Re: ghc-2.04 compiling problems
> > I downloaded ghc-2.04 patch level 2. I then tried compiling it. I'm > using an i386-unknown-solaris2 machine. I get the error shown below. > I tried compiling using ghc-0.29. I was trying for profiled, concurrent > and concurrent profiled builds. ...[compiling ArrBase].. > /tmp/ghc27445.hc:953: fixed or forbidden register was spilled. > This may be due to a compiler bug or to impossible asm > statements or clauses. Hi, you'll need to compile ArrBase with -monly-3-regs, the backend is trying to steal more x86 registers than gcc can handle. If you're still getting the same error, try -monly-2-regs hope that helps, --Sigbjorn
ghc-2.04 compiling problems
I downloaded ghc-2.04 patch level 2. I then tried compiling it. I'm
using an i386-unknown-solaris2 machine. I get the error shown below.
I tried compiling using ghc-0.29. I was trying for profiled, concurrent
and concurrent profiled builds.
Cheers
Meurig
rm -f ghc/ArrBase.o ; if [ ! -d ghc/ArrBase ]; then mkdir ghc/ArrBase ;
else exit 0; fi; find ghc/ArrBase -name '*.o' -print | xargs rm -f
__rm_food;
../../ghc/driver/ghc -recomp -cpp -fglasgow-exts -fvia-C -Rghc-timing
-split-objs -odir ghc/ArrBase -H20m -c ghc/ArrBase.lhs -o
ghc/ArrBase.o -osuf o
Warning: GENERATE_SPECS pre-processing pragma ignored:
{-# GENERATE_SPECS array a{~,Int,IPr} b{} #-}
Warning: GENERATE_SPECS pre-processing pragma ignored:
{-# GENERATE_SPECS (!) a{~,Int,IPr} b{} #-}
Warning: GENERATE_SPECS pre-processing pragma ignored:
{-# GENERATE_SPECS (//) a{~,Int,IPr} b{} #-}
Warning: GENERATE_SPECS pre-processing pragma ignored:
{-# GENERATE_SPECS accum a{~,Int,IPr} b{} c{} #-}
Warning: GENERATE_SPECS pre-processing pragma ignored:
{-# GENERATE_SPECS accumArray a{~,Int,IPr} b{} c{} #-}
Warning: InitTime not found in stats file
Warning: InitElapsed not found in stats file
<>
ghc: module version changed to 1; reason: no old .hi file
/tmp/ghc27445.hc:953: fixed or forbidden register was spilled.
This may be due to a compiler bug or to impossible asm
statements or clauses.
make[2]: *** [ghc/ArrBase.o] Error 1
make[1]: *** [all] Error 2
--
Meurig Sage
Dept of Computing Science
University of Glasgow
http://www.dcs.gla.ac.uk/~meurig
mailto:[EMAIL PROTECTED]
Re: ghc-2.04-pl2 on Linux/HP-UX
>
> Yesterday I successfully installed ghc-2.04-pl2 on Linux and HP-UX10.
> Just a few remarks:
>
>* autoheader complains:
> Symbol `HAVE_ALTZONE' is not covered by
/soft/share/autoconf/acconfig.h
> Ignoring this didn't hurt, but I was just wondering if our
autoheader
> is too old (from autoconf 2.12, latest version, I think)?
>
autoheader needs some help with that one as it is a custom autoconf macro,
I've
appended the necessary acconfig.h stub (which should have been included in
the distrib..)
>* fptools/ghc/lib/cbits/timezone.h still makes some trouble on HPs.
> After commenting out line 31 ('extern time_t timezone;'),
> everything went smooth. The problem seems to be the OS-dependent
> type of timezone: Some OS declare it as long, others as time_t.
> But either way, timezone.h already includes the relevant header
> files, so this redeclaration seems unneccessary.
>
Thanks for the HPUX info, I'll take another stab at getting it Right.
>* makeinfo, ugen and hp2ps are compiled without -O. Not that these
> are the most important and time-consuming commands, but anyway...
>
No good reason why they shouldn't be, let's got for broke :-)
--Sigbjorn
--acconfig.h--
/* Leave this blank line here -- autoheader needs it! */
/* Define this if your time.h defines altzone */
#undef HAVE_ALTZONE
bind error in decl (2)
[Sorry for sending this message twice but I accidentally hit
the deliver button.]
Dear bug chasers,
ghc's latest version sometimes misses conflicting function definitions.
Consider the following *wrong* program:
%--
> module Small ( module Small )
> where
> data MinView t a = Min a (t a)
> | Infty
> newtype ToppedTree a = P (MinView BinTree a)
> data BinTree a= Bin a (BinTree a) (BinTree a)
> | Empty
> P Infty /\ u = P Infty
> t@(P (Min _ _)) /\ P Infty= P Infty
> P (Min a t) /\ P (Min b u)
> | a <= b = P (Min a (Bin b u t))
> | otherwise = P (Min b (Bin a t u))
> data Tree a = Root a (Forest a)
> | Void
> type Forest a = [Tree a]
> Void /\ u= Void
> t@(Root _ _) /\ Void = Void
> t@(Root a ts) /\ u@(Root b us)
> | a <= b = Root a (u : ts)
> | otherwise = Root b (t : us)
%--
Note that /\ is defined twice at different types. ghc replies
bind error in decl (2)
bind error in decl (2)
bind error in decl (2)
Module version unchanged at 3
Ralf
P.S.: Here is the verbose compiler output:
The Glorious Glasgow Haskell Compilation System, version 2.04, patchlevel 0
literate pre-processor:
echo '#line 1 "Small.lhs"' > /tmp/ghc23960.lpp &&
/home/III/a/ralf/FP/ghc-2.04/lib/unlit Small.lhs - >> /tmp/ghc23960.lpp
real0.0
user0.0
sys 0.0
Effective command line: -i/home/III/a/ralf/TBoA/GHC/Import
-L/home/III/a/ralf/TBoA/GHC/lib -lbn -v -c
Ineffective C pre-processor:
echo '#line 1 "Small.lhs"' > /tmp/ghc23960.cpp && cat /tmp/ghc23960.lpp >>
/tmp/ghc23960.cpp
real0.0
user0.0
sys 0.0
Haskell compiler:
/home/III/a/ralf/FP/ghc-2.04/lib/hsc ,-W ,/tmp/ghc23960.cpp
-fignore-interface-pragmas -fomit-interface-pragmas -fsimplify \(
-ffloat-lets-exposing-whnf -ffloat-primops-ok -fcase-of-case -freuse-con
-fpedantic-bottoms -fsimpl-uf-use-threshold3 -fmax-simplifier-iterations4 \)
-himap=/home/III/a/ralf/TBoA/GHC/Import%.hi:.%.hi:/home/III/a/ralf/FP/ghc-2.04/lib/imports%.hi
-v -hifile=/tmp/ghc23960.hi -S=/tmp/ghc23960.s +RTS -H600 -K100
Glasgow Haskell Compiler, version 2.03, for Haskell 1.4
bind error in decl (2)
bind error in decl (2)
bind error in decl (2)
real2.3
user1.9
sys 0.1
Pin on Haskell consistency info:
echo '
.text
hsc.Small.lhs.33.0..:' >> /tmp/ghc23960.s
real0.0
user0.0
sys 0.0
*** New hi file follows...
{-# GHC_PRAGMA INTERFACE VERSION 20 #-}
_interface_ Small
_instance_modules_
ArrBase IO PrelNum
_usages_
PrelBase 1 :: $d14 1 $d15 1 $d26 1 $d27 1 $d32 1 $d34 1 $d37 1 $d39 1 $d41 1 $d44 1
$d45 1 $d46 1 $d49 1 $d51 1 $d6 1 $d7 1 otherwise 1 Eq 1 Eval 1 Ord 1 Ordering 1;
PrelNum 1 :: $d17 1 $d18 1;
_exports_
Small /\ BinTree(Bin Empty) Forest MinView(Min Infty) ToppedTree(P) Tree(Root Void);
_instances_
instance _forall_ [a] => {PrelBase.Eval (Tree a)} = $d1;
instance _forall_ [a] => {PrelBase.Eval (BinTree a)} = $d2;
instance _forall_ [a] => {PrelBase.Eval (ToppedTree a)} = $d3;
instance _forall_ [a :: (* -> *) b] => {PrelBase.Eval (MinView a b)} = $d4;
_declarations_
type Forest r3H = [Tree r3H] ;
data Tree r3J = Root r3J (Forest r3J) | Void ;
data BinTree r3L = Bin r3L (BinTree r3L) (BinTree r3L) | Empty ;
newtype ToppedTree r3N = P (MinView BinTree r3N) ;
data MinView r3P :: (* -> *) r3Q = Min r3Q (r3P r3Q) | Infty ;
/\ _:_ _forall_ [a] {PrelBase.Ord a} => ToppedTree a -> ToppedTree a -> ToppedTree a ;;
$d1 _:_ _forall_ [a] => {PrelBase.Eval (Tree a)} ;;
$d2 _:_ _forall_ [a] => {PrelBase.Eval (BinTree a)} ;;
$d3 _:_ _forall_ [a] => {PrelBase.Eval (ToppedTree a)} ;;
$d4 _:_ _forall_ [a :: (* -> *) b] => {PrelBase.Eval (MinView a b)} ;;
Module version unchanged at 3
Replace .hi file, if changed:
cmp -s Small.hi /tmp/ghc23960.hi-new || ( rm -f Small.hi && cp
/tmp/ghc23960.hi-new Small.hi )
real0.0
user0.0
sys 0.0
Unix assembler:
gcc -o Small.o -c /tmp/ghc23960.s
real0.2
user0.0
sys 0.0
rm -f /tmp/ghc23960*
bind error in decl (2)
Dear bug chasers, ghc's latest version sometimes misses conflicting function definitions. Consider the following *wrong* program: %-- > module Small ( module Small ) > where > data MinView t a = Min a (t a) > | Infty > newtype ToppedTree a = P (MinView BinTree a) > data BinTree a= Bin a (BinTree a) (BinTree a) > | Empty > P Infty /\ u = P Infty > t@(P (Min _ _)) /\ P Infty= P Infty > P (Min a t) /\ P (Min b u) > | a <= b = P (Min a (Bin b u t)) > | otherwise = P (Min b (Bin a t u)) > data Tree a = Root a (Forest a) > | Void > type Forest a = [Tree a] > Void /\ u= Void > t@(Root _ _) /\ Void = Void > t@(Root a ts) /\ u@(Root b us) > | a <= b = Root a (u : ts) > | otherwise = Root b (t : us) %-- Note that /\ is defined twice at different types.
Re: Pattern-matching error within GHC
Thanks for chasing this down, Einar - this (mis)behaviour is reproducable. Fixed in the next release. --Sigbjorn > > The module that caused GHC to fail attempted to reexport a name > from a module that was imported as a qualified module, i.e.: > > module Silly ( > Posix.dupTo, > ... > > ) > import qualified Posix > .. > > Lesson to be learned: dont reexport names that are imported as > qualified names. The following works better: > > > module Silly ( > dupTo, > ... > > ) > import qualified Posix > import Posix(dupTo) > .. > > Einar
Re: Pattern-matching error within GHC
> > While compiling a larger configuration of GHC code (30 KLOC's) with > the brand new compiler 2.04 on a Ultra-Sparc, I got the following > > *** Pattern-matching error within GHC! > > This is a compiler bug; please report it to [EMAIL PROTECTED] > > Fail: "rename/RnNames.lhs", line 453: pattern-match failed on an irrefutable pattern > > I will see if I can reconstruct the error on a smaller configuration of > Haskell code. > Hi, if you recompile rename/RnNames.o with EXTRA_HC_OPTS=-DDEBUG set, this should at least give you the name of the thing in the export list it is processing at the point of failure. It may help to turn off the pruning of type constructors for the module that it is failing on, i.e., add the option -fno-prune-tydecls --Sigbjorn
ghc-2.04-pl2 on Linux/HP-UX
Yesterday I successfully installed ghc-2.04-pl2 on Linux and HP-UX10.
Just a few remarks:
* autoheader complains:
Symbol `HAVE_ALTZONE' is not covered by /soft/share/autoconf/acconfig.h
Ignoring this didn't hurt, but I was just wondering if our autoheader
is too old (from autoconf 2.12, latest version, I think)?
* fptools/ghc/lib/cbits/timezone.h still makes some trouble on HPs.
After commenting out line 31 ('extern time_t timezone;'),
everything went smooth. The problem seems to be the OS-dependent
type of timezone: Some OS declare it as long, others as time_t.
But either way, timezone.h already includes the relevant header
files, so this redeclaration seems unneccessary.
* makeinfo, ugen and hp2ps are compiled without -O. Not that these
are the most important and time-consuming commands, but anyway...
Apart from these really minor glitches, this was the most painless and
easy installation of ghc I had. Moving away from Jmakefiles to GNU make
was clearly the right decision!
Today I will give ghc-2.04-pl2 a try compiling itself, so stay tuned!
:-)
--
Sven PanneTel.: +49/89/2178-2235
LMU, Institut fuer Informatik FAX : +49/89/2178-2211
LFE Programmier- und Modellierungssprachen Oettingenstr. 67
mailto:[EMAIL PROTECTED]D-80538 Muenchen
http://www.pms.informatik.uni-muenchen.de/mitarbeiter/panne
