February 2014 - ghc-tickets

[GHC] #8741: `System.Directory.getPermissions` fails on read-only filesystem
by GHC 24 Apr '14

24 Apr '14

#8741: `System.Directory.getPermissions` fails on read-only filesystem -------------------------------------+------------------------------------- Reporter: hvr | Owner: Type: bug | Status: new Priority: high | Milestone: Component: | Version: 7.6.3 libraries/directory | Operating System: POSIX Keywords: | Type of failure: Incorrect result Architecture: Unknown/Multiple | at runtime Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | -------------------------------------+------------------------------------- Alain O'Dea reports in an [http://permalink.gmane.org/gmane.comp.lang.haskell.libraries/21078 email] that: {{{ Prelude> System.Directory.getPermissions "/usr/bin/ld" *** Exception: /usr/bin/ld: fileAccess: permission denied (Read-only file system) }}} > That seems wrong. > > An `access(*, W_OK)` syscall by design should return `EROFS` on a read- only file system by specification. > > This breaks Cabal on SmartOS since `/usr` is read-only by design and Cabal calls `getPermissions "/usr/bin/ld"`. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8741> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 12

[GHC] #8641: ghc with -odir and -prof cannot create the specified output directory
by GHC 22 Apr '14

22 Apr '14

#8641: ghc with -odir and -prof cannot create the specified output directory -----------------------------------+--------------------------------------- Reporter: wmarshall | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.6.3 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Compile-time crash Unknown/Multiple | Test Case: Difficulty: Unknown | Blocking: Blocked By: | Related Tickets: | -----------------------------------+--------------------------------------- Calling ghc with -odir set to a non-existent directory causes ghc to create the directory. However, if the -prof option is also specified, ghc fails to create the directory and quits with an error. Steps to reproduce: {{{ $ echo 'main = print "Hello world"' > Main.hs $ ghc Main.hs -odir bin [1 of 1] Compiling Main ( Main.hs, bin\Main.o ) Linking Main.exe ... $ rm -Rvf bin removed `bin/Main.o' removed directory: `bin' $ ghc Main.hs -odir bin -prof [1 of 1] Compiling Main ( Main.hs, bin\Main.o ) C:/Program Files (x86)/Haskell Platform/2013.2.0.0/mingw/bin/ld.exe: cannot open output file bin\Main.o: No such file or directory collect2: ld returned 1 exit status }}} Expected result: The output directory is created regardless of whether -prof is specified. The above commands were executed from within Cygwin, but the error occurs in the Windows command shell as well. (I used Cygwin here because it is easier to give precise instructions to reproduce the error.) I have only tested this on Windows. I do not know if the issue exists on Linux or other platforms. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8641> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 8

[GHC] #8458: T5435_dyn_asm fails on x86_64 Linux
by GHC 22 Apr '14

22 Apr '14

#8458: T5435_dyn_asm fails on x86_64 Linux --------------------------+------------------------------------------------ Reporter: | Owner: simonmar jstolarek | Status: new Type: bug | Milestone: Priority: normal | Version: 7.7 Component: | Operating System: Linux Runtime System | Type of failure: Incorrect result at runtime Keywords: | Test Case: Architecture: x86_64 | Blocking: (amd64) | Difficulty: | Unknown | Blocked By: | Related Tickets: | --------------------------+------------------------------------------------ I keep getting failure from `T5435_dyn_asm` test: {{{ make WAY=normal TEST=T5435_dyn_asm =====> T5435_dyn_asm(normal) 2622 of 3829 [0, 0, 0] cd ./rts && $MAKE -s --no-print-directory T5435_dyn_asm </dev/null >T5435_dyn_asm.run.stdout 2>T5435_dyn_asm.run.stderr Actual stdout output differs from expected: --- ./rts/T5435_dyn_asm.stdout 2013-10-16 09:49:22.236712001 +0200 +++ ./rts/T5435_dyn_asm.run.stdout 2013-10-18 19:26:11.076223002 +0200 @@ -1,5 +1,5 @@ -initArray1 -initArray2 ctors1 ctors2 +initArray1 +initArray2 success *** unexpected failure for T5435_dyn_asm(normal) }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8458> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 5

[GHC] #8141: Liberalising IncoherentInstances
by GHC 21 Apr '14

21 Apr '14

#8141: Liberalising IncoherentInstances ------------------------------------+------------------------------------- Reporter: nomeata | Owner: Type: feature request | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.6.3 Keywords: | Operating System: Unknown/Multiple Architecture: Unknown/Multiple | Type of failure: None/Unknown Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | ------------------------------------+------------------------------------- As [http://www.haskell.org/pipermail/glasgow-haskell- users/2013-July/022651.html|SPJ argued], -XIncoherentInstances could be even more liberal. This ticket is to give the proposal a number; I am currently preparing a patch for this. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8141> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 14

[GHC] #8491: Make BlockAlloc.c comments slightly more accurate
by GHC 13 Apr '14

13 Apr '14

#8491: Make BlockAlloc.c comments slightly more accurate ------------------------------------+------------------------------------- Reporter: ezyang | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.6.3 Keywords: | Operating System: Unknown/Multiple Architecture: Unknown/Multiple | Type of failure: None/Unknown Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | ------------------------------------+------------------------------------- I think this patch makes the comments for BlockAlloc slightly more accurate, but I'd like someone to double check for me. {{{ diff --git a/rts/sm/BlockAlloc.c b/rts/sm/BlockAlloc.c index 18c167f..41a2279 100644 --- a/rts/sm/BlockAlloc.c +++ b/rts/sm/BlockAlloc.c @@ -45,7 +45,7 @@ static void initMBlock(void *mblock); bd->free is either: - zero for a non-group-head; bd->link points to the head - (-1) for the head of a free block group - - or it points within the block + - or it points within the block (group) bd->blocks is either: - zero for a non-group-head; bd->link points to the head }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8491> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 3

[GHC] #8153: Implement AES intrinsics
by GHC 10 Apr '14

10 Apr '14

#8153: Implement AES intrinsics ----------------------------------------------+---------------------------- Reporter: schyler | Owner: Type: feature request | Status: new Priority: low | Milestone: Component: Compiler | Version: Keywords: | Operating System: Architecture: Unknown/Multiple | Unknown/Multiple Difficulty: Moderate (less than a day) | Type of failure: Blocked By: | None/Unknown Related Tickets: | Test Case: | Blocking: ----------------------------------------------+---------------------------- LLVM has the following intrinsics for working with AES encryption; @llvm.x86.aesni_{aesimc,aesenc,aesenclast,aesdec,aesdeclast,aeskeygenassist} It would be nice if at some stage these are supported through primops/intrinsics, as it would mean the crypto libraries can be rewritten to not use C and we'd get some nicer inlining on these. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8153> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 5

[GHC] #8157: Add a broader set of (C/CMM-based) atomic memory operations
by GHC 10 Apr '14

10 Apr '14

#8157: Add a broader set of (C/CMM-based) atomic memory operations ------------------------------------+------------------------------------- Reporter: rrnewton | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.6.3 Keywords: | Operating System: Unknown/Multiple Architecture: Unknown/Multiple | Type of failure: None/Unknown Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | ------------------------------------+------------------------------------- This task is a precursor to ticket #7883. The goal is to standardize a broader set of atomic ops -- mainly CAS and fetch-and-add -- and then in #7883 we will optimize those primops to have native support in the LLVM backend and thus avoid the extra C calls. Currently, the goal is only to support word-sized (4 or 8 byte) compare-and-swap and fetch-and-add operations. In the future, this goal could be expanded to support 16-byte CAS, implemented by CMPXCHG16B on x86. Even supporting word sized CAS requires three different primops for operating on `MutVar`, `MutableArray`, and `MutableByteArray`, respectively. Here are signatures for those primOps: {{{ #!C gcptr stg_casMutVarzh ( gcptr mv, gcptr old, gcptr new ) /* MutVar# s a -> a -> a -> State# s -> (# State#, Int#, Any a #) */ gcptr stg_casArrayzh ( gcptr arr, W_ ind, gcptr old, gcptr new ); /* MutableArray# s a -> Int# -> a -> a -> State# s -> (# State# s, Int#, Any a #) */ W_ stg_casIntArrayzh( gcptr arr, W_ ind, W_ old, W_ new ); /* MutableByteArray# s -> Int# -> Int# -> Int# -> State# s -> (# State# s, Int# #) */ }}} Fetch-and-add, on the other hand, only makes sense on a `MutableByteArray`: {{{ #!C W_ stg_fetchAddIntArrayzh( gcptr arr, W_ ind, W_ incr ); /* MutableByteArray# s -> Int# -> Int# -> State# s -> (# State# s, Int# #) */ }}} Some systems, such as gcc and LLVM, provide other variants, such as "fetch-and-multiply". However, these do not have direct support on x86 and are usually just simulated with CAS. For example, see the [http://llvm.org/docs/Atomics.html#atomics-and-codegen LLVM documentation here]. User-exposed Haskell libraries for atomic operations may employ the same approach using the CAS primOps above to simulate other fetch-and-X operations. Finally, a design question which bears thinking about is how to implement opaque "ticket" approach to CAS. The basic idea is that a user does NOT present a regular pointer-based data value (e.g. `Int`) to CAS as the "old" value. Rather, observations of prior values are kept in an opaque form (`Any` type) to prevent the compiler changing the pointer identity by, for example, unboxing and reboxing the value. The way this affects the **primOp** design is as follows. The current implementation of casMutVar# and casArray# returns the NEW value rather than the old one if the CAS is successful. The Haskell code that imports the primOps treats this new value as the opaque ticket. The user can in fact safely retrieve the real value from the `Any`-based "ticket" with a `peekTicket :: Ticket a -> a` operation. But [http://hackage.haskell.org/packages/archive/atomic-primops/0.4/doc/html /Data-Atomics.html#g:1 peekTicket] MUST [https://github.com/rrnewton /haskell-lockfree- queue/blob/880fb11de354d84f6c9a8ee06d5d4a887bd449a4/AtomicPrimops/Data/Atomics.hs#L114 be marked NOINLINE] to hide the `unsafeCoerce` operation inside it from compiler optimizations. Otherwise, the code path that carries the ticket from one CAS operation to the next is subject to optimization [1]. In particular, `unsafeCoerce` simply becomes a type-level operation, not a function call with the usual constraints on control and dataflow. Returning to the issue of primOp design, if we wished to avoid the somewhat non-standard design of returning the new value from CAS, we //could// allow users to create `Ticket` values themselves, rather retrieving them as the result of previous read and CAS IO operations. Yet this would entail an additional `NOINLINE` (unsafeCoerce) function call, and I believe it would be prone to error. Further, if you look at [https://github.com/ghc/ghc/blob/25ad01533aaaf1ccf69e1b7216e66de025b8131b/rt… the primOp implementation for stg_casMutVarzh], you will see that there is //already// a branch, which is necessary for updating GC-related meta-data. Thus there is no additional overhead to return the new rather than the old value from CAS operations in the successful case. Haskell-level CAS is already a compound operation, not merely a single machine code instruction. By contrast, the C intrinsic `__sync_val_compare_and_swap` can become only a single machine operation, and it returns the "old" value irrespective of whether the CAS succeeded or failed. In fact, an equality comparison on that old value is necessary to determine if the operation succeeded. In contrast, the Haskell CAS primOps above return an unboxed tuple with an explicit success bit. Thus returning the old value in the success case is completely redundant, and returning the new value instead allows us to acheive two goals (the CAS and the opacifying coercion) with one operation at no additional cost. [1] P.S. As a concrete example of what happens when `unsafeCoerce` is not hidden, consider the following code as a starting point: {{{ x = unsafeCoerce y z = f y }}} `x` would seem to have nothing to do with `f`, and yet later in the compiler, one can end up with Core like this: {{{ z = f x }}} This exposes the type information and activates optimizations, which was the source of [https://github.com/rrnewton/haskell-lockfree- queue/issues/5 one bug] with the `lockfree-queue` package that was fixed by adding NOINLINE to `peekTicket`. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8157> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 3

[GHC] #8745: GeneralizedNewtypeDeriving is still not Safe
by GHC 24 Mar '14

24 Mar '14

#8745: GeneralizedNewtypeDeriving is still not Safe ------------------------------------+------------------------------------- Reporter: goldfire | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.8.1-rc1 Keywords: Safe | Operating System: Unknown/Multiple Architecture: Unknown/Multiple | Type of failure: None/Unknown Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | ------------------------------------+------------------------------------- In GHC 7.8, !GeneralizedNewtypeDeriving (GND) is type-safe, due to the advent of roles. Although GND allows looking through unimported constructors, the notion of abstraction has changed -- for a type to be fully abstract in 7.8, it must not export its constructors, and the roles of all its parameters must be nominal. Thus, it would seem that GND should be Safe; that is, in the Safe Haskell subset. See also #5498. Currently, GND is not considered Safe. Along similar lines, the `Coercible` and `coerce` mechanism could be considered Safe, but there is no way to import these into a Safe module (they are in `GHC.Exts`). -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8745> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 27

[GHC] #8378: Cross-compiling from x86_64-unknown-linux-gnu to x86_64-sun-solaris2 with mkGmpConstants workaround fails to build objects for integer-gmp
by GHC 24 Mar '14

24 Mar '14

#8378: Cross-compiling from x86_64-unknown-linux-gnu to x86_64-sun-solaris2 with mkGmpConstants workaround fails to build objects for integer-gmp -------------------------------------+------------------------------------- Reporter: AlainODea | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Build System | Version: Keywords: solaris,integer- | Operating System: Unknown/Multiple gmp | Type of failure: Building GHC Architecture: Unknown/Multiple | failed Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: 8373,8366,8361 | -------------------------------------+------------------------------------- Here is the script I used to get build dependencies: {{{ sudo apt-get update sudo apt-get install --assume-yes ghc libncurses5-dev \ mesa-common-dev freeglut3-dev libedit-dev cabal install happy alex haddock }}} Here is the build bootstrap script I am using on a clean Ubuntu 13.04 install: {{{ #!/bin/bash export TARGET=x86_64-sun-solaris2.11 export PREFIX=$HOME/Documents/gcc/cross/$TARGET export SYSROOT=$PREFIX/sysroot export PATH=$PATH:$PREFIX/bin export GCC=$PREFIX/bin/$TARGET-gcc export LD=$PREFIX/bin/$TARGET-ld export NM=$PREFIX/bin/$TARGET-nm export OBJDUMP=$PREFIX/bin/$TARGET-objdump git clone git://github.com/ghc/ghc pushd ghc echo "Stage1Only = YES" >> mk/build.mk cp mk/build.mk{.sample,} ./sync-all --no-dph get perl boot patch -d libraries/haskeline/ -p1 < \ ../ghc-patches/0001-Include-termios.h-for- solaris2-host.-Fixes-8366.patch cp $SYSROOT/opt/local/include/gmp.h libraries/integer-gmp/gmp/ ./configure \ --host=x86_64-unknown-linux-gnu \ --build=x86_64-unknown-linux-gnu \ --target=x86_64-sun-solaris2 \ --with-gcc=$GCC \ --with-ld=$LD \ --with-nm=$NM \ --with-objdump=$OBJDUMP make # will fail with failure to run mkGmpDerivedConstants scp inplace/lib/bin/mkGmpDerivedConstants root@ns2:. ssh root@ns2 "./mkGmpDerivedConstants" > \ libraries/integer-gmp/mkGmpDerivedConstants/dist/GmpDerivedConstants.h make cat << __EOF__ >> libraries/base/include/HsBaseConfig.h #define HTYPE_DOUBLE Double #define HTYPE_FLOAT Float __EOF__ make #make -j $(nproc) popd }}} The log of the run is attached. It is too large for the ticket body and apparently too large for Gist as well. Somewhere along the way the build forgets integer-gmp and no objects are built for it. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8378> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 3

[GHC] #8831: Cannot use Template Haskell splice in ghci
by GHC 24 Mar '14

24 Mar '14

#8831: Cannot use Template Haskell splice in ghci ------------------------------------+--------------------------------- Reporter: goldfire | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Template Haskell | Version: 7.8.1-rc1 Keywords: | Operating System: MacOS X Architecture: x86_64 (amd64) | Type of failure: None/Unknown Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | ------------------------------------+--------------------------------- I have this module: {{{ {-# LANGUAGE TemplateHaskell #-} module A where foo = [| 3 |] }}} Now, I do this in ghci: {{{ > :seti -XTemplateHaskell > :load A.hs [1 of 1] Compiling A ( A.hs, interpreted ) Ok, modules loaded: A. > $foo Loading package array-0.5.0.0 ... linking ... done. Loading package deepseq-1.3.0.2 ... linking ... done. Loading package containers-0.5.4.0 ... linking ... done. Loading package pretty-1.1.1.1 ... linking ... done. Loading package template-haskell ... linking ... done. unknown package: main }}} The problem is the last line, failing to find the `main` package. I can reproduce this issue in 7.8rc1 and in HEAD. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8831> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 11