Known GHC Bugs and Problems

You also wish to consult the archive of the glasgow-haskell-bugs mailing list...

Contents

• Building 0.29 breaks on compiling PreludeIO_p.hc
• Empty arrays crashes compacting collector
• Building 0.29 hangs at qp2ap
• Profiling build `not available' on Solaris
• Stable pointer problems during major GCs
• Assembler trouble under mips-sgi-irix
• HSC can't find getrusage under Solaris 2.4
• Problems with the GLASGOW_HASKELL_ROOT environment variable
• getCPUTime doesn't work poperly for ghc-2.01 (SunOs 4, Solaris 2.x)
• The profiling setup of GHC 2.01 hates PreludeGlaST

This is the FEB (Frequently Encountered Bugs) page for the Glasgow Haskell Compiler, suggesting solutions/workarounds to commonly encountered bugs with GHC (versions 0.29 onward).

You may also wish to consult the archive of the glasgow-haskell-bugs mailing list...

Contents:

• GHC 2.08 bugs and workarounds
  • Typechecking bug with overloaded constructor arguments.

• GHC 2.07 bugs and workarounds
  • happy/src/Parser.hs won't compile with ghc-0.29
  • Can't access () when importing the Prelude qualified
  • Can't compile happy/src/Main.lhs with -O
  • GHC always recompiles modules even if nothing has changed.
  • @seq@ doesn't work with the native code generator

• GHC 2.05 bugs and workarounds
  • hsp shortcoming
  • Crash when discarding polymorphic case
  • Crash in absApply
  • Compiling prelude libraries on x86 machines
  • x86 backend bug
  • Big tuples
  • Fails on 1M threads
  • Interface file parse error for ccalls
  • Record construction for overloaded data types
  • Misleading error message for guards
  • Missed type error for rank-2 types
  • compiling with ghc-0.29 -O
  • missing functions from List
  • Exporting (->) instances

• GHC 2.04 bugs and workarounds
  • "->" isn't an instance of Eval
  • Specialiser doesn't work
  • Same field in two constructors of data decl
  • Exporting qualified imports
  • Trouble compiling up ghc-2.04-src.tar.gz
  • Compiling prelude libraries on x86 machines

• GHC 2.03 bugs and workarounds
  • Function arrow not exported from Prelude
  • Two record bugs
  • Bug in Semaphore.signalQSemN
  • Turning off incomplete pattern match warnings is not possible
  • Compiler loops on type error
  • Cannot link code using CPUTime
  • Undefined symbols when using -prof -auto-all
  • Cannot build user documentation

• GHC 2.02 bugs and workarounds
  • Duplicate definitions in instances
  • Do notation for lists
  • Unqualified Prelude names aren't in scope
  • Multiple type signatures for mutually recursive overloaded functions don't work
  • Duplicate top-level declarations not detected.
  • Wrong type for seq
  • make install breaks when creating directories
  • Compiler panic when compiling profiling code
  • Converting profile data with hp2psfails
  • Unreported symbols when linking -concurrent code
  • GHC for Win32: linker reports unresolved symbol VirtualProtect
  • Cannot use -syslib hbc under Linux
  • What? Perfectly Reasonable Haskell programs don't compile anymore
  • Checksum errors when un-tar'ing the 2.02 distribution
  • Help! I cannot find the function maybe
  • The compiler cannot find a Glasgow extension like unboxed types
  • Using letrec as identifier does not work
  • Strictness annotations in data types aren't parsed correctly
  • Installation fails during copying of library files
  • Empty comment ending the file breaks the parser

• GHC 0.29 bugs and workarounds
  • Using dupChan does not work(2.01 and 0.29)
  • Foreign objects prematurely released
  • Building 0.29 breaks on compiling PreludeIO_p.hc
  • Empty arrays crashes compacting collector
  • Building 0.29 hangs at qp2ap
  • Profiling build `not available' on Solaris
  • Assembler trouble under mips-sgi-irix
  • HSC can't find getrusage under Solaris 2.4
  • Problems with the GLASGOW_HASKELL_ROOT environment variable

GHC 2.08 bugs and workarounds

Typechecking bug with overloaded constructor arguments.

The following module fails to typecheck:

module Test where
data Empty q			=  Empty (Ord a => q a)
q				:: (Ord a) => [a]
q				=  []
e				:: Empty []
e				=  Empty q

This bug is fixed in later versions.

GHC 2.07 bugs and workarounds

happy/src/Parser.hs won't compile with ghc-0.29

If you get something like this while compiling happy:

	ghc-0.29 -cpp -fhaskell-1.3 -fglasgow-exts   -Onot  -c Parser.hs -o Parser.o -osuf o
	"Parser.hs", line 572, column 14: parse error on input: "->"

Edit Parser.hs and change newtype HappyState to data HappyState.

Can't access () when importing the Prelude qualified

Programs such as the following don't work:

	import qualified Prelude

	f :: Prelude.IO ()
	f = Prelude.print Prelude.True

Workaround: use

	import qualified Prelude hiding (())
	import Prelude (())

Can't compile happy/src/Main.lhs with -O

This bug manifests itself as a non-terminating compilation when GHC (version 2.xx) is applied to happy/src/Main.lhs with the -O flag. It may happen on other modules, but we haven't found any yet. Workaround: add the -fno-update-analysis flag.

GHC always recompiles modules, even if nothing has changed

The recompilation checker in GHC 2.07 was accidentally turned off. Apply the following patch to the driver:

*** fptools/ghc/driver/ghc.lprl	1997/09/03 23:48:25	1.35
--- fptools/ghc/driver/ghc.lprl	1997/10/05 20:13:24
***************
*** 2546,2550 ****
      #---------- what phases are to be run ----------------------------------
!     /^-(no-)?recomp/	    && do { $Do_recomp_chkr = ($1 ne '') ? 1 : 0; next arg; };
  
      /^-cpp$/	    && do { $Cpp_flag_set = 1; next arg; };
--- 2546,2550 ----
  
      #---------- what phases are to be run ----------------------------------
!     /^-(no-)?recomp/	    && do { $Do_recomp_chkr = ($1 eq '') ? 1 : 0; next arg; };
  
      /^-cpp$/	    && do { $Cpp_flag_set = 1; next arg; };

seq doesn't work with the native code generator

There currently isn't any support for seq in the native code generator. The workaround is to compile with the -fvia-C flag.

GHC 2.05 bugs and workarounds

hsp shortcoming

The standalone parser (hsp) does not take precedence information into consideration when parsing. The reason being that precedence parsing was moved into the compiler proper in 2.0x.

This shortcoming only affects people using hsp, users of GHC are not affected.

Crash when discarding polymorphic case

GHC 2.05 sometimes emits the message

	Warning: discarding polymorphic case: tpl_s2vz

and then subsequently dies with

	panic! (the `impossible' happened):
	lookupBindC:no info!
	 for: tpl_s2vz

This is a straightforward bug in simplCore/SimplCore.lhs. Workaround: don't use "!" in data type declarations.

You can easily fix the bug if you have source code: just replace this equation of tidyCoreExpr. The lines marked "!!" are the changed ones.

     -- Eliminate polymorphic case, for which we can't generate code just yet
    tidyCoreExpr (Case scrut (AlgAlts [] (BindDefault deflt_bndr rhs)))
      | not (typeOkForCase (idType deflt_bndr))
      = pprTrace "Warning: discarding polymorphic case:" (ppr PprDebug scrut) $
	case scrut of
    !!	Var v -> lookupId v	`thenTM` \ v' ->
    !!		 extendEnvTM deflt_bndr v' (tidyCoreExpr rhs)
	    other -> tidyCoreExpr (Let (NonRec deflt_bndr scrut) rhs)

Crash in absApply

When compiling with -O, under certain circumstances too tiresome to describe in detail, you may see this:

 Fail: "stranal/SaAbsInt.lhs", line 671: incomplete pattern(s) to match in function "absApply"

The fix is to change the call to splitForAllTy to a call to splitForAllTyExpandingDicts in the body of mkWrapper in stranal/WwLib.lhs. Fixed in 2.06.

Compiling prelude libraries on x86 machines

This bug has persisted from 2.04. See the description there.

x86 backend -monly-n-regs bug

The x86 backend has got a bug related to the handling of the -monly-<x>-regs option, the code it generates for calling the garbage collector does under some specific circumstances cause a crash. To fix this, apply the following patch to the mangler:

*** fptools/ghc/driver/ghc-asm.lprl~	Wed Jul 30 14:40:09 1997
--- fptools/ghc/driver/ghc-asm.lprl	Wed Aug 27 15:53:33 1997
***************
*** 1149,1154 ****
--- 1149,1156 ----
      local($exit_patch)	= '';
      local($call_entry_patch)= '';
      local($call_exit_patch)	= '';
+     local($gc_call_entry_patch)= '';	# Patches before and after calls to Perform_GC_wrapper
+     local($gc_call_exit_patch)	= '';
  
  #OLD:	# first, convert calls to *very magic form*: (ToDo: document
      # for real!)  from
***************
*** 1238,1253 ****
--- 1240,1265 ----
  
      # OK, now we can decide what our patch-up code is going to
      # be:
+ 
+ 	# Note funky ".=" stuff; we're *adding* to these _patch guys
      if ( $StolenX86Regs <= 2
  	 && ( /32\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
  	$entry_patch .= "\tmovl \%esi,32(\%ebx)\n";
  	$exit_patch  .= "\tmovl 32(\%ebx),\%esi\n";
+ 
+ 	$gc_call_entry_patch  .= "\tmovl \%esi,32(\%ebx)\n";
+ 	$gc_call_exit_patch .= "\tmovl 32(\%ebx),\%esi\n";
+ 
  	# nothing for call_{entry,exit} because %esi is callee-save
      }
      if ( $StolenX86Regs <= 3
  	 && ( /64\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # SpA (edi)
  	$entry_patch .= "\tmovl \%edi,64(\%ebx)\n";
  	$exit_patch  .= "\tmovl 64(\%ebx),\%edi\n";
+ 
+ 	$gc_call_entry_patch  .= "\tmovl \%edi,64(\%ebx)\n";
+ 	$gc_call_exit_patch .= "\tmovl 64(\%ebx),\%edi\n";
+ 
  	# nothing for call_{entry,exit} because %edi is callee-save
      }
  #=  if ( $StolenX86Regs <= 4
***************
*** 1265,1271 ****
      # next, here we go with non-%esp patching!
      #
      s/^(\t[a-z])/$entry_patch$1/; # before first instruction
!     s/^(\tcall .*\n(\taddl \$\d+,\%esp\n)?)/$call_exit_patch$1$call_entry_patch/g; # _all_ calls
  
      # fix _all_ non-local jumps:
  
--- 1277,1290 ----
      # next, here we go with non-%esp patching!
      #
      s/^(\t[a-z])/$entry_patch$1/; # before first instruction
! 
! # Actually, call_entry_patch and call_exit_patch never get set,
! # so let's nuke this one
! #    s/^(\tcall .*\n(\taddl \$\d+,\%esp\n)?)/$call_exit_patch$1$call_entry_patch/g; # _all_ calls
! 
! # Before calling GC we must set up the exit condition before the call
! # and entry condition when we come back
!     s/^(\tcall ${T_DO_GC}\n(\taddl \$\d+,\%esp\n)?)/$gc_call_exit_patch$1$gc_call_entry_patch/g; # _all_ calls
  
      # fix _all_ non-local jumps:

After having applied it, you need to remake the contents of ghc/driver, followed by a complete rebuild of the prelude (and hslibs) libraries.

[Thanks to David Barton for reporting a problem that (eventually!) turned out to be caused by the above bug.]

Big tuples

If your program needs very large tuples (more than about 37 components) you'll get a rather horrid message from the linker, something like:

  ld: Software Generation Utilities (SGU) SunOS/ELF (LK-2.0 (S/I) - versioning)
  Undefined                       first referenced
   symbol                             in file
  PrelTup_Z40Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44
  Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z4
  4Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z44Z41_con_info 
  Galileo6.o
  ld: fatal: Symbol referencing errors.

To solve this you need to add an extra data declaration to ghc/lib/ghc/PrelTup.lhs and rebuild libHS.a. You can't do this from a binary distribution; you have to work from source (at least source for the libraries).

This has been the case with every version of GHC, and we don't have any immediate plans to fix it. Solving it "properly" requires more work than it appears to justify....but please tell us if you think differently!

Fails on 1M threads

Einar reports that GHC 2.04 (and hence almost certainly 2.05) finds "No runnable threads" after sparking 1M threads with the following program:

module Main (
	main 
	) 
where

import Concurrent

main :: IO ()
main = spawner forkIO 1000000 	

spawner :: (IO () -> IO ()) -> Int -> IO ()
spawner c 0 = print "done"
spawner c n = do { {- print n ; -} c (spawner c (n-1)); return ()}

Interface file parse error for ccalls

GHC 2.05 fails to parse an interface file with ccalls that involve polymorphic types. The error message is a bit confusing, and it happens when you compile hslibs/ghc/src/Readline.lhs with -O.

    src/Readline.lhs:13:warning:
        Warning: failed to find (optional) interface decl for
        `Foreign.ts9Tr'
 
    panic! (the `impossible' happened):
            tcLookupTyVar: Foreign.ts9Tr

Workaround: compile either the exporting module or the importing module without -O.

Record construction for overloaded data types

Record construction using overloaded data constructors causes a compiler crash. Simple workaround: omit the context in the data declaration.

Misleading error message for guards

When you have a type error in a guard, GHC 2.05's error message misleadingly mentions list comprehensions. Harmless, but confusing.

Missed type error for rank 2 types

GHC 2.05 fails to report an error on some programs involving rank-2 polymorphism. This can only bite you if you are using runST or the GHC extensions involving local universal quantification.

compiling with ghc-0.29 -O

When compiling up the 2.05 sources with ghc-0.29 compiler with -O turned on, the compilation of ghc/compiler/absCSyn/PprAbsC.lhs fails as follows:

"codeGen/ClosureInfo.hi", line 72:
    undefined type constructor: StgSyn.StgBinderInfo
 
"codeGen/ClosureInfo.hi", line 74:
    undefined type constructor: CgMonad.CgInfoDownwards
 
"codeGen/ClosureInfo.hi", line 74: undefined type constructor: CgMonad.CgState
 
"codeGen/ClosureInfo.hi", line 74: undefined type constructor: CgMonad.CgState

[...and many more like it...]

The reason for the break is that the ClosureInfo interface isn't properly closed for the Haskell 1.2 module system to be happy. To fix this, either compile the PprAbsC without optimisations turned on, or apply the following patch:

*** ghc/compiler/ClosureInfo.lhs	1997/07/26 09:49:29	1.23
--- ghc/compiler/ClosureInfo.lhs	1997/08/02 21:27:13	1.24
***************
*** 17,22 ****
--- 17,23 ----
  
  	mkClosureLFInfo, mkConLFInfo, mkSelectorLFInfo, mkVapLFInfo,
  	mkLFImported, mkLFArgument, mkLFLetNoEscape,
+ 	UpdateFlag,
  
  	closureSize, closureHdrSize,
  	closureNonHdrSize, closureSizeWithoutFixedHdr,
***************
*** 27,38 ****
  	layOutStaticNoFVClosure, layOutPhantomClosure,
  	mkVirtHeapOffsets,
  
! 	nodeMustPointToIt, getEntryConvention,
  	blackHoleOnEntry,
  
  	staticClosureRequired,
  	slowFunEntryCodeRequired, funInfoTableRequired,
  	stdVapRequired, noUpdVapRequired,
  
  	closureId, infoTableLabelFromCI, fastLabelFromCI,
  	closureLabelFromCI,
--- 28,42 ----
  	layOutStaticNoFVClosure, layOutPhantomClosure,
  	mkVirtHeapOffsets,
  
! 	nodeMustPointToIt, getEntryConvention, 
! 	SYN_IE(FCode), CgInfoDownwards, CgState, 
! 
  	blackHoleOnEntry,
  
  	staticClosureRequired,
  	slowFunEntryCodeRequired, funInfoTableRequired,
  	stdVapRequired, noUpdVapRequired,
+ 	StgBinderInfo,
  
  	closureId, infoTableLabelFromCI, fastLabelFromCI,
  	closureLabelFromCI,
***************
*** 40,45 ****
--- 44,50 ----
  	closureLFInfo, closureSMRep, closureUpdReqd,
  	closureSingleEntry, closureSemiTag, closureType,
  	closureReturnsUnboxedType, getStandardFormThunkInfo,
+ 	GenStgArg,
  
  	isToplevClosure,
  	closureKind, closureTypeDescr,		-- profiling

[Thanks to Sven Panne for reporting this problem]

missing functions from List

The functions unionBy and intersectBy are missing from the standard library List.

[Reported by Bjarte M. Oestvold - thanks!]

Exporting (->) instances

Exporting (->) instances breaks the interface file parser in 2.05:

instance Foo (->) where
    ....

To fix this, apply the following patch to the 2.05 sources:

*** ghc/compiler/reader/Lex.lhs.~1~    Wed Jul 30 14:35:11 1997
--- ghc/compiler/reader/Lex.lhs        Tue Aug 19 19:18:59 1997
****************
*** 664,669 ****
--- 664,676 ----
       ','# -> lex_tuple module_dot (stepOnBy# buf 2#)
       _    -> lex_id3 module_dot buf
    ':'# -> lex_id3 module_dot (incLexeme buf)
+   '-'# ->
+     case module_dot of
+       Nothing  -> lex_id3 module_dot buf
+       Just ghc -> -- this should be "GHC" (current home of (->))
+         case lookAhead# buf 1# of
+          '>'# -> end_lex_id module_dot (ITconid SLIT("->")) (stepOnBy# buf 2#)
+          _    -> lex_id3 module_dot buf
     _    -> lex_id3 module_dot buf

[Thanks to David Barton for reporting this]

GHC 2.04 bugs and workarounds

"->" isn't an instance of Eval

"->" isn't an instance of Eval. It should be.

Specialiser doesn't work

The specialiser, which removes unnecessary overloading, still isn't working in 2.04. So your programs won't get specialised and -fshow-specialisations work work.

Same field in two constructors of data decl

GHC 2.04 won't let you have the same field name in two constructors of the same data type. For example

  data T = T1 { a :: Int } | { b :: Int }

Alas, there's no workaround. Tell us if this is a real nuisance. (It'll be fixed in the next release, of course.)

Exporting qualified imports

GHC 2.04 dies if you try to export something that's only imported qualified. Workaround: import it unqualified).

Trouble compiling up ghc-2.04-src.tar.gz

The distribution ghc-2.04-src.tar.gz has a couple of build-related problems that will cause some trouble when you try to compile it up. A patched release (patchlevel 2) is available from the GHC downloading page

Compiling prelude libraries on x86 machines

When compiling the prelude libraries on x86 machines, you're more than likely to run into a panic from gcc when compiling ghc/ArrBase.o, it saying something like:

/tmp/ghc29172.hc:953: fixed or forbidden register was spilled.
This may be due to a compiler bug or to impossible asm
statements or clauses.

The solution is to compile that module with the switch -monly-2-regs added, i.e., do make EXTRA_HC_OPTS=-monly-2-regs ghc/ArrBase.o ; make to continue. For more info as to what went wrong, see Section 7.4.3 of the user's guide.

GHC 2.03 bugs and workarounds

Function arrow not exported from Prelude

The Prelude doesn't export (->). Workaround: get it from PrelBase.

Two record bugs

GHC 2.03 complains erroneously if you have the same field in multiple constructors:

	data X = A {a :: Int} | B {a :: Int}

In addition using a record pattern on the left hand side of a binding fails:

     let
	A {a} = x
     in ...

Thanks to Tomasz Cholewo.

Bug in Semaphore.signalQSemN

There's a bug in the signalQSemN function in the Semaphore module. (In versions ghc-0.29 through 2.03). The fixed code is below; the one line that differs from 2.03 is indicated.

 signalQSemN :: QSemN -> Int  -> IO ()
 signalQSemN (QSemN sem) n
  = takeMVar sem                  >>= \ (avail,blocked) ->
    free (avail+n) blocked        >>= \ (avail',blocked') ->
    putMVar sem (avail',blocked') >>
    return ()
    where
     free avail [] = return (avail,[])
     free avail ((req,block):blocked) =
!     if avail >= req then
        putMVar block () >>
        free (avail-req) blocked
      else
        free avail blocked >>= \ (avail',blocked') ->
        return (avail',(req,block):blocked')

[Thanks to Meurig Sage.]

Turning off incomplete pattern match warnings is not possible

For 2.02 and 2.03, the command line option to ghc, -fno-warn-incomplete-patterns does not have any effect, warnings will still be issued. Fixed in the next release. [Thanks to Satnam Singh for reporting this one.]

Compiler loops on type error

It is possible to make GHC 2.03 loop in the type checker! The error recovery mechanism isn't quite right, and it retries the erroneous thing it's just found an error in. This doesn't happen very often, and you should be able to tell if it does because you get lots of identical type-error messages. [Thanks to Marc van Dongen for this one.]

Cannot link code using CPUTime

The 2.03 distribution is missing a pair of files containing vital C stubs for executing actions related to CPUTime, so trying to use CPUTime.getCPUTime and friends will result in abject failure.

Help is at hand though, pick up

ftp://ftp.dcs.gla.ac.uk/pub/haskell/glasgow/working/ghc-2.03-src-MISSING.tar.gz

and unpack it on top of the source distribution. If you've got a working a build, typing make all in fptools/ghc/lib/cbits will suffice.

Undefined symbols when using -prof -auto-all

Symptom: Under 2.03, when compiling code with the profiling options -prof -auto-all, undefined symbols (starting with CC_) are reported when linking.

Solution:

Apply the following source patch and recompile the compiler proper: (patch also available from ftp://ftp.dcs.gla.ac.uk/pub/haskell/glasgow/working/ghc-2.03-prof.patch)

*** profiling/CostCentre.lhs	Mon Mar 10 23:41:44 1997
--- profiling/CostCentre.lhs-CURRENT	Mon May 12 10:43:19 1997
***************
*** 389,396 ****
      do_cc (NormalCC kind mod_name grp_name is_dupd is_caf)
        = let
! 	    basic_kind = do_caf is_caf ++ do_kind kind
  	in
  	if friendly_sty then
! 	    do_dupd is_dupd (basic_kind ++ ('/': moduleString mod_name) ++ ('/': _UNPK_ grp_name))
  	else
  	    basic_kind
--- 389,399 ----
      do_cc (NormalCC kind mod_name grp_name is_dupd is_caf)
        = let
! 	    basic_kind = do_caf is_caf           ++ 
! 	                 moduleString mod_name   ++ 
! 			 ('/' : _UNPK_ grp_name) ++ 
! 			 ('/' : do_kind kind)
  	in
  	if friendly_sty then
! 	    do_dupd is_dupd basic_kind
  	else
  	    basic_kind
***************
*** 403,411 ****
  	do_kind (DictCC id)   = do_id id ++ (if friendly_sty then "/DICT" else "")
  
  	do_id :: Id -> String
! 	do_id id
! 	  = if print_as_string
! 	    then  getOccString id		-- use occ name
! 	    else showId sty id	 		-- we really do
  
      ---------------
--- 406,415 ----
  	do_kind (DictCC id)   = do_id id ++ (if friendly_sty then "/DICT" else "")
  
+         {-
+ 	 do_id is only applied in a (not print_as_string) context for local ids,
+ 	 hence using the occurrence name is enough.
+ 	-}
  	do_id :: Id -> String
! 	do_id id = getOccString id
  
      ---------------
*** reader/Lex.lhs	Mon Apr  7 02:00:02 1997
--- reader/Lex.lhs-NEW	Mon Apr 28 13:56:53 1997
***************
*** 385,397 ****
  	                        lexIface (stepOn (stepOverLexeme buf''))
                       Nothing ->
                        case prefixMatch (stepOn buf) "CAF:" of
!                        Just buf' ->		  
! 	                case untilChar# (stepOverLexeme buf') '\"'# of
! 	                 buf'' -> ITscc (cafifyCC (mkUserCC (lexemeToFastString buf'') _NIL_ _NIL_)): 
! 	                          lexIface (stepOn (stepOverLexeme buf''))
                         Nothing ->
! 	                case untilChar# (stepOn buf) '\"'# of
! 	                   buf' -> ITscc (mkUserCC (lexemeToFastString buf') _NIL_ _NIL_): 
!                                    lexIface (stepOn (stepOverLexeme buf'))
    c -> ITunknown [C# c] : lexIface (stepOn buf)
  
--- 385,409 ----
  	                        lexIface (stepOn (stepOverLexeme buf''))
                       Nothing ->
+ 		      let
+ 		       match_user_cc buf =
+ 	                case untilChar# buf '/'# of
+ 	                 buf' -> 
+                           let mod_name = lexemeToFastString buf' in
+ 	                  case untilChar# (stepOn (stepOverLexeme buf')) '/'# of
+ 	                   buf'' -> 
+                             let grp_name = lexemeToFastString buf'' in
+ 			    case untilChar# (stepOn (stepOverLexeme buf'')) '\"'# of
+ 			     buf''' ->
+ 			       let cc_name = lexemeToFastString buf''' in
+ 			       (mkUserCC cc_name mod_name grp_name, 
+ 			        stepOn (stepOverLexeme buf'''))
+                       in
                        case prefixMatch (stepOn buf) "CAF:" of
!                        Just buf' ->
! 			 case match_user_cc (stepOverLexeme buf') of
! 			  (cc, buf'') -> ITscc (cafifyCC cc) : lexIface buf''
                         Nothing ->
! 	                 case match_user_cc (stepOn buf) of
! 			  (cc, buf'') -> ITscc cc : lexIface buf''
    c -> ITunknown [C# c] : lexIface (stepOn buf)

Cannot build user documentation

The user documentation that is bundled with the 2.03 source release has got a couple of files missing, a new, fixed copy can be found at

ftp://ftp.dcs.gla.ac.uk/pub/haskell/glasgow/ghc-2.03-user-docs.tar.gz

Install this on top of the 2.03 documentation, i.e., in fptools/ghc/docs/users_guide/

GHC 2.02 bugs and workarounds

Duplicate definitions in instances

If you give two separate declarations for the same method in an instance declaration, you'll get a compiler failure. Workaround: don't.

Do notation for lists

If you use the do notation at the list type you'll elicit a compiler crash. For example:

	foo = do x <- [1..10]
		 return (x+1)

elicits the crash:

*** Pattern-matching error within GHC!
 
This is a compiler bug; please report it to glasgow-haskell-bugs@dcs.gla.ac.uk.
 
Fail: "deSugar/DsListComp.lhs", line 118: incomplete pattern(s) to match in function "deListComp"

The workaround is to use the list comprehension notation instead. (This tiresome bug was caused by my trying to retain some old list-comprehension code that does deforestation.)

Unqualified Prelude names aren't in scope

When you give an explicit import of Prelude, the unqualified names are meant to be in scope regardless. For example:

  module M (myfoldr) where
  import Prelude hiding (foldr)
  myfoldr = Prelude.foldr

GHC 2.02 complains that Prelude.foldr isn't in scope. There was an ambiguity in the Haskell Report, which is now cleared up. In general, a hiding clause now does not ever hide qualified names, so Prelude.foldr should be in scope; and this applies to all modules, not just the Prelude.

Multiple type signatures for mutually recursive overloaded functions don't work

If you give type signatures for more than one overloaded function in a mutually recursive group you'll get a bogus type error. For example:

a :: (Ord f) => f 
a = b

b :: (Ord f) => f 
b = a 

elicits the message

tmp.lhs:1: A group of type signatures have mismatched contexts
               Abf.a ::  (PrelBase.Ord f{-aX6-}) => ...
               Abf.b ::  (PrelBase.Ord f{-aX2-}) => ... 

>>>>>>> 1.41

<<<<<<< ghc-bugs.html
Symptom:


Utter misery when building 0.29 a non-sequential (i.e., profiling, concurrent etc.)
prelude from sources: 

foo% pwd;make
/ghc-BUILDS/0.29/ghc/lib
[....]
../../ghc/driver/ghc -c   -hi-diffs -dcore-lint -link-chk  -iprelude:glaExts   
-dcore-lint -fshow-pragma-name-errs -fshow-import-specs -fomit-default-instance-methods       
-cpp -fglasgow-exts -genSPECS -DUSE_FOLDR_BUILD -D__OVERLAPPING_INSTANCES__
-H40m -prelude -fno-implicit-prelude -fomit-reexported-instances 
-split-objs PIO -ohi prelude/PreludeIO.hi -fno-omit-reexported-instances 
'-#include"stgio.h"' -hisuf _p.hi  -O  -prof -GPrelude prelude/PreludeIO_p.hc 
-odir prelude/PreludeIO -osuf    '_p.o'

make[1]: *** No rule to make target 
`/local/grasp_tmp3/hwloidl/sparc-sun-sunos4/./ghc/lib/../../ghc/includes/error.h',
 needed by `prelude/PreludeIOError_p.hc'.  Stop.
foo%


Solution:


The ghc/lib/Jmake.inc5 in the source distribution contains some links
to local Glasgow files. To fix, do `make print_file_lists' and do
make again (this will still add some hard-wired paths to the Jmake.inc? files,
but at least they're yours now :-)



Empty arrays crashes compacting collector

Symptom:


Programs compiled under either the 1.2 compiler (version 0.xx) or the 1.3 compiler (v. 2.xx)
that uses empty arrays (i.e., arrays with bounds (a,b) that define an empty
range (a>b)) crash and burn with the (default) generational garbage collector,
reporting "Segmentation fault" (or the like). Using the two space collector works
fine - what gives? 

Solution:


The problem is caused by the marking code in the garbage collector assumed
that arrays had to be non-empty - which is not the case for arrays. Fix for 0.29
and 2.01 in their next releases - but if you've got the source installed locally,
apply the following patch and recompile the RTS:

*** ghc/runtime/storage/SMmark.lhc      Mon Aug  5 06:14:38 1996
--- ../../../ghc-0.29/ghc/runtime/storage/SMmark.lhc    Tue Jul 23 17:32:21 1996
***************
*** 666,674 ****
          JUMP_MARK_RETURN;
      } else {
!         INIT_MARK_NODE("MUT ", MUTUPLE_CLOSURE_NoPTRS(Mark));
!       if (MUTUPLE_CLOSURE_NoPTRS(Mark) > 0)
!          INIT_MSTACK(MUTUPLE_CLOSURE_PTR);
!       else
!          JUMP_MARK;
      }
      FUNEND;
--- 666,671 ----
          JUMP_MARK_RETURN;
      } else {
!     INIT_MARK_NODE("MUT ", MUTUPLE_CLOSURE_NoPTRS(Mark));
!       INIT_MSTACK(MUTUPLE_CLOSURE_PTR);
      }
      FUNEND;




Building 0.29 hangs at qp2ap

Symptom:


On some platforms, when the STARTUP script gets to
ghc/utils/parallel/, it hangs at...
   
   rm -f qp2ap.pl
   echo "#!"/bin/perl > qp2ap.pl
   ../../../glafp-utils/msub/msub  qp2ap.pl >> qp2ap.pl || ( rm -f qp2ap.pl && exit 1 )


Solution:


Drop the spurious first .pl in the qp2ap in
the Jmakefile:
*** ghc-0.29/ghc/utils/parallel/Jmakefile.~1~
--- ghc-0.29/ghc/utils/parallel/Jmakefile
***************
*** 40,46 ****
  MsubProgramScriptTarget(PerlCmd,gr2java,gr2java.pl,,)
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2jv,gr2jv.bash,,)
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2ap,gr2ap.bash,,)
! MsubProgramScriptTarget(PerlCmd,qp2ap.pl,qp2ap.pl,,)
  
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2gran,gr2gran.bash,,)
  MsubProgramScriptTarget(PerlCmd,gr2RTS,gr2RTS.pl,,)
--- 40,46 ----
  MsubProgramScriptTarget(PerlCmd,gr2java,gr2java.pl,,)
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2jv,gr2jv.bash,,)
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2ap,gr2ap.bash,,)
! MsubProgramScriptTarget(PerlCmd,qp2ap,qp2ap.pl,,)
  
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2gran,gr2gran.bash,,)
  MsubProgramScriptTarget(PerlCmd,gr2RTS,gr2RTS.pl,,)




Profiling build `not available' on Solaris

Symptom:


When trying the pre-built profiling stuff for sparc-sun-solaris2,
you may get:
   % ghc-0.29 -prof Main.hs
   ghc-0.29: a `profiling' "build" is not available with your GHC setup.
   (It was not configured for it at your site.)


Solution:


Simply edit your driver (ghc-0.29), around about
line 207, to read:
%BuildAvail	= ('',	    '-build-normal-defined',
		   '_p',    '-build-p-defined', # NB: this line here
		   ...




Stable Pointers problems

Symptom:


Programs that use the whizzy stable pointers (references from the outside
into the Haskell heap) break during the first major garbage collection.
=======

Workaround: just one type signature for each mutually recursive group.

Duplicate top-level declarations not detected.

GHC 2.02 fails to detect duplicate top-level declarations. This can actually lead to a core dump, as the type checker picks one, and the code generator picks the other.

Wrong type for seq

GHC 2.02 erroneously thinks that seq has type a -> b -> b instead of Eval a => a -> b -> b.

This bug is unlikely to break any programs, especially since 2.02 treats seq as the identity function. (This is a serious, albeit advertised, shortcoming.)

make install breaks when creating directories

The install script that comes with the binary bundles is faulty - failing to create anything but the simplest of directory paths.

A substitute mkdirhier script is included below, to use it just replace it with the old mkdirhier script:

#!/bin/sh

#
# create a hierarchy of directories, a mixture between
# the glafp-utils/mkdirhier/ and Noah Friedman's mkinstalldirs
#

errs=0

for f in $*; do
    parts=`echo ":$f" | sed -ne 's/^:\//#/;s/^://;s/\// /g;s/^#/\//;p'`
    path="";
    for p in $parts; do
        path="$path$p"
        case "$path" in
          -* ) path=./$path ;;
        esac

        if test ! -d "$path"; then
           echo "mkdir $path" #1>&2

           mkdir "$path" || lasterr=$?
	   
	   if test ! -d "$path"; then
	      errs=$lasterr
           fi 
        fi
	path="$path/";
    done;
done

exit $errs

# end of story

Unreported symbols when linking -concurrent code

The driver script (ghc.prl) as supplied with the various 2.02 bundles fails to tell the linker about which symbols the RTS need from the prelude libraries, so that these will always be included (the archive for the prelude libraries appear before the archive for the run-time system on the linker command line - rather brittle setup). This problem is only going to strike in Concurrent code that doesn't make use of lists, Ints, Chars or the IO/ST monad - i.e., not many, but the distributed driver script is broken nonetheless.

Two ways to recover from this impassé, either add -lHSrts_{mc,mr} (mc for -concurrent code, mr for -prof -concurrent) to the cmd. line when linking or patch the driver script with the patch available from

 ftp://ftp.dcs.gla.ac.uk/pub/haskell/glasgow/working/ghc-2.02-ghc.prl-patch

applying it (patch -p < ghc-2.02-ghc.prl-patch) to the driver script in fptools/bin/<your platform>/ghc-2.02/, and then re-run make in-place or re-install. Patching the driver is the preferred solution.

Thanks to Stefan Westmeier <stefan@stefan.hb.north.de> for reporting the problem.

Converting profile data with hp2psfails

The run-time system tags foreign objects with incorrect profiling label, which will flummox the hp2ps utility when converting the raw profiling data into PostScript. Patch included - the following sed script can be used to massage the profiling data into the right shape.

sed -e 's/FOREIGN OBJ/FOREIGN_OBJ/g' file.hp > file.hp.new 
mv file.hp.new file.hp

Source patch:

*** fptools/ghc/runtime/main/StgStartup.lhc-old     Thu Mar  6 20:12:53 1997
--- fptools/ghc/runtime/main/StgStartup.lhc   Mon Apr  7 15:02:05 1997
***************
*** 126,128 ****
  
! ForeignObj_ITBL(ForeignObj_info,ForeignObj_entry,UpdErr,0,INFO_OTHER_TAG,,,const,EF_,ForeignObj_K,"FOREIGN OBJ","ForeignObj");
  
--- 126,128 ----
  
! ForeignObj_ITBL(ForeignObj_info,ForeignObj_entry,UpdErr,0,INFO_OTHER_TAG,,,const,EF_,ForeignObj_K,"FOREIGN_OBJ","ForeignObj");
  
***************
*** 147,149 ****
  
! STATIC_ITBL(UnusedSP_info,UnusedSP_entry,UpdErr,0,INFO_OTHER_TAG,0,0,const,IF_,CON_K,"UNUSED STABLE PTR","USP");
  
--- 147,149 ----
  
! STATIC_ITBL(UnusedSP_info,UnusedSP_entry,UpdErr,0,INFO_OTHER_TAG,0,0,const,IF_,CON_K,"UNUSED_STABLE_PTR","USP");

Compiler panic when compiling profiling code

2.02 has got a bug in the lexer for interface files, causing it to break when compiling optimised profiling code (i.e., -O -prof). Source patches included below that fixes this.

If you don't have a source distribution, the easiest workaround is not to use -O and -prof in conjunction.

*** fptools/ghc/compiler/reader/Lex.lhs-old Tue Mar 18 23:55:12 1997
--- fptools/ghc/compiler/reader/Lex.lhs     Mon Apr  7 02:00:02 1997
***************
*** 304,307 ****
           's'# -> case prefixMatch (stepOnBy# buf 2#) "cc_" of
!                    Just buf' -> lex_scc (stepOnUntil (not . isSpace) 
!                                                      (stepOverLexeme buf'))
                     Nothing   -> lex_keyword (stepOnBy# buf 1#) -- drop the '_' and assume
--- 304,306 ----
           's'# -> case prefixMatch (stepOnBy# buf 2#) "cc_" of
!                    Just buf' -> lex_scc (stepOnUntil (not . isSpace) (stepOverLexeme buf'))
                     Nothing   -> lex_keyword (stepOnBy# buf 1#) -- drop the '_' and assume
***************
*** 376,378 ****
                   buf'' -> ITscc (mkAllCafsCC ({-module-}lexemeToFastString buf'') _NIL_): 
!                           lexIface (stepOverLexeme buf'')
                   Nothing ->
--- 375,377 ----
                   buf'' -> ITscc (mkAllCafsCC ({-module-}lexemeToFastString buf'') _NIL_): 
!                           lexIface (stepOn (stepOverLexeme buf''))
                   Nothing ->
***************
*** 385,387 ****
                       buf'' -> ITscc (mkAllDictsCC (lexemeToFastString buf'') _NIL_ True): 
!                               lexIface (stepOverLexeme buf'')
                       Nothing ->
--- 384,386 ----
                       buf'' -> ITscc (mkAllDictsCC (lexemeToFastString buf'') _NIL_ True): 
!                               lexIface (stepOn (stepOverLexeme buf''))
                       Nothing ->
***************
*** 391,393 ****
                         buf'' -> ITscc (cafifyCC (mkUserCC (lexemeToFastString buf'') _NIL_ _NIL_)): 
!                                 lexIface (stepOverLexeme buf'')
                         Nothing ->
--- 390,392 ----
                         buf'' -> ITscc (cafifyCC (mkUserCC (lexemeToFastString buf'') _NIL_ _NIL_)): 
!                                 lexIface (stepOn (stepOverLexeme buf''))
                         Nothing ->
***************
*** 395,397 ****
                           buf' -> ITscc (mkUserCC (lexemeToFastString buf') _NIL_ _NIL_): 
!                                  lexIface (stepOverLexeme buf')
    c -> ITunknown [C# c] : lexIface (stepOn buf)
--- 394,396 ----
                           buf' -> ITscc (mkUserCC (lexemeToFastString buf') _NIL_ _NIL_): 
!                                    lexIface (stepOn (stepOverLexeme buf'))
    c -> ITunknown [C# c] : lexIface (stepOn buf)
***************
*** 708,709 ****
--- 707,709 ----
         ,("@_",                        ITatsign)
+        ,("letrec_",           ITletrec)
         ,("interface_",                ITinterface)
***************
*** 751,753 ****
         ,("let",                       ITlet)
-        ,("letrec",            ITletrec)
         ,("deriving",          ITderiving)
--- 751,752 ----

*** ptools/ghc/compiler/rename/ParseUnfolding.y-old   Thu Mar 13 16:55:56 1997
--- fptools/ghc/compiler/rename/ParseUnfolding.y      Mon Apr  7 12:30:35 1997
***************
*** 167,169 ****
                                                                }
!               | SCC OPAREN core_expr CPAREN   {  UfSCC $1 $3  }
  
--- 167,169 ----
                                                                }
!               | SCC core_expr                                 {  UfSCC $1 $2  }
  

GHC for Win32: linker reports unresolved symbol VirtualProtect

With the Win32 build of GHC, programs fail to link; the linker reporting the symbol VirtualProtect as undefined. The reason for this is that the beta17 release as shipped by Cygnus contains a version of libgcc.a that has an incorrect implementation of of mprotect() (or, more precisely, incorrectly compiled).

To get around this problem, you will have to pick up a patched version of libgcc.a from

  ftp://ftp.dcs.gla.ac.uk/pub/haskell/glasgow/working/

from the file libgcc.a.gz See ghc-2.02-cygwin32.README for quick instructions of how and where to install the archive.

Cannot use -syslib hbc under Linux

The original binary bundle for the Linux, ghc-2.02-i386-unknown-linux.tar.gz did not have the hbc libraries included. An updated binary bundle has been put up:

-rw-rw-r-- sof 2327819 Mar 28 21:25 /local/ftp/pub/haskell/glasgow/ghc-2.02-conc-i386-unknown-linux.tar.gz
-rw-rw-r-- sof 3991335 Mar 28 21:25 /local/ftp/pub/haskell/glasgow/ghc-2.02-i386-unknown-linux.tar.gz
-rw-rw-r-- sof 8361827 Mar 26 04:50 /local/ftp/pub/haskell/glasgow/ghc-2.02-prof-i386-unknown-linux.tar.gz

that includes the library plus interface files.

What? Perfectly Reasonable Haskell programs don't compile anymore

With the move to Haskell 1.3/1.4, programs written in Haskell 1.2 or earlier may very well fail to compile out of the box. The compiler documentation contains material on some of the incompatibilities between the language versions. Potentially useful.

Checksum errors when un-tar'ing the 2.02 distribution

Non-GNU tar programs may report checksum errors when unpacking the 2.02 binary bundles, due to overly long names in the archive. To get around this un-tar with the -i option added.

The entries that have too long names are very much on-the-fringe stuff, interface files for one or two of the hslibs/contrib/ numerical codes. If you're bothered, un-tar with the verbose option added and manually rename the interface files where tar trips up. (or, even better, get hold of GNU tar).

Help! I cannot find the function maybe

Perturbations in the Haskell 1.4 Prelude/Library caused the function maybe to slip through the net, and ended up not being exported by neither the (implicit) Prelude nor through the Maybe interface in 2.02. Sigh.

To get at maybe in 2.02, you'll have to say

import PrelBase (maybe)

Thanks to Alex Ferguson <abf@cs.ucc.ie> for reporting this.

The compiler cannot find a Glasgow extension like unboxed types

In 2.02, the Glasgow system libraries have again been re-arranged (we'll stop doing this some day..honest!) One outcome of all this, is that now all Glasgow extensions can be reached from the GlaExts interface.

Strictness annotations in data types aren't parsed correctly

Gruff in the parser causes definitions like the following

data A = B !Int Char Int

to be parsed incorrectly, the Char is parsed as a type constructor applied to Int. Fix is just-around the corner.

Using letrec as identifier does not work

Usages of letrec may seriously confuse the interface file parser, as it treats letrec as a keyword when parsing unfoldings. Avoid for now.

Installation fails during copying of library files

The Makefile supplied with the binary bundles (erroneously) assumes that cp supports the -f option, which may cause make install to fail. To fix, simply remove the -f option from the definition of CP in the Makefile.in and re-run the configure script.

Empty comment ending the file breaks the parser

If you should happen to write the following

module Main(main) where

main = print "Hello, world!" --

and don't add a newline after the --, the parser complains. Parser fix is in the pipeline, for the moment, the work-around is, Don't Do That Then.

GHC 0.29 bugs and workarounds

Using dupChan does not work(2.01 and 0.29)

 getChan :: Chan a -> IO a
 getChan (Chan read write)
  = takeMVar read            >>= \ rend ->
    -- WRONG: takeMVar rend  >>= \ res@(ChItem val new_rend) ->
    readMVar rend            >>= \ res@(ChItem val new_rend) ->
    putMVar read new_rend    >>
    return val

Foreign objects prematurely released

*** ghc/compiler/PrimKind.lhs-old	1996/05/15 17:35:06
--- ghc/compiler/PrimKind.lhs	1997/08/07 20:42:19
***************
*** 104,109 ****
  isFollowableKind ArrayKind  	= True
  isFollowableKind ByteArrayKind 	= True
! -- why is a MallocPtr followable? 4/96 SOF
! --isFollowableKind MallocPtrKind = True
  
  isFollowableKind StablePtrKind  = False
--- 104,113 ----
  isFollowableKind ArrayKind  	= True
  isFollowableKind ByteArrayKind 	= True
! -- why is a ForeignObj followable? 4/96 SOF
! --
! -- A: they're followable because these objects
! -- should be lugged around by the storage manager
! -- (==> we need to generate code that identify them as such) -- 3/97 SOF
! isFollowableKind ForeignObjKind = True
  
  isFollowableKind StablePtrKind  = False

Assembler trouble under IRIX

The output from GHC-0.26 under mips-sgi-irix is not liked by the latest versions of GCC and GAS (i.e., 2.7.2 and 2.6):

    % ghc -v hello.hs
    /usr/tmp/ghc10322.s:44: Error: Can not represent relocation in this object file format
    /usr/tmp/ghc10322.s:46: Error: Can not represent relocation in this object file format
    [..]

(or variations thereof). For now, try sticking with an older version of GAS (2.5.2 has been used with success)

HSC can't find getrusage under Solaris 2.4

Symptom:

Running GHC as it comes out of the binary distribution fails with the message

    ld.so.1: hsc: fatal: relocation error: symbol not found:
    getrusage: referenced in hsc

under Solaris 2.4.

Solution:

Preferably, upgrade to Solaris 2.5 (this problem is specific to Solaris 2.4).

Alternatively, create a dummy definition of getrusage and turn it into a shared object (for use with hsc) and add it to the runtime-system library. Jon Mountjoy gives a recipe for this glorious hack.

Problems with the GLASGOW_HASKELL_ROOT environment variable

Symptom:

After boldly hacking the ghc driver script and setting the GLASGOW_HASKELL_ROOT variable any compilation fails with the message

    GLASGOW_HASKELL_ROOT environment variable is set;
    But can't untangle /local/fp/lib/ghc/2.01/i386-unknown-linux.
    (Installation error)

Solution:

Don't do it! Set the variables

    $InstLibDirGhc  
    $InstDataDirGhc 

Calls to getCPUTime in ghc-2.01 return strange (and very unlikely) results (see bug report).

Solution:

Patched versions of the ghc library for a standard, profiling and concurrent setup are available from the Glasgow FTP server:

 ftp://www.dcs.gla.ac.uk/pub/haskell/glasgow/working/libHSghc.a-ghc-2.01-sparc-sun-solaris2
 ftp://www.dcs.gla.ac.uk/pub/haskell/glasgow/working/libHSghc_p.a-ghc-2.01-sparc-sun-solaris2
 ftp://www.dcs.gla.ac.uk/pub/haskell/glasgow/working/libHSghc_mc.a-ghc-2.01-sparc-sun-solaris2

Symptom:

When compiling a program that has an import PreludeGlaST with GHC 2.01 and profiling turned on (-prof) an error message like this appears:

ld:
/usr/haskell/ghc-2.01/lib/ghc/2.01/sparc-sun-sunos4/libHSrts_p.a(StgStartup.p_o):
__regPreludeGlaST: multiply defined

Solution:

Pick one of the following:

• Do a setenv LD_OPTIONS "-z muldefs" in your working environment.
• Replace import PreludeGlaST by import GHCbase.cvs cy

directly to the directories containing platform-dependent and platform-independent GHC support files, respectively. Using GLASGOW_HASKELL_ROOT is an officially Plain Bad Idea.

Profiling build `not available' on Solaris

Symptom:

When trying the pre-built profiling stuff for sparc-sun-solaris2, you may get:

   % ghc-0.29 -prof Main.hs
   ghc-0.29: a `profiling' "build" is not available with your GHC setup.
   (It was not configured for it at your site.)

Solution:

Simply edit your driver (ghc-0.29), around about line 207, to read:

%BuildAvail	= ('',	    '-build-normal-defined',
		   '_p',    '-build-p-defined', # NB: this line here
		   ...

Building 0.29 breaks on compiling PreludeIO_p.hc

Symptom:

Utter misery when building 0.29 a non-sequential (i.e., profiling, concurrent etc.) prelude from sources:

foo% pwd;make
/ghc-BUILDS/0.29/ghc/lib
[....]
../../ghc/driver/ghc -c   -hi-diffs -dcore-lint -link-chk  -iprelude:glaExts   
-dcore-lint -fshow-pragma-name-errs -fshow-import-specs -fomit-default-instance-methods       
-cpp -fglasgow-exts -genSPECS -DUSE_FOLDR_BUILD -D__OVERLAPPING_INSTANCES__
-H40m -prelude -fno-implicit-prelude -fomit-reexported-instances 
-split-objs PIO -ohi prelude/PreludeIO.hi -fno-omit-reexported-instances 
'-#include"stgio.h"' -hisuf _p.hi  -O  -prof -GPrelude prelude/PreludeIO_p.hc 
-odir prelude/PreludeIO -osuf    '_p.o'

make[1]: *** No rule to make target 
`/local/grasp_tmp3/hwloidl/sparc-sun-sunos4/./ghc/lib/../../ghc/includes/error.h',
 needed by `prelude/PreludeIOError_p.hc'.  Stop.
foo%

Solution:

The ghc/lib/Jmake.inc5 in the source distribution contains some links to local Glasgow files. To fix, do `make print_file_lists' and do make again (this will still add some hard-wired paths to the Jmake.inc? files, but at least they're yours now :-)

Empty arrays crashes compacting collector

Symptom:

Programs compiled under either the 1.2 compiler (version 0.xx) or the 1.3 compiler (v. 2.xx) that uses empty arrays (i.e., arrays with bounds (a,b) that define an empty range (a>b)) crash and burn with the (default) generational garbage collector, reporting "Segmentation fault" (or the like). Using the two space collector works fine - what gives?

Solution:

The problem is caused by the marking code in the garbage collector assumed that arrays had to be non-empty - which is not the case for arrays. Fix for 0.29 and 2.01 in their next releases - but if you've got the source installed locally, apply the following patch and recompile the RTS:

*** ghc/runtime/storage/SMmark.lhc      Mon Aug  5 06:14:38 1996
--- ../../../ghc-0.29/ghc/runtime/storage/SMmark.lhc    Tue Jul 23 17:32:21 1996
***************
*** 666,674 ****
          JUMP_MARK_RETURN;
      } else {
!         INIT_MARK_NODE("MUT ", MUTUPLE_CLOSURE_NoPTRS(Mark));
!       if (MUTUPLE_CLOSURE_NoPTRS(Mark) > 0)
!          INIT_MSTACK(MUTUPLE_CLOSURE_PTR);
!       else
!          JUMP_MARK;
      }
      FUNEND;
--- 666,671 ----
          JUMP_MARK_RETURN;
      } else {
!     INIT_MARK_NODE("MUT ", MUTUPLE_CLOSURE_NoPTRS(Mark));
!       INIT_MSTACK(MUTUPLE_CLOSURE_PTR);
      }
      FUNEND;

Building 0.29 hangs at qp2ap

Symptom:

On some platforms, when the STARTUP script gets to ghc/utils/parallel/, it hangs at...

   
   rm -f qp2ap.pl
   echo "#!"/bin/perl > qp2ap.pl
   ../../../glafp-utils/msub/msub  qp2ap.pl >> qp2ap.pl || ( rm -f qp2ap.pl && exit 1 )

Solution:

Drop the spurious first .pl in the qp2ap in the Jmakefile:

*** ghc-0.29/ghc/utils/parallel/Jmakefile.~1~
--- ghc-0.29/ghc/utils/parallel/Jmakefile
***************
*** 40,46 ****
  MsubProgramScriptTarget(PerlCmd,gr2java,gr2java.pl,,)
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2jv,gr2jv.bash,,)
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2ap,gr2ap.bash,,)
! MsubProgramScriptTarget(PerlCmd,qp2ap.pl,qp2ap.pl,,)
  
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2gran,gr2gran.bash,,)
  MsubProgramScriptTarget(PerlCmd,gr2RTS,gr2RTS.pl,,)
--- 40,46 ----
  MsubProgramScriptTarget(PerlCmd,gr2java,gr2java.pl,,)
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2jv,gr2jv.bash,,)
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2ap,gr2ap.bash,,)
! MsubProgramScriptTarget(PerlCmd,qp2ap,qp2ap.pl,,)
  
  MsubProgramScriptTarget(/usr/local/bin/bash,gr2gran,gr2gran.bash,,)
  MsubProgramScriptTarget(PerlCmd,gr2RTS,gr2RTS.pl,,)

>>>>>>> 1.41 <<<<<<< ghc-bugs.html Last modified: Wed Nov 6 00:15:15 1996 Stardate: [-31]8379.28 ======= Last modified: Fri 10 Oct 18:39 BST 1997 by sof@dcs.gla.ac.uk >>>>>>> 1.41