- ghc-devs - Haskell.org

Link error with head Re: [commit: ghc] master: GHCi: Properly generate jump code for ARM (#8380) (5bab1a5)
by Amos Robinson 26 Nov '13

26 Nov '13

Hi, I'm having trouble linking with head: > [1 of 1] Compiling Main ( bindisttest/HelloWorld.lhs, bindisttest/HelloWorld.o ) > Linking bindisttest/HelloWorld ... > Undefined symbols for architecture x86_64: > "___builtin___clear_cache", referenced from: > _flushExec in libHSrts.a(Storage.o) > ld: symbol(s) not found for architecture x86_64 > collect2: ld returned 1 exit status This is on OSX 10.8.5, where gcc --version gives me > i686-apple-darwin11-llvm-gcc-4.2 (GCC) 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2336.11.00) I get the impression that this commit is related. Is it the case that Apple switched from gcc to clang recently? Could that be why ___builtin__clear_cache is missing? I'm sorry, I don't really know anything about the runtime system, so am kind of stuck. On Sat, Nov 23, 2013 at 1:22 AM, <git(a)git.haskell.org> wrote: > Repository : ssh://git@git.haskell.org/ghc > > On branch : master > Link : http://ghc.haskell.org/trac/ghc/changeset/5bab1a57f572e29dfdffd6d1ce8e53a27… > >>--------------------------------------------------------------- > > commit 5bab1a57f572e29dfdffd6d1ce8e53a2772b18fd > Author: Austin Seipp <austin(a)well-typed.com> > Date: Mon Nov 11 10:26:03 2013 -0600 > > GHCi: Properly generate jump code for ARM (#8380) > > This adds code for jumping to given addresses for ARM, written by Ben > Gamari. > > However, when allocating new infotables for bytecode (which is where > this jump code occurs), we need to be sure to flush the cache on the > execute pointer returned from allocateExec() - on systems like ARM, the > processor won't reliably read back code or automatically cache flush, > where x86 will. > > So we add a new flushExec primitive to call out to GCC's > __builtin___clear_cache primitive, which will properly generate the > correct code (nothing on x86, and a call to libgcc's __clear_cache on > ARM) and make sure we use it after writing the code out. > > Authored-by: Ben Gamari <bgamari.foss(a)gmail.com> > Authored-by: Austin Seipp <austin(a)well-typed.com> > Signed-off-by: Austin Seipp <austin(a)well-typed.com> > > >>--------------------------------------------------------------- > > 5bab1a57f572e29dfdffd6d1ce8e53a2772b18fd > compiler/ghci/ByteCodeItbls.lhs | 21 ++++++++++++++++++++- > includes/rts/storage/GC.h | 1 + > rts/Linker.c | 1 + > rts/sm/Storage.c | 27 +++++++++++++++++++++++++++ > 4 files changed, 49 insertions(+), 1 deletion(-) > > diff --git a/compiler/ghci/ByteCodeItbls.lhs b/compiler/ghci/ByteCodeItbls.lhs > index 0d07be5..2180f87 100644 > --- a/compiler/ghci/ByteCodeItbls.lhs > +++ b/compiler/ghci/ByteCodeItbls.lhs > @@ -227,6 +227,20 @@ mkJumpToAddr dflags a = case platformArch (targetPlatform dflags) of > , fromIntegral (w64 .&. 0x0000FFFF) > , fromIntegral ((w64 `shiftR` 32) .&. 0x0000FFFF) ] > > + ArchARM { } -> > + -- Generates Thumb sequence, > + -- ldr r1, [pc, #0] > + -- bx r1 > + -- > + -- which looks like: > + -- 00000000 <.addr-0x8>: > + -- 0: 4900 ldr r1, [pc] ; 8 <.addr> > + -- 4: 4708 bx r1 > + let w32 = fromIntegral (ptrToInt a) :: Word32 > + in Left [ 0x49, 0x00 > + , 0x47, 0x08 > + , byte0 w32, byte1 w32, byte2 w32, byte3 w32] > + > arch -> > panic ("mkJumpToAddr not defined for " ++ show arch) > > @@ -374,11 +388,16 @@ load = do addr <- advance > newExecConItbl :: DynFlags -> StgConInfoTable -> IO (FunPtr ()) > newExecConItbl dflags obj > = alloca $ \pcode -> do > - wr_ptr <- _allocateExec (fromIntegral (sizeOfConItbl dflags obj)) pcode > + let sz = fromIntegral (sizeOfConItbl dflags obj) > + wr_ptr <- _allocateExec sz pcode > ex_ptr <- peek pcode > pokeConItbl dflags wr_ptr ex_ptr obj > + _flushExec sz ex_ptr -- Cache flush (if needed) > return (castPtrToFunPtr ex_ptr) > > foreign import ccall unsafe "allocateExec" > _allocateExec :: CUInt -> Ptr (Ptr a) -> IO (Ptr a) > + > +foreign import ccall unsafe "flushExec" > + _flushExec :: CUInt -> Ptr a -> IO () > \end{code} > diff --git a/includes/rts/storage/GC.h b/includes/rts/storage/GC.h > index 8133496..f8b8afe 100644 > --- a/includes/rts/storage/GC.h > +++ b/includes/rts/storage/GC.h > @@ -161,6 +161,7 @@ typedef void* AdjustorWritable; > typedef void* AdjustorExecutable; > > AdjustorWritable allocateExec(W_ len, AdjustorExecutable *exec_addr); > +void flushExec(W_ len, AdjustorExecutable exec_addr); > #if defined(ios_HOST_OS) > AdjustorWritable execToWritable(AdjustorExecutable exec); > #endif > diff --git a/rts/Linker.c b/rts/Linker.c > index 77943a5..14ebac3 100644 > --- a/rts/Linker.c > +++ b/rts/Linker.c > @@ -1350,6 +1350,7 @@ typedef struct _RtsSymbolVal { > SymI_HasProto(g0) \ > SymI_HasProto(allocate) \ > SymI_HasProto(allocateExec) \ > + SymI_HasProto(flushExec) \ > SymI_HasProto(freeExec) \ > SymI_HasProto(getAllocations) \ > SymI_HasProto(revertCAFs) \ > diff --git a/rts/sm/Storage.c b/rts/sm/Storage.c > index 112ad83..c1a1a5a 100644 > --- a/rts/sm/Storage.c > +++ b/rts/sm/Storage.c > @@ -1152,6 +1152,15 @@ AdjustorWritable allocateExec (W_ bytes, AdjustorExecutable *exec_ret) > return (ret + 1); > } > > +void flushExec (W_ len, AdjustorExecutable exec_addr) > +{ > + /* On ARM and other platforms, we need to flush the cache after > + writing code into memory, so the processor reliably sees it. */ > + unsigned char* begin = (unsigned char*)exec_addr; > + unsigned char* end = begin + len; > + __builtin___clear_cache(begin, end); > +} > + > // freeExec gets passed the executable address, not the writable address. > void freeExec (AdjustorExecutable addr) > { > @@ -1198,6 +1207,15 @@ AdjustorWritable execToWritable(AdjustorExecutable exec) > return writ; > } > > +void flushExec (W_ len, AdjustorExecutable exec_addr) > +{ > + /* On ARM and other platforms, we need to flush the cache after > + writing code into memory, so the processor reliably sees it. */ > + unsigned char* begin = (unsigned char*)exec_addr; > + unsigned char* end = begin + len; > + __builtin___clear_cache(begin, end); > +} > + > void freeExec(AdjustorExecutable exec) > { > AdjustorWritable writ; > @@ -1251,6 +1269,15 @@ AdjustorWritable allocateExec (W_ bytes, AdjustorExecutable *exec_ret) > return ret; > } > > +void flushExec (W_ len, AdjustorExecutable exec_addr) > +{ > + /* On ARM and other platforms, we need to flush the cache after > + writing code into memory, so the processor reliably sees it. */ > + unsigned char* begin = (unsigned char*)exec_addr; > + unsigned char* end = begin + len; > + __builtin___clear_cache(begin, end); > +} > + > void freeExec (void *addr) > { > StgPtr p = (StgPtr)addr - 1; > > _______________________________________________ > ghc-commits mailing list > ghc-commits(a)haskell.org > http://www.haskell.org/mailman/listinfo/ghc-commits

2 6

Cannot build tagged with HEAD
by Edsko de Vries 25 Nov '13

25 Nov '13

I'm getting the errors below if I try to build tagged with head: src/Data/Tagged.hs:57:5: 'Tagged' is applied to too many type arguments In an expression type signature: Tagged k s b -> Tagged k s b -> Bool In the expression: ghc-prim:GHC.Prim.coerce ((==) :: b -> b -> Bool) :: Tagged k s b -> Tagged k s b -> Bool In an equation for '==': (==) = ghc-prim:GHC.Prim.coerce ((==) :: b -> b -> Bool) :: Tagged k s b -> Tagged k s b -> Bool src/Data/Tagged.hs:57:9: 'Tagged' is applied to too many type arguments In an expression type signature: Tagged k s b -> Tagged k s b -> Ordering In the expression: ghc-prim:GHC.Prim.coerce (compare :: b -> b -> Ordering) :: Tagged k s b -> Tagged k s b -> Ordering In an equation for 'compare': compare = ghc-prim:GHC.Prim.coerce (compare :: b -> b -> Ordering) :: Tagged k s b -> Tagged k s b -> Ordering src/Data/Tagged.hs:57:14: 'Tagged' is applied to too many type arguments In an expression type signature: (,) (Tagged k s b) (Tagged k s b) -> [] (Tagged k s b) In the expression: ghc-prim:GHC.Prim.coerce (range :: (,) b b -> [] b) :: (,) (Tagged k s b) (Tagged k s b) -> [] (Tagged k s b) In an equation for 'range': range = ghc-prim:GHC.Prim.coerce (range :: (,) b b -> [] b) :: (,) (Tagged k s b) (Tagged k s b) -> [] (Tagged k s b) src/Data/Tagged.hs:57:18: 'Tagged' is applied to too many type arguments In an expression type signature: Tagged k s b In the expression: ghc-prim:GHC.Prim.coerce (minBound :: b) :: Tagged k s b In an equation for 'minBound': minBound = ghc-prim:GHC.Prim.coerce (minBound :: b) :: Tagged k s b Line 57 is the "(" in newtype Tagged s b = Tagged { unTagged :: b } deriving ( Eq, Ord, Ix, Bounded #if __GLASGOW_HASKELL__ >= 707 , Typeable #endif ) Anyone any ideas what's going on? -E

3 3

RE: bundle panic
by Simon Peyton-Jones 22 Nov '13

22 Nov '13

I've looked into this myself. I don't think it's actually anything wrong in gmp-wrappers. It's because CmmBuildInfoTables expects all the info tables in the (CmmProc info_tbls lbl g) to be defined in the graph g. But Jan's new optimisation to the stack overflow check (Note [Always false stack check] in CmmLayoutStack) ends up dropping a conditional branch, so a call to gc is eliminated, and hence a proc point is eliminated, and we end up with an unused thing in info_tbls, which confuses CmmBuildInfoTables Sigh. I don't know why this only happens on Windows. I'm looking into a fix Simon From: Simon Peyton-Jones Sent: 15 November 2013 22:18 To: ghc-devs(a)haskell.org Subject: bundle panic Windows build is failing again in a new way. It was fine a couple of days ago. Does this ring any bells? ghc-stage1.exe: panic! (the 'impossible' happened) (GHC version 7.7.20131107 for i386-unknown-mingw32): bundle Simon "inplace/bin/ghc-stage1.exe" -static -H32m -O -Werror -Wall -H64m -O0 -package-name integer-gmp-0.5.1.0 -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-0.3.1.0 -Wall -package-name integer-gmp -XHaskell2010 -dcore-lint -no-user-package-db -rtsopts -Ilibraries/integer-gmp/mkGmpDerivedConstants/dist -c libraries/integer-gmp/cbits/gmp-wrappers.cmm -o libraries/integer-gmp/dist-install/build/cbits/gmp-wrappers.o ghc-stage1.exe: panic! (the 'impossible' happened) (GHC version 7.7.20131107 for i386-unknown-mingw32): bundle c8 integer_cmm_importIntegerFromByteArrayzh [(c8, label: block{v c8}_info rep:StackRep [False, True, True, True]), (cg, label: block{v cg}_info rep:StackRep [False, True, True, True])] Please report this as a GHC bug: http://www.haskell.org/ghc/reportabug HEAD (master)$

3 3

Travis now also tests with -DDEBUG
by Joachim Breitner 21 Nov '13

21 Nov '13

Hi, after I got confused by panics in the testsuite that were not caused by my changes I configured travis to build GHC (stage2) with and without -DDEBUG. The current result can be viewed here: https://travis-ci.org/nomeata/ghc-complete/builds/14332594 You see that it works without -DDEBUG (this is also what vanilla validate does), but fails with (due to http://ghc.haskell.org/trac/ghc/ticket/8553) I hope this will make such issues known sooner. Again: I believe someone is working or planning on a proper infrastructure to do GHC QA, this is a crutch until then. Greetings, Joachim -- Joachim “nomeata” Breitner mail(a)joachim-breitner.de • http://www.joachim-breitner.de/ Jabber: nomeata(a)joachim-breitner.de • GPG-Key: 0x4743206C Debian Developer: nomeata(a)debian.org

1 0

RE: Too many symbols
by Simon Peyton-Jones 21 Nov '13

21 Nov '13

Right GhcLibWays=v did the job, thanks Simon | -----Original Message----- | From: Austin Seipp [mailto:aseipp@pobox.com] | Sent: 21 November 2013 12:13 | To: Simon Peyton-Jones | Cc: ghc-devs(a)haskell.org Devs | Subject: Re: Too many symbols | | Hi Simon, | | I believe I've found the problem. The issue is that even with | DYNAMIC_* variables set, GhcLibWays still contains the 'dyn' way as | opposed to just the 'vanilla' way. If you do: | | $ make show VALUE=GhcLibWays | | in your build tree, it will probably say "v dyn", indicating it'll | build the GHC DLL as well as vanilla objects. This makes sense since | we rebuild using different ways in a lot of cases - dynamic-too is | just a special case, and the other variables (DYNAMIC_TOO, | DYNAMIC_GHC_PROGRAMS and DYNAMIC_BY_DEFAULT) don't make a difference | here. | | In the short term, you can hardcode GhcLibWays=v in your build | settings, but the default templates should perhaps expose something | more sensible. (They should also precisely document the impact of | these variables too.) | | On Wed, Nov 20, 2013 at 3:11 PM, Simon Peyton-Jones | <simonpj(a)microsoft.com> wrote: | > Friends | > | > I’m getting the dreaded “Too many symbols in DLL” problem on Windows | (see | > below). | > | > I believe I can get round it by using static linking only, yes? But | how do | > I do that? I have | > | > DYNAMIC_BY_DEFAULT = NO | > | > DYNAMIC_GHC_PROGRAMS = NO | > | > in my validate.mk, but it still happens. Static only will do me | fine! | > | > Thanks! | > | > Simon | > | > | > | > Number of symbols in | > compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll: | 69595 | > | > case `C:/code/HEAD/inplace/mingw/bin/objdump.exe -p | > compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll | | sed -n | > "1,/^.Ordinal\/Name Pointer/ D; p; /^$/ q" | grep "\[ *0\]" | wc -l` | in 1) | > echo DLL compiler/stage2/build/libHSghc-7.7.20131119-0- | ghc7.7.20131119.dll | > OK;; 0) echo No symbols in DLL | > compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll; | exit 1;; | > [0-9]*) echo Too many symbols in DLL | > compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll; | > C:/code/HEAD/inplace/mingw/bin/objdump.exe -p | > compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll | | sed -n | > "1,/^.Ordinal\/Name Pointer/ D; p; /^$/ q" | tail; exit 1;; *) echo | bad DLL | > compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll; | exit 1;; | > esac | > | > Too many symbols in DLL | > compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll | > | > [4050] ghczm7zi7zi20131119_Var_zdfUniquableVar_closure | > | > [4051] ghczm7zi7zi20131119_Var_zdfUniquableVar_info | > | > [4052] ghczm7zi7zi20131119_Var_zdwsetVarUnique_closure | > | > [4053] ghczm7zi7zi20131119_Var_zdwsetVarUnique_info | > | > [4054] ghczm7zi7zi20131119_Var_zdwupdateTyVarKindM_closure | > | > [4055] ghczm7zi7zi20131119_Var_zdwupdateTyVarKindM_info | > | > [4056] ghczm7zi7zi20131119_Var_zdwzdcgmapMp_closure | > | > [4057] ghczm7zi7zi20131119_Var_zdwzdcgmapMp_info | > | > [4058] setHeapSize | > | > | > | > compiler/ghc.mk:472: recipe for target | > 'compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll' | failed | > | > make[1]: *** | > [compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll] | Error 1 | > | > make[1]: *** Deleting file | > 'compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll' | > | > Makefile:64: recipe for target 'all' failed | > | > make: *** [all] Error 2 | > | > HEAD $ | > | > | > _______________________________________________ | > ghc-devs mailing list | > ghc-devs(a)haskell.org | > http://www.haskell.org/mailman/listinfo/ghc-devs | > | | | | -- | Regards, | Austin - PGP: 4096R/0x91384671

1 0

Too many symbols
by Simon Peyton-Jones 21 Nov '13

21 Nov '13

Friends I'm getting the dreaded "Too many symbols in DLL" problem on Windows (see below). I believe I can get round it by using static linking only, yes? But how do I do that? I have DYNAMIC_BY_DEFAULT = NO DYNAMIC_GHC_PROGRAMS = NO in my validate.mk, but it still happens. Static only will do me fine! Thanks! Simon Number of symbols in compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll: 69595 case `C:/code/HEAD/inplace/mingw/bin/objdump.exe -p compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll | sed -n "1,/^.Ordinal\/Name Pointer/ D; p; /^$/ q" | grep "\[ *0\]" | wc -l` in 1) echo DLL compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll OK;; 0) echo No symbols in DLL compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll; exit 1;; [0-9]*) echo Too many symbols in DLL compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll; C:/code/HEAD/inplace/mingw/bin/objdump.exe -p compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll | sed -n "1,/^.Ordinal\/Name Pointer/ D; p; /^$/ q" | tail; exit 1;; *) echo bad DLL compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll; exit 1;; esac Too many symbols in DLL compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll [4050] ghczm7zi7zi20131119_Var_zdfUniquableVar_closure [4051] ghczm7zi7zi20131119_Var_zdfUniquableVar_info [4052] ghczm7zi7zi20131119_Var_zdwsetVarUnique_closure [4053] ghczm7zi7zi20131119_Var_zdwsetVarUnique_info [4054] ghczm7zi7zi20131119_Var_zdwupdateTyVarKindM_closure [4055] ghczm7zi7zi20131119_Var_zdwupdateTyVarKindM_info [4056] ghczm7zi7zi20131119_Var_zdwzdcgmapMp_closure [4057] ghczm7zi7zi20131119_Var_zdwzdcgmapMp_info [4058] setHeapSize compiler/ghc.mk:472: recipe for target 'compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll' failed make[1]: *** [compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll] Error 1 make[1]: *** Deleting file 'compiler/stage2/build/libHSghc-7.7.20131119-0-ghc7.7.20131119.dll' Makefile:64: recipe for target 'all' failed make: *** [all] Error 2 HEAD $

2 1

[PATCH] Fix typo in User's Guide
by Pedro Rodrigues 21 Nov '13

21 Nov '13

--- docs/users_guide/glasgow_exts.xml | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/docs/users_guide/glasgow_exts.xml b/docs/users_guide/glasgow_exts.xml index ce07806..d61ff7a 100644 --- a/docs/users_guide/glasgow_exts.xml +++ b/docs/users_guide/glasgow_exts.xml @@ -8673,7 +8673,7 @@ proc (x,y) -> which is translated to <screen> arr (\ (x,y) -> if f x y then Left x else Right y) >>> - (arr (\x -> x+1) >>> f) ||| (arr (\y -> y+2) >>> g) + (arr (\x -> x+1) >>> g) ||| (arr (\y -> y+2) >>> h) </screen> Since the translation uses <function>|||</function>, the arrow concerned must belong to the <literal>ArrowChoice</literal> class. -- 1.7.4.1

5 6

Re: [commit: testsuite] master: Much better perf on 5030, higher peak on 3064 (53b3199)
by Simon Marlow 21 Nov '13

21 Nov '13

On 12/11/13 15:07, git(a)git.haskell.org wrote: > Repository : ssh://git@git.haskell.org/testsuite > > On branch : master > Link : http://ghc.haskell.org/trac/ghc/changeset/53b3199b03bd2e52c63babffd33d47824… > >> --------------------------------------------------------------- > > commit 53b3199b03bd2e52c63babffd33d478242ae74d5 > Author: Simon Peyton Jones <simonpj(a)microsoft.com> > Date: Tue Nov 12 15:06:56 2013 +0000 > > Much better perf on 5030, higher peak on 3064 > > The peak-megabytes is vulnerable to when GC runs Juat a reminder to everyone: when peak_megabytes or max_bytes_used changes, please run with +RTS -h -i0.01 to get a heap profile and check whether it's a real change or not. You don't need to compile for profiling or anything. Cheers, Simon

2 1

Behaviour of readProcess, and code simplicity (or lack thereof)
by Duncan Coutts 21 Nov '13

21 Nov '13

Hi possibly-interested souls, In this patch to System.Process.readProcess from last year: https://ghc.haskell.org/trac/ghc/changeset/b5ee908863882d18e4110d96b43ccc1b… Bas changed things so that if an async exception is received by the thread calling readProcess (or readProcessWithExitCode) then the external process gets forcibly terminated. I do mean forcibly, it's TerminateProcess() on Windows and SIGKILL (not SIGTERM) on Unix. I was wondering if this is really desirable behaviour. There are certainly programs out there that will leave stale lock files (or worse) if forcibly terminated, that can shut down reasonably cleanly if asked more nicely. I have to say, it does make me a bit nervous. That said, terminating programs cleanly is a tricky business. While there is SIGTERM on Unix, there's no simple way to do it on Windows; there's no softer equivalent of TerminateProcess(). But if we think it *is* the right behaviour then I think we ought to be doing it consistently. There are other System.Process functions that run processes synchronously that should have the same behaviour. And it turns out that cleaning up is rather subtle. Takano recently sent in a patch to fix a deadlock due to the order in which the cleanup happens: https://ghc.haskell.org/trac/ghc/changeset/32223a9ab174c22e939c81e24b6f4822… So if it is desirable behaviour and is so subtle we ought to provide the users with some help to do it (or indeed to do alternatives if we think no one policy fits all). For example, we could provide: -- | A 'bracketOnError'-style resource handler for 'createProcess'. -- -- In normal operation it adds nothing, you are still responsible for -- waiting for (or forcing) process termination and closing any -- 'Handle's. It only does automatic cleanup if there is an exception. -- If there is an exception in the body then it ensures that the process -- gets terminated and any 'CreatePipe' 'Handle's are closed. In -- particular this means that if the Haskell thread is killed (e.g. -- 'killThread'), that the external process is also terminated. -- withCreateProcess :: CreateProcess -> (Maybe Handle -> Maybe Handle -> Maybe Handle -> ProcessHandle -> IO a) -> IO a withCreateProcess c action = bracketOnError (createProcess c) cleanupProcess (\(m_in, m_out, m_err, ph) -> action m_in m_out m_err ph) cleanupProcess :: (Maybe Handle, Maybe Handle, Maybe Handle, ProcessHandle) -> IO ExitCode cleanupProcess (mb_stdin, mb_stdout, mb_stderr, ph) = do -- We rely on the fact that terminateProcess really does terminate it -- and not just ask it nicely. Otherwise we could deadlock if there -- are other threads holding the lock on the stdin/out/err Handles. terminateProcess ph maybe (return ()) hClose mb_stdin maybe (return ()) hClose mb_stdout maybe (return ()) hClose mb_stderr waitForProcess ph As it is, the docs say: 'readProcess' and 'readProcessWithExitCode' are fairly simple wrappers around 'createProcess'. Constructing variants of these functions is quite easy: follow the link to the source code to see how 'readProcess' is implemented. but if you do look at the source you see that while it may originally have been true, it certainly is not anymore! Providing and using a withCreateProcess-style would help quite a bit. Also, the ignoring EPIPE stuff is rather unsatisfactory and non-portable. It's not used consistently and it's impossible for anyone else to implement it themselves without using GHC internals. It's almost possible in a nice way, since you could check if it's ResourceVanished rather than errno = EPIPE, but actually ResourceVanished is not exposed via System.IO.Error either. I think it's worth if at all possible making it true that people can write useful and reliable special cases of createProcess, but we'd need to provide more help.

3 2

clang/wrapper cannot build GHC head with integer-gmp on Mavericks
by Kazu Yamamoto 21 Nov '13

21 Nov '13

Hi, I'm try to make GHC head stably buildable with clang/wrapper on Mavericks because we cannot ask all Mavericks users to install apple-gcc or GCC. The last big item for me is integer-gmp. GHC head cannot be built with clang/wrapper if integer-gmp is used. (integer-simple is OK.) This is because "configure" of libgmp fails. Now I understand why this happens. The source of this problem is "/usr/bin/gcc" which is a wrapper for clang (not gcc actually). This cheats "configure" of libgmp. If we type "CC=clang ./configure", it works. The problem is how to specify "CC=clang" to configure of libgmp when we build GHC head. If we modified "libraries/integer-gmp/gmp/ghc.mk" manually, we can build GHC head. But I don't like this one since "make maintainer-clean" removes "ghc.mk". Any ideas? Ticket is here: https://ghc.haskell.org/trac/ghc/ticket/8497 --Kazu

2 3