int totalSize = hpSize+spSize;
Int tableSize = (totalSize+WORDSIZE)/(WORDSIZE+1)+1; /* Last one for end of marked */

- if(NULL == (hpStart = malloc ((int)totalSize * sizeof(Node)))) {
+ if (
+#if SBRK
+ ((NodePtr)-1)== (hpStart = (NodePtr)sbrk((int)totalSize * sizeof(Node)))
+#else
+ NULL == (hpStart = malloc ((int)totalSize * sizeof(Node)))
+#endif
+ ) {
fprintf(stderr,"Not enough memory for heap and stack.\n");
exit(-1);
}

_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Re: EnumFromThen

2006-04-10T04:23:22Z

hanatani@... wrote:

> I found a bug in nhc98-1.18.

Yes indeed! I wonder how that one escaped attention for so long...

> $ diff src/prelude/Prelude/Enum_Integer.hs src/prelude/Prelude/Enum_Integer.hs
> 14c14
> < enumFromThen x y = x : enumFromThen y (2*y-x)
> ---
> > enumFromThen x y = x : enumFromThen y (2*y+x)

Your patch format is the wrong way round, but I get the idea.
Thanks!

Regards,
Malcolm
_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

EnumFromThen

2006-04-08T01:29:09Z

Hello,

I found a bug in nhc98-1.18.

$ diff src/prelude/Prelude/Enum_Integer.hs src/prelude/Prelude/Enum_Integer.hs
~
14c14
< enumFromThen x y = x : enumFromThen y (2*y-x)
---
> enumFromThen x y = x : enumFromThen y (2*y+x)

--
Yoichi Hanatani

_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs

Linux install troubles

2006-03-27T13:14:42Z

On Fedora Linux, I fetched nhc98-1.18.tar.gz (Linux executable).

Got some relatively innocuous diagnostics from configure and make install, but one stood out:

./configure: line 594: src/runtime/nhc98heap.c: No such file or directory

Bulling through anyway, I did an installation. and compiled a hello world program.
The result appears to be related to the configure trouble:

OS allocated a heap in high memory (>0x80000000)
which breaks this program's run-time system.
hpStart=0xb7d96008, hpEnd=0xb7e0b308

_______________________________________________
Nhc-bugs mailing list
Nhc-bugs@...
http://www.haskell.org/mailman/listinfo/nhc-bugs