July 2015 - ghc-tickets

[GHC] #10667: '-g' option generates invalid assembly when '*/*' operator is used
by GHC 14 Nov '15

14 Nov '15

#10667: '-g' option generates invalid assembly when '*/*' operator is used -------------------------------------+------------------------------------- Reporter: slyfox | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.10.2-rc2 (CodeGen) | Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Compile-time Unknown/Multiple | crash Test Case: | Blocked By: Blocking: | Related Tickets: Differential Revisions: | -------------------------------------+------------------------------------- Bug is observed when building cpphs-1.19 {{{#!hs module A where x */* y = 42 }}} {{{ $ ghc -fforce-recomp A -g [1 of 1] Compiling A ( A.hs, A.o ) /tmp/ghc23923_0/ghc_2.s: Assembler messages: /tmp/ghc23923_0/ghc_2.s:17:0: Error: bad expression /tmp/ghc23923_0/ghc_2.s:17:0: Warning: missing operand; zero assumed ... }}} The problem here is the following assembly snippet: {{{ .text .align 8 .loc 1 3 1 /* */* */ .quad 12884901911 .quad 0 .quad 15 }}} Would it be worthwile using ';' as a comment instead? Don't know if it's universally portable. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10667> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 7

[GHC] #10690: Save merged signatures to disk
by GHC 14 Nov '15

14 Nov '15

#10690: Save merged signatures to disk -------------------------------------+------------------------------------- Reporter: ezyang | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.10.1 Keywords: backpack | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Revisions: | -------------------------------------+------------------------------------- The Backpack model of development encourages users to mix signatures together. For example: {{{ unit p where signature H where x :: Int unit q where signature H where y :: Int unit r where include p include q module A where import H(x,y) }}} When `A` imports `H`, it should see both `x` and `y` (i.e. a merged version of the two signatures.) There are two possible ways to implement this: 1. The old strategy: When `A` imports `H`, it somehow "sees" both the interface for H from `p`, and the interface from `q`, and it merges them together on-the-fly. 2. The new strategy: Before compiling `A`, we merge the interfaces together, and use that merged interface as what A imports and gets its export from. Strategy (2) has a few advantages: in separate compilation, it avoids the need to repeatedly merge interfaces together, and it also simplifies GHC in that we don't have to manage on the fly signature merging. Now the challenge: GHC currently assumes that every build product is associated with a source file: however, merged interface files don't have this property. Both `--make` and one-shot `-c` compilation infrastructure is organized around this property. Here's is the proposal for how to slot this in: 1. Eliminate `HscSource` distinction between `HsBootFile` and `HsigFile`: the file is always called `A.hs-boot`. Instead, the type checker simply behaves differently depending on whether or not we know what the underlying implementation of `A` is (passed in using the `-sig-of` flag. If we know the underlying implementation is `containers:Data.Map`, we compile the hs-boot file "signature-style", if we don't, we assume it's `this-package:A` and compile it "boot-style" (Conceivably for cross-unit mutual recursion this will need to be more sophisticated, but this will do for now.) By the way, this means that all signature files now compile to A.hi-boot. 2. We introduce a new `HscSource` type, `HsBootMerge`, representing an `hi` (and stub `o`) file that will be generated by a signature merge. Intuitively, this operation will merge together A.hi-boot as well as any signatures which were brought into scope by a `-package` flag. Unlike `HsSrcFile` and `HsBootFile`, `ModSummary`s that are `HsBootMerge` don't have a source file. 3. When compiling `--make`, for all "signature-style" hs-boot files in the module graph, we generate an `HsBootMerge` summary to serve as the `NotBoot` node of that module name. This means if you `import A`, you're actually importing the `HsBootMerge` that produces `A.hi`. 4. When compiling one-shot, we have a new command for generating the `HsBootMerge`: `ghc --merge-requirements A`. This will go and look for an A.hi-boot as well as any external signatures and merge them together. Note that we can't use the existing compilation pipeline `-c` to generate merged hs-boot files, since these presuppose that there is some input source file, which there is none in this case. 5. When generating a Makefile dependency graph, an entry for an `HsBootMerge` looks like this: {{{ A.o : A.hi-boot A.o : /path/to/ghc/foo-0.1/A.hi # if -include-pkg-deps is added }}} That is to say, an `HsBootMerge` looks like a normal source file that imports each signature it's merging together, but doesn't have a source file associated with it. One downside with this representation is that it makes Makefile rules a little more complicated: how do you build `A.o`? Not with `ghc -c A.hs`, but with `ghc --merge-requirements A`. I'm not sure what the best way of solving this is. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10690> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 2

[GHC] #10622: Rename Backpack packages to units
by GHC 14 Nov '15

14 Nov '15

#10622: Rename Backpack packages to units -------------------------------------+------------------------------------- Reporter: ezyang | Owner: ezyang Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.11 Keywords: backpack | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Blocked By: Test Case: | Related Tickets: Blocking: | Differential Revisions: | -------------------------------------+------------------------------------- After today's weekly Backpack call, we have come to the conclusion that we have two different types of "packages" in the Backpack world: 1. Cabal packages, which have a single `.cabal` file and are a unit of distribution which get uploaded to Hackage, and 2. Backpack packages, of which there may be multiple defined in a Backpack file shipped with a Cabal package; and are the building blocks for modular development in the small. It's really confusing to have both of these called packages: thus, we propose to rename all occurrences of Backpack package to unit. A Cabal package may contain MULTIPLE Cabal units, although old-style Cabal files will only define one unit. Here are some of the consequences: 1. We rename `PackageKey` to `UnitId`, as it identifies a unit rather than a Cabal package. (I think this actually makes the function of these identifiers clearer.) We'll also distinguish Cabal-file level `PackageName`s from Backpack-file `UnitName`s. Finally, any given unit will be uniquely identified by an `InstalledUnitId`. 2. The source-level syntax of Backpack files will use `unit` in place of where `package` was used before. 3. For backwards compatibility reasons, we'll sometimes arrange for `PackageName`/`UnitName` and `InstalledUnitId`/`InstalledPackageId` to coincide. Specifically, the unit of a package which has the same `UnitName` as the `PackageName` is treated specially: its `InstalledUnitId` is guaranteed to be the same as the `InstalledPackageId` and it is what is "visible" when a user uses old concepts such as `-package foo-0.1` to select a package. 4. The installed package database is extended to record some number of units per an installed package: the public facing API is that you can register a package CONTAINING some number of units. For old-style packages, there will be only one unit with a matching `UnitName`, so this will be isomorphic to an old style package; however, GHC will grow some new APIs for selecting specific units to bring into scope. GHC really mostly only cares about units, but if a user asks for a package using, e.g. `-package` it will translate this request into a request for the appropriate unit. 5. For old-style packages, Cabal will continue to write and register a package configuration file which implicitly defines a single unit. However, the plan is to give GHC the capacity to generate unit description files (like package description files, but per unit), which Cabal can interpret and use to register packages in the global database using a `ghc-pkg` which can register units and "unitless" package description files which don't implicitly define a unit. (NB: we must create unit description files, because `ghc-pkg recache` is expected to be able to regenerate the database.) (NB: we want to be able to add units for a package separately, because this is how units for indefinite packages are created as they are instantiated with new implementations.) 6. Cabal could also finally grow the "multiple libraries per single Cabal file" support people have wanted; it's just a stylized use of the Backpack facilities. The work plan: 1. Modify `bin-package-db` to reflect the unit/package split, but otherwise keep ghc-pkg and GHC the same (so old-style package description still supported, and interpreted as a package containing one unit.) Source modifications to GHC assume that a package only has one unit. 2. Generalize GHC to work with packages with multiple units 3. Add capability to Cabal/ghc-pkg to register just units. For compilation of Backpack files, GHC will write out units which Cabal will then install to the real registry 4. Backpack! -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10622> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 13

[GHC] #10678: integer-gmp's runS seems unnecessarily expensive
by GHC 13 Nov '15

13 Nov '15

#10678: integer-gmp's runS seems unnecessarily expensive -------------------------------------+------------------------------------- Reporter: rwbarton | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.10.1 (CodeGen) | Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Runtime Unknown/Multiple | performance bug Test Case: | Blocked By: Blocking: | Related Tickets: Differential Revisions: | -------------------------------------+------------------------------------- integer-gmp uses an unsafePerformIO-like operation to work with mutable BigNats (unsafePerformIO and even the IO type are not yet available, since integer-gmp is a dependency of base): {{{ type S s a = State# s -> (# State# s, a #) -- NB: equivalent of GHC.IO.unsafeDupablePerformIO, see notes there runS :: S RealWorld a -> a runS m = lazy (case m realWorld# of (# _, r #) -> r) {-# NOINLINE runS #-} }}} It's tempting to think of such an operation as "free" like an unsafeCoerce, but it is actually somewhat expensive. Consider `plusBigNat` for instance. (Most BigNat operations have a similar structure.) {{{ plusBigNat :: BigNat -> BigNat -> BigNat plusBigNat x y | isTrue# (eqBigNatWord# x 0##) = y | isTrue# (eqBigNatWord# y 0##) = x | isTrue# (nx# >=# ny#) = go x nx# y ny# | True = go y ny# x nx# where go (BN# a#) na# (BN# b#) nb# = runS $ do mbn@(MBN# mba#) <- newBigNat# na# (W# c#) <- liftIO (c_mpn_add mba# a# na# b# nb#) case c# of 0## -> unsafeFreezeBigNat# mbn _ -> unsafeSnocFreezeBigNat# mbn c# nx# = sizeofBigNat# x ny# = sizeofBigNat# y }}} The assembly for `go` begins {{{ 00000000000001d0 <integerzmgmp_GHCziIntegerziType_zdwgo_info>: 1d0: 49 83 c4 28 add $0x28,%r12 1d4: 4d 3b a5 58 03 00 00 cmp 0x358(%r13),%r12 1db: 77 26 ja 203 <integerzmgmp_GHCziIntegerziType_zdwgo_info+0x33> 1dd: 49 c7 44 24 e0 00 00 movq $0x0,-0x20(%r12) 1e4: 00 00 1e2: R_X86_64_32S .text+0x38 1e6: 4d 89 74 24 e8 mov %r14,-0x18(%r12) 1eb: 49 89 7c 24 f0 mov %rdi,-0x10(%r12) 1f0: 49 89 74 24 f8 mov %rsi,-0x8(%r12) 1f5: 4d 89 04 24 mov %r8,(%r12) 1f9: 4d 8d 74 24 e1 lea -0x1f(%r12),%r14 1fe: e9 00 00 00 00 jmpq 203 <integerzmgmp_GHCziIntegerziType_zdwgo_info+0x33> 1ff: R_X86_64_PC32 integerzmgmp_GHCziIntegerziType_runS_info-0x4 203: ... ; heap overflow }}} This allocates a 5-word closure (containing `a#`, `na#`, `b#`, `nb#`) whose code is at `.text+0x38` and passes it to `runS`, which does some `stg_ap`-y things to call back into the closure, which reads its free variables back from the heap and finally does all the real work. Altogether it's around two dozen instructions compared to if we could call directly from `go` to the argument of `runS`. The old integer-gmp somehow avoided this particular overhead by instead using the implicit "unsafePerformIO" of a foreign import prim which performed both the allocation and the addition. Is this overhead a necessary consequence of doing the work in multiple steps in Haskell? I understand that we cannot allow everything to be inlined and, for example, the `newBigNat#` to be shared between a `plusBigNat` and `minusBigNat` with the same arguments. But once `runS` has done its job of keeping the `newBigNat#/c_mpn_add/unsafeFreeze*` together, it would be nice to eliminate it completely in the backend when compiling `go`, or any inlined version of `go`. I'm not sure whether this should be fixed in the code generator or in integer-gmp itself. I'm also aware that this is a tricky subject but haven't really done my homework on the related tickets, so I might be missing something important! -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10678> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 20

[GHC] #10061: Remove fun_infix from Funbind, as it is now in Match
by GHC 13 Nov '15

13 Nov '15

#10061: Remove fun_infix from Funbind, as it is now in Match -------------------------------------+------------------------------------- Reporter: alanz | Owner: Type: task | Status: new Priority: normal | Milestone: 7.12.1 Component: Compiler | Version: 7.10.1-rc2 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Blocked By: Test Case: | Related Tickets: #9988 Blocking: | Differential Revisions: | -------------------------------------+------------------------------------- One of the changes Phab:D538 introduced is to add m_fun_id_infix to Match {{{#!hs data Match id body = Match { m_fun_id_infix :: (Maybe (Located id,Bool)), -- fun_id and fun_infix for functions with multiple equations -- only present for a RdrName. See note [fun_id in Match] m_pats :: [LPat id], -- The patterns m_type :: (Maybe (LHsType id)), -- A type signature for the result of the match -- Nothing after typechecking m_grhss :: (GRHSs id body) } deriving (Typeable) }}} This was done to track the individual locations and fixity of the fun_id for each of the defining equations for a function when there are more than one. For example, the function (&&&) is defined with some prefix and some infix equations below. {{{#!hs (&&& ) [] [] = [] xs &&& [] = xs ( &&& ) [] ys = ys }}} This means that the fun_infix is now superfluous in the FunBind. This has not been removed as a potentially risky change just before 7.10 RC2, and so must be done after. This ticket captures that task, which includes processing these fields through the renamer and beyond. Ticket #9988 introduced these fields into Match through renaming, this ticket it to continue through type checking and then remove it from FunBind completely. The split happened so that #9988 could land in 7.10 -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10061> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 7

[GHC] #10172: Cross-platform sed
by GHC 11 Nov '15

11 Nov '15

#10172: Cross-platform sed -------------------------------------+------------------------------------- Reporter: ezyang | Owner: Type: task | Status: new Priority: normal | Milestone: Component: Build | Version: 7.11 System | Operating System: Solaris Keywords: | Type of failure: None/Unknown Architecture: | Blocked By: Unknown/Multiple | Related Tickets: Test Case: | Blocking: | Differential Revisions: | -------------------------------------+------------------------------------- It's quite easy to accidentally use GNU extensions in sed expressions, which means Solaris builds start failing. See https://phabricator.haskell.org/D740 for an example. What should we do? Here are a few possibilities: * Force use of POSIX sed with `--posix` flag. I haven't tested, but maybe this won't work if Solaris sed chokes on the `--posix`. Alternately, use `$(SED)` and add `--posix` only if it's supported * Detect GNU sed in autoconfigure and use `$(SED)` in all Makefile scripts * Stop using sed in the scripts for something else I don't know if this admonition applies to other random POSIX tools we use. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10172> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 5

[GHC] #9988: Remove fun_id, is_infix from FunBind, as they are now in Match
by GHC 11 Nov '15

11 Nov '15

#9988: Remove fun_id, is_infix from FunBind, as they are now in Match -------------------------------------+------------------------------------- Reporter: alanz | Owner: Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.10.1-rc1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Blocked By: Test Case: | Related Tickets: Blocking: | Differential Revisions: | -------------------------------------+------------------------------------- One of the changes Phab:D538 introduced is to add m_fun_id_infix to Match {{{#!hs data Match id body = Match { m_fun_id_infix :: (Maybe (Located id,Bool)), -- fun_id and fun_infix for functions with multiple equations -- only present for a RdrName. See note [fun_id in Match] m_pats :: [LPat id], -- The patterns m_type :: (Maybe (LHsType id)), -- A type signature for the result of the match -- Nothing after typechecking m_grhss :: (GRHSs id body) } deriving (Typeable) }}} This was done to track the individual locations and fixity of the fun_id for each of the defining equations for a function when there are more than one. For example, the function (&&&) is defined with some prefix and some infix equations below. {{{#!hs (&&& ) [] [] = [] xs &&& [] = xs ( &&& ) [] ys = ys }}} This means that the fun_id / is_infix is now superfluous in the FunBind. This has not been removed as a potentially risky change just before 7.10 RC2, and so must be done after. This ticket captures that task. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9988> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 16

[GHC] #10583: Chaos in Lexeme.hs
by GHC 09 Nov '15

09 Nov '15

#10583: Chaos in Lexeme.hs -------------------------------------+------------------------------------- Reporter: goldfire | Owner: goldfire Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.10.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Blocked By: Test Case: | Related Tickets: Blocking: | Differential Revisions: | -------------------------------------+------------------------------------- I've been looking at the `Lexeme` module (in `basicTypes`), where -- as far as I can tell -- utter chaos reigns. (Full disclosure: I wrote this module some time ago, inheriting its code from various places. But I clearly did a poor job of it.) Here is a sampling of the chaos: * `isLexConSym` claims to recognize type and data constructor infix symbols. But it requires symbols to start with a `:` (or be `->`). This is out-of-date with respect to the change in type constructor infix symbols in 7.6(?), which now do not need to start with a `:`. * `isVarSymChar` and `okSymChar` both purport to recognize characters that are valid parts of symbolic identifiers. But they have entirely different, unrelated implementations. These should be the '''same''' function, I believe. * The `notFollowedBySymbol` function defined in `parser/Lexer.x` overlaps with the functions above. But it has a '''third''' implementation, different than either of these other two. * The `isLexXXX` functions all just look at first characters, except for `isLexVarSym`, which looks at all characters. There is a reason for this -- that GHC-generated names start with a `$` but should be printed prefix -- but I'm not sure I buy it. Is it sufficient to look at the first two characters instead of the first one? I'm happy to make the code changes around this, but I need some advice from someone who has more knowledge about both Haskell's lexical structure and quite possibly Unicode. Happily, the function in `Lexeme` are not used much. But it would be awfully nice if they did the right thing when they are used. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10583> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 6

[GHC] #10633: GHCi segfaults on arm
by GHC 07 Nov '15

07 Nov '15

#10633: GHCi segfaults on arm ---------------------------------+------------------------------- Reporter: Thra11 | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: GHCi | Version: 7.10.1 Keywords: | Operating System: Linux Architecture: arm | Type of failure: GHCi crash Test Case: | Blocked By: Blocking: | Related Tickets: Differential Revisions: | ---------------------------------+------------------------------- I define a data type: {{{#!hs data Tree a = Leaf a | Tree { left :: Tree a , right :: Tree a } deriving Show }}} Then I create a few named trees: {{{#!hs testTree0 = Leaf 0 testTree1 = Tree (Leaf 0) (Leaf 0) testTree2 = Tree testTree0 testTree0 testTree3 = Tree testTree1 testTree1 testTree4 = Tree (Tree (Leaf 0) (Leaf 0)) (Tree (Leaf 0) (Leaf 0)) }}} This segfaults in GHCi (works fine with non-interactive GHC): {{{#!hs main = print testTree3 }}} == Notes == * testTree3 and testTree4 represent the same tree, constructed in different ways. However, testTree3 segfaults when it tries to print the second testTree1; testTree4 does not. * entering 'print testTree1' twice segfaults, but only if it is twice in a row. * strangely enough, printing testTree4 repeatedly does not segfault. * printing testTree0 many times in a row doesn't segfault either. * printing testTree2 segfaults. It seems to be something to do with repetition or reevaluation of the same named structure. I haven't yet managed to determine whether the print is relevant or whether it's forcing the evaluation of the data structure that causes the segfault (Can't remember how to force a list but discard the result). I have attached the the output from running ghci with +RTS -Di. Let me know if there's anything else I should try. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10633> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 3

[GHC] #9591: Add custom "Wiki" field in Trac
by GHC 06 Nov '15

06 Nov '15

#9591: Add custom "Wiki" field in Trac -------------------------------------+------------------------------------- Reporter: jstolarek | Owner: hvr Type: feature request | Status: new Priority: normal | Milestone: Component: Trac & Git | Version: Keywords: | Operating System: Architecture: Unknown/Multiple | Unknown/Multiple Difficulty: Unknown | Type of failure: Blocked By: | None/Unknown Related Tickets: | Test Case: | Blocking: | Differential Revisions: -------------------------------------+------------------------------------- I want to request a new custom field to Trac called "Wiki page"? Tickets that are about implementing a larger feature have an accompanying wiki page that summarizes the design. It would help my workflow if I could just get to that wiki page from the top of a Trac ticket. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9591> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 2