ghc-7.2.1-rc1 broke unregisterised build

[slyfox]

Tried to build unregisterised ghc:

"inplace/bin/ghc-stage1" -keep-tmp-files  -H32m -O -optc-mieee -opta-mieee -optc-mcpu=ev4 -opta-mcpu=ev4 -w    -package-name integer-gmp-0.3.0.0 -hide-all-packages -i -ilibraries/integer-gmp/. -ilibraries/integer-gmp/dist-install/build -ilibraries/integer-gmp/dist-install/build/autogen -Ilibraries/integer-gmp/dist-install/build -Ilibraries/integer-gmp/dist-install/build/autogen -Ilibraries/integer-gmp/.    -optP-include -optPlibraries/integer-gmp/dist-install/build/autogen/cabal_macros.h -package ghc-prim-0.2.0.0  -package-name integer-gmp -XHaskell98 -XCPP -XMagicHash -XUnboxedTuples -XNoImplicitPrelude -XForeignFunctionInterface -XUnliftedFFITypes -O2 -no-user-package-conf -rtsopts     -odir libraries/integer-gmp/dist-install/build -hidir libraries/integer-gmp/dist-install/build -stubdir libraries/integer-gmp/dist-install/build -hisuf hi -osuf  o -hcsuf hc -c libraries/integer-gmp/./GHC/Integer/Type.hs -o libraries/integer-gmp/dist-install/build/GHC/Integer/Type.o

/var/tmp/portage/dev-lang/ghc-7.2.0.20110728/temp/ghc11901_0/ghc11901_0.hc:33:0:
     error: static declaration of ‘integerzmgmp_GHCziIntegerziType_Szh_info’ follows non-static declaration

/var/tmp/portage/dev-lang/ghc-7.2.0.20110728/temp/ghc11901_0/ghc11901_0.hc:6:0:
     error: previous declaration of ‘integerzmgmp_GHCziIntegerziType_Szh_info’ was here

/var/tmp/portage/dev-lang/ghc-7.2.0.20110728/temp/ghc11901_0/ghc11901_0.hc:64:0:
     error: static declaration of ‘integerzmgmp_GHCziIntegerziType_Jzh_info’ follows non-static declaration

/var/tmp/portage/dev-lang/ghc-7.2.0.20110728/temp/ghc11901_0/ghc11901_0.hc:36:0:
     error: previous declaration of ‘integerzmgmp_GHCziIntegerziType_Jzh_info’ was here
make[1]: *** [libraries/integer-gmp/dist-install/build/GHC/Integer/Type.o] Error 1

And yes, the .hs file contains incoherent declarations:

EI_(integerzmgmp_GHCziIntegerziType_Szh_info);
StgWord integerzmgmp_GHCziIntegerziType_Szh_closure[] = {
(W_)&integerzmgmp_GHCziIntegerziType_Szh_info
};
...
static StgWord integerzmgmp_GHCziIntegerziType_Szh_info[] = {
(W_)&integerzmgmp_GHCziIntegerziType_Szh_entry, 0x0, 0xfUL, 0x100000004UL
};

The architecture is alpha-linux, but I think it's a generic problem. Looks like it's occured after Max's (CCed) _info export change:

commit dd391759500ab2b6abd23d502ade7ff8946c780f
Author: Max Bolingbroke <batterseapower@hotmail.com>
Date:   Wed Jul 6 18:01:02 2011 +0100

    Don't export the _info symbol for the data constructor worker bindings

    This is safe because GHC never generates a fast call to a data constructor
    worker: if the call is seen statically it will be eta-expanded and the
    allocation of the data will be inlined. We still need to export the _closure
    in case the constructor is used in an unapplied fashion.

Thanks!

[igloo]

Thanks for the report

[batterseapower]

I couldn't reproduce this with HEAD, but I *could* reproduce with 834dbd9ac41f8e40b31d9d2045765d03fc210d50. I think that my recent series of "refactoring" patches inadvertently fixed this problem:

81c6183dca435a0f03ec3342f8c116d5f9de2ea6..06dc1813f5cfc42010925ad7d36dff1791d846ae

If you cherry-pick these onto stable it should fix the problem.

[igloo]

These patches don't merge cleanly. Do you know exactly which change fixed it?

[batterseapower]

I don't know which one fixed it, and I wouldn't like to try to guess since the patches are interdependent. Instead I've merged them onto ghc-7.2 myself (fixing the single conflict) - is that OK?

[igloo]

OK, thanks!

[dterei]

Reopening as the unregisterised build is pretty broken in HEAD.

First error I get is:

/tmp/ghc30253_0/ghc30253_0.hc:1004:2:
     error: ‘free’ redeclared as different kind of symbol

/tmp/ghc30253_0/ghc30253_0.hc:473:1:
     note: previous declaration of ‘free’ was here

C code in question is:

EI_(free);
StgWord base_ForeignziMarshalziAlloc_finalizzerFree_closure[] = {
(W_)&base_GHCziPtr_FunPtr_static_info,
(W_)&free
};

// [ ... snip ...]

FN_(base_ForeignziMarshalziAlloc_zdwccall_entry) {                                                                                                                                                            
W_ _c1O6;
FB_
_c1O6 = *Sp;
Sp=Sp+1;
;EF_(free);
{void (*ghcFunPtr)(void *);
ghcFunPtr = ((void (*)(void *))(W_)&free);
ghcFunPtr((void *)_c1O6);;}
JMP_(*((P_)(*Sp)));
FE_
}

This error seems to be caused by the fact that free is imported twice in Foreign/Marshall/Alloc

foreign import ccall unsafe "stdlib.h free"    _free    :: Ptr a -> IO ()

-- | A pointer to a foreign function equivalent to 'free', which may be
-- used as a finalizer (cf 'Foreign.ForeignPtr.ForeignPtr') for storage
-- allocated with 'malloc', 'mallocBytes', 'realloc' or 'reallocBytes'.
foreign import ccall unsafe "stdlib.h &free" finalizerFree :: FinalizerPtr a 

After coming up with a hacky way to fix that error I get this second error:

/tmp/ghc9007_0/ghc9007_0.hc:16360:1:
     error: static declaration of ‘bytestringzm0zi9zi2zi0_DataziByteStringziChar8_zdwlines_entry’ follows non-static declaration

/tmp/ghc9007_0/ghc9007_0.hc:16314:1:
     note: previous declaration of ‘bytestringzm0zi9zi2zi0_DataziByteStringziChar8_zdwlines_entry’ was here

C code in question for that error is:

IF_(bytestringzm0zi9zi2zi0_DataziByteStringziChar8_zdwlines_entry); 
FN_(bytestringzm0zi9zi2zi0_DataziByteStringziChar8_zdwlines_slow) {
FB_Sp=Sp+0;
JMP_((W_)&bytestringzm0zi9zi2zi0_DataziByteStringziChar8_zdwlines_entry);
FE_
}

// [... snip ...]

FN_(bytestringzm0zi9zi2zi0_DataziByteStringziChar8_zdwlines_entry) {W_ _c6uU;W_ _c6uV;                                                                                                                                                                                                     W_ _s4XO;

As well as getting this bug fixed is it possible Ian to get an unregisterized build bot running? Every time I try using the unregisterized build it seems to have bitrotted since no one really monitors it.

[dterei]

Building GHC in unregisterised with the LLVM backend still works fine at least.

[simonmar]

I have a fix for the errors about free, validating now. I'm also running an unreg build so I'll see if anything else pops up.

[simonmar]

commit 0545b5500c19b7266046447c1352e936a9ee7c99

Author: Simon Marlow <marlowsd@gmail.com>
Date:   Wed Jan 4 13:29:14 2012 +0000

    dsCImport (CLabel): drop foralls when checking for FunPtr
    
    We were mislabelling some foreign imports as IsData rather than
    IsFunction, because the type was forall a. FunPtr (... a ...).
    
    Eventually this gave rise to these obscure error when compiling
    unregisterised:
    
    libraries/base/Foreign/Marshal/Alloc.hc:473:0:
         warning: built-in function ‘free’ declared as non-function
    
    libraries/base/Foreign/Marshal/Alloc.hc:1004:0:
         error: ‘free’ redeclared as different kind of symbol

[marlowsd@…]

commit 4464c92badaedc45ce53d6349f6790f6d2298103

Author: Simon Marlow <marlowsd@gmail.com>
Date:   Wed Jan 4 13:59:06 2012 +0000

    mkSlowEntryCode: enterLocalIdLabel should be enterIdLabel (#5357)

 compiler/codeGen/CgClosure.lhs |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

[marlowsd@…]

commit 974f45103b930ed4310f9ec67b20399e3f289adf

Author: Simon Marlow <marlowsd@gmail.com>
Date:   Wed Jan 4 15:40:41 2012 +0000

    Fixup to 4464c92badaedc45ce53d6349f6790f6d2298103
    
    Instead of enterLocalIdLabel we should get the label from the
    ClosureInfo, because that knows better whether the label should be
    local or not.
    
    Needed by #5357

 compiler/codeGen/CgClosure.lhs   |    2 +-
 compiler/codeGen/ClosureInfo.lhs |    5 ++++-
 2 files changed, 5 insertions(+), 2 deletions(-)

[marlowsd@…]

commit dc179f54d895dc560be857a89c991bd11cac100f

Author: Simon Marlow <marlowsd@gmail.com>
Date:   Wed Jan 4 15:23:27 2012 +0000

    Give the correct type to CCCS
    
    Needed by #5357

 includes/stg/Regs.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

[marlowsd@…]

commit 53fad2ab6cd5cd39717e74231d49444f1759fcd1

Author: Simon Marlow <marlowsd@gmail.com>
Date:   Wed Jan 4 15:23:15 2012 +0000

    Rename struct _CostCentreStack to struct CostCentreStack_ for consistency
    
    Needed by #5357

 includes/rts/prof/CCS.h |   16 ++++++++--------
 includes/stg/Regs.h     |    2 +-
 2 files changed, 9 insertions(+), 9 deletions(-)

[marlowsd@…]

commit e04da8a48ea5e8823b09f536330d6c7d796a93b5

Author: Simon Marlow <marlowsd@gmail.com>
Date:   Wed Jan 4 15:14:31 2012 +0000

    Fix for unregisterised compilation: we want registers when defined(USE_MINIINTERPRETER)
    
    Needed by #5357

 rts/StgCRun.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

[simonmar]

commit 3478213ac422912df85308029e16c3f31b5a91d2

Author: Simon Marlow <marlowsd@gmail.com>
Date:   Wed Jan 4 15:13:23 2012 +0000

    Rename the CCCS field of StgTSO so as not to conflict with the CCCS pseudo-register
    
    Needed by #5357

[simonmar]

commit 2d91e04b9cedac7bdb772defc7bd23a114c4dbdb

Author: Simon Marlow <marlowsd@gmail.com>
Date:   Wed Jan 4 15:13:56 2012 +0000

    Fix for unregisterised compilation: isStrangeTypeGlobal CCCS = True
    
    Needed by #5357

[simonmar]

I think I'm done here. Let's see if the overnight unregisterised build works before closing the ticket though. In the meantime all these patches need to be merged.

[simonmar]

Looks like the overnight build worked, modulo a few test failures of which I've fixed the serious ones.

[igloo]

Merged as 5cbd9a2f9727017929eef1b7ade1253e1566e6dd, 3a6fe38c9a986c03a7946647dc52f5f87bec9921, ff8eaac6f7e18aae257efcf70ecb324fd00ce308, cf207c538527b69e6dab6ad3111e085e2476da1a, edbeeeadf3ab44c4190797512ab2f60650f2accf, 8a9d80eb2816b460c8d1cffd1d36e978e9d14d7c, 647f8a20e66fc3d11d7bdd8e1eaa3a9c25ab31d9, 0ea1e77a8d404ac09fea740a43da9ae880ab8187 respectively.