This section lists Glasgow Haskell infelicities in its implementation of Haskell 98 and Haskell 2010. See also the “when things go wrong” section (What to do when something goes wrong) for information about crashes, space leaks, and other undesirable phenomena.
The limitations here are listed in Haskell Report order (roughly).
By default, GHC mainly aims to behave (mostly) like a Haskell 2010 compiler, although you can tell it to try to behave like a particular version of the language with the -XHaskell98 and -XHaskell2010 flags. The known deviations from the standards are described below. Unless otherwise stated, the deviation applies in Haskell 98, Haskell 2010 and the default modes.
In Haskell 98 mode and by default (but not in Haskell 2010 mode), GHC is a little less strict about the layout rule when used in do expressions. Specifically, the restriction that “a nested context must be indented further to the right than the enclosing context” is relaxed to allow the nested context to be at the same level as the enclosing context, if the enclosing context is a do expression.
For example, the following code is accepted by GHC:
main = do args <- getArgs
if null args then return [] else do
ps <- mapM process args
mapM print ps
This behaviour is controlled by the NondecreasingIndentation extension.
GHC doesn’t do the fixity resolution in expressions during parsing as required by Haskell 98 (but not by Haskell 2010). For example, according to the Haskell 98 report, the following expression is legal:
let x = 42 in x == 42 == True
and parses as:
(let x = 42 in x == 42) == True
because according to the report, the let expression “extends as far to the right as possible”. Since it can’t extend past the second equals sign without causing a parse error (== is non-fix), the let-expression must terminate there. GHC simply gobbles up the whole expression, parsing like this:
(let x = 42 in x == 42 == True)
In its default mode, GHC makes some programs slightly more defined than they should be. For example, consider
f :: [a] -> b -> b
f [] = error "urk"
f (x:xs) = \v -> v
main = print (f [] `seq` True)
This should call error but actually prints True. Reason: GHC eta-expands f to
f :: [a] -> b -> b
f [] v = error "urk"
f (x:xs) v = v
This improves efficiency slightly but significantly for most programs, and is bad for only a few. To suppress this bogus “optimisation” use -fpedantic-bottoms.
In its default mode, GHC does not accept datatype contexts, as it has been decided to remove them from the next version of the language standard. This behaviour can be controlled with the DatatypeContexts extension. See Data type contexts.
The Haskell Report specifies that a group of bindings (at top level, or in a let or where) should be sorted into strongly-connected components, and then type-checked in dependency order (Haskell Report, Section 4.5.1). As each group is type-checked, any binders of the group that have an explicit type signature are put in the type environment with the specified polymorphic type, and all others are monomorphic until the group is generalised (Haskell Report, Section 4.5.2).
Following a suggestion of Mark Jones, in his paper Typing Haskell in Haskell, GHC implements a more general scheme. In GHC the dependency analysis ignores references to variables that have an explicit type signature. As a result of this refined dependency analysis, the dependency groups are smaller, and more bindings will typecheck. For example, consider:
f :: Eq a => a -> Bool
f x = (x == x) || g True || g "Yes"
g y = (y <= y) || f True
This is rejected by Haskell 98, but under Jones’s scheme the definition for g is typechecked first, separately from that for f, because the reference to f in g‘s right hand side is ignored by the dependency analysis. Then g‘s type is generalised, to get
g :: Ord a => a -> Bool
Now, the definition for f is typechecked, with this type for g in the type environment.
The same refined dependency analysis also allows the type signatures of mutually-recursive functions to have different contexts, something that is illegal in Haskell 98 (Section 4.5.2, last sentence). GHC only insists that the type signatures of a refined group have identical type signatures; in practice this means that only variables bound by the same pattern binding must have the same context. For example, this is fine:
f :: Eq a => a -> Bool
f x = (x == x) || g True
g :: Ord a => a -> Bool
g y = (y <= y) || f True
GHC requires the use of hs-boot files to cut the recursive loops among mutually recursive modules as described in How to compile mutually recursive modules. This more of an infelicity than a bug: the Haskell Report says (Section 5.7)
“Depending on the Haskell implementation used, separate compilation of mutually recursive modules may require that imported modules contain additional information so that they may be referenced before they are compiled. Explicit type signatures for all exported values may be necessary to deal with mutual recursion. The precise details of separate compilation are not defined by this Report.”
The Num class does not have Show or Eq superclasses.
You can make code that works with both Haskell98/Haskell2010 and GHC by:
Show and Eq instances, and
constraint, also give it Show t and Eq t constraints.
The Bits class does not have a Num superclass. It therefore does not have default methods for the bit, testBit and popCount methods.
You can make code that works with both Haskell 2010 and GHC by:
Num instance, and
constraint, also give it a Num t constraint, and
in Bits instances.
The following extra instances are defined:
instance Functor ((->) r)
instance Monad ((->) r)
instance Functor ((,) a)
instance Functor (Either a)
instance Monad (Either e)
This code fragment should elicit a fatal error, but it does not:
main = print (array (1,1) [(1,2), (1,3)])
GHC’s implementation of array takes the value of an array slot from the last (index,value) pair in the list, and does no checking for duplicates. The reason for this is efficiency, pure and simple.
Tuples are currently limited to size 100. However, standard instances for tuples (Eq, Ord, Bounded, Ix, Read, and Show) are available only up to 16-tuples.
This limitation is easily subvertible, so please ask if you get stuck on it.
Data.List.splitAt is more strict than specified in the Report. Specifically, the Report specifies that
splitAt n xs = (take n xs, drop n xs)
which implies that
splitAt undefined undefined = (undefined, undefined)
but GHC’s implementation is strict in its first argument, so
splitAt undefined [] = undefined
The Haskell 2010 definition of Show stipulates that the rendered string should only include parentheses which are necessary to unambiguously parse the result. For historical reasons, Show instances derived by GHC include parentheses around records despite the fact that record syntax binds more tightly than function application; e.g.,
data Hello = Hello { aField :: Int } deriving (Show)
-- GHC produces...
show (Just (Hello {aField=42})) == "Just (Hello {aField=42})"
-- whereas Haskell 2010 calls for...
show (Just (Hello {aField=42})) == "Just Hello {aField=42}"
GHC’s implementation of the Read class for integral types accepts hexadecimal and octal literals (the code in the Haskell 98 report doesn’t). So, for example,
read "0xf00" :: Int
works in GHC.
A possible reason for this is that readLitChar accepts hex and octal escapes, so it seems inconsistent not to do so for integers too.
The Haskell 98 definition of isAlpha is:
isAlpha c = isUpper c || isLower c
GHC’s implementation diverges from the Haskell 98 definition in the sense that Unicode alphabetic characters which are neither upper nor lower case will still be identified as alphabetic by isAlpha.
The Haskell Report demands that, for infix operators %, the following identities hold:
(% expr) = \x -> x % expr
(expr %) = \x -> expr % x
However, the second law is violated in the presence of undefined operators,
(%) = error "urk"
(() %) `seq` () -- urk
(\x -> () % x) `seq` () -- OK, result ()
The operator section is treated like function application of an undefined function, while the lambda form is in WHNF that contains an application of an undefined function.
This section documents GHC’s take on various issues that are left undefined or implementation specific in Haskell 98.
Following the ISO-10646 standard, maxBound :: Char in GHC is 0x10FFFF.
In GHC the Int type follows the size of an address on the host architecture; in other words it holds 32 bits on a 32-bit machine, and 64-bits on a 64-bit machine.
Arithmetic on Int is unchecked for overflowoverflowInt, so all operations on Int happen modulo 2⟨n⟩ where ⟨n⟩ is the size in bits of the Int type.
The fromInteger (and hence also fromIntegral) is a special case when converting to Int. The value of fromIntegral x :: Int is given by taking the lower ⟨n⟩ bits of (abs x), multiplied by the sign of x (in 2’s complement ⟨n⟩-bit arithmetic). This behaviour was chosen so that for example writing 0xffffffff :: Int preserves the bit-pattern in the resulting Int.
Negative literals, such as -3, are specified by (a careful reading of) the Haskell Report as meaning Prelude.negate (Prelude.fromInteger 3). So -2147483648 means negate (fromInteger 2147483648). Since fromInteger takes the lower 32 bits of the representation, fromInteger (2147483648::Integer), computed at type Int is -2147483648::Int. The negate operation then overflows, but it is unchecked, so negate (-2147483648::Int) is just -2147483648. In short, one can write minBound::Int as a literal with the expected meaning (but that is not in general guaranteed).
The fromIntegral function also preserves bit-patterns when converting between the sized integral types (Int8, Int16, Int32, Int64 and the unsigned Word variants), see the modules Data.Int and Data.Word in the library documentation.
The bug tracker lists bugs that have been reported in GHC but not yet fixed: see the GHC Trac. In addition to those, GHC also has the following known bugs or infelicities. These bugs are more permanent; it is unlikely that any of them will be fixed in the short term.
GHC’s runtime system implements cooperative multitasking, with context switching potentially occurring only when a program allocates. This means that programs that do not allocate may never context switch. This is especially true of programs using STM, which may deadlock after observing inconsistent state. See Trac #367 for further discussion.
If you are hit by this, you may want to compile the affected module with -fno-omit-yields (see -f*: platform-independent flags). This flag ensures that yield points are inserted at every function entrypoint (at the expense of a bit of performance).
GHC does not allow you to have a data type with a context that mentions type variables that are not data type parameters. For example:
data C a b => T a = MkT a
so that MkT‘s type is
MkT :: forall a b. C a b => a -> T a
In principle, with a suitable class declaration with a functional dependency, it’s possible that this type is not ambiguous; but GHC nevertheless rejects it. The type variables mentioned in the context of the data type declaration must be among the type parameters of the data type.
GHC’s inliner can be persuaded into non-termination using the standard way to encode recursion via a data type:
data U = MkU (U -> Bool)
russel :: U -> Bool
russel u@(MkU p) = not $ p u
x :: Bool
x = russel (MkU russel)
The non-termination is reported like this:
ghc: panic! (the 'impossible' happened)
(GHC version 8.2.1 for x86_64-unknown-linux):
Simplifier ticks exhausted
When trying UnfoldingDone x_alB
To increase the limit, use -fsimpl-tick-factor=N (default 100)
with the panic being reported no matter how high a -fsimpl-tick-factor you supply.
We have never found another class of programs, other than this contrived one, that makes GHC diverge, and fixing the problem would impose an extra overhead on every compilation. So the bug remains un-fixed. There is more background in Secrets of the GHC inliner.
On 32-bit x86 platforms when using the native code generator, the -fexcess-precision option is always on. This means that floating-point calculations are non-deterministic, because depending on how the program is compiled (optimisation settings, for example), certain calculations might be done at 80-bit precision instead of the intended 32-bit or 64-bit precision. Floating-point results may differ when optimisation is turned on. In the worst case, referential transparency is violated, because for example let x = E1 in E2 can evaluate to a different value than E2[E1/x].
One workaround is to use the -msse2 option (see Platform-specific Flags), which generates code to use the SSE2 instruction set instead of the x87 instruction set. SSE2 code uses the correct precision for all floating-point operations, and so gives deterministic results. However, note that this only works with processors that support SSE2 (Intel Pentium 4 or AMD Athlon 64 and later), which is why the option is not enabled by default. The libraries that come with GHC are probably built without this option, unless you built GHC yourself.
The state hack optimization can result in non-obvious changes in evaluation ordering which may hide exceptions, even with -fpedantic-bottoms (see, e.g., Trac #7411). For instance,
import Control.Exception
import Control.DeepSeq
main = do
evaluate (('a' : undefined) `deepseq` return () :: IO ())
putStrLn "Hello"
Compiling this program with -O results in Hello to be printed, despite the fact that evaluate should have bottomed. Compiling with -O -fno-state-hack results in the exception one would expect.
Programs compiled with -fdefer-type-errors may fail a bit more eagerly than one might expect. For instance,
{-# OPTIONS_GHC -fdefer-type-errors #-}
main = do
putStrLn "Hi there."
putStrLn True
Will emit no output, despite the fact that the ill-typed term appears after the well-typed putStrLn "Hi there.". See Trac #11197.
Despite appearances * and Constraint aren’t really distinct kinds in the compiler’s internal representation and can be unified producing unexpected results. See Trac #11715 for one example.
Because of a toolchain limitation we are unable to support full Unicode paths on Windows. On Windows we support up to Latin-1. See Trac #12971 for more.
GHCi does not respect the default declaration in the module whose scope you are in. Instead, for expressions typed at the command line, you always get the default default-type behaviour; that is, default(Int,Double).
It would be better for GHCi to record what the default settings in each module are, and use those of the ‘current’ module (whatever that is).
On Windows, there’s a GNU ld/BFD bug whereby it emits bogus PE object files that have more than 0xffff relocations. When GHCi tries to load a package affected by this bug, you get an error message of the form
Loading package javavm ... linking ... WARNING: Overflown relocation field (# relocs found: 30765)
The last time we looked, this bug still wasn’t fixed in the BFD codebase, and there wasn’t any noticeable interest in fixing it when we reported the bug back in 2001 or so.
The workaround is to split up the .o files that make up your package into two or more .o’s, along the lines of how the base package does it.