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.
.
⟨reservedop⟩, such as M.\
, GHC will interpret it as a single qualified operator rather than the two lexemes M
and .\
.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 nonfix), the let
expression must terminate there. GHC simply gobbles up the whole expression, parsing like this:
(let x = 42 in x == 42 == True)

preceding certain expressions headed by keywords, allowing constructs like  case x of ...
or  do { ... }
. GHC does not allow this. Instead, unary 
is allowed before only expressions that could potentially be applied as a function. 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 etaexpands 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 fpedanticbottoms
.
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 stronglyconnected components, and then typechecked in dependency order (Haskell Report, Section 4.5.1). As each group is typechecked, 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 mutuallyrecursive 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 hsboot
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)
Num superclasses
The Num
class does not have Show
or Eq
superclasses.
You can make code that works with both Haskell98/Haskell2010 and GHC by:
Whenever you make a Num instance of a type, also make
Show
and Eq
instances, and
Whenever you give a function, instance or class a Num t
constraint, also give it Show t
and Eq t
constraints.
Bits superclass
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:
Whenever you make a Bits instance of a type, also make a
Num
instance, and
Whenever you give a function, instance or class a Bits t
constraint, also give it a Num t
constraint, and
Always define the bit, testBit and popCount methods
in Bits
instances.
Read class methods
Read
class has two extra methods, readPrec
and readListPrec
, that are not found in the Haskell 2010 since they rely on the ReadPrec
data type, which requires the XRankNTypes
extension. GHC also derives Read
instances by implementing readPrec
instead of readsPrec
, and relies on a default implementation of readsPrec
that is defined in terms of readPrec
. GHC adds these two extra methods simply because ReadPrec
is more efficient than ReadS
(the type on which readsPrec
is based).Monad superclass
Monad
class has an Applicative
superclass. You cannot write Monad
instances that work for GHC and also for a Haskell 2010 implementation that does not define Applicative
.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.
Prelude
supportTuples are currently limited to size 100. However, standard instances for tuples (Eq
, Ord
, Bounded
, Ix
, Read
, and Show
) are available only up to 16tuples.
This limitation is easily subvertible, so please ask if you get stuck on it.
splitAt semantics
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
Showing records
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}"
Reading integers
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.
isAlpha
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
.
hGetContents
System.IO.Error.catch
or Control.Exception.catch
.hs_init(), hs_exit()
hs_exit()
, but GHC does not currently support that. See Trac #13693.This section documents GHC’s take on various issues that are left undefined or implementation specific in Haskell 98.
Char
Following the ISO10646 standard, maxBound :: Char
in GHC is 0x10FFFF
.
Int
In GHC the Int
type follows the size of an address on the host architecture; in other words it holds 32 bits on a 32bit machine, and 64bits on a 64bit 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 bitpattern 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 bitpatterns 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.
Float
and Double
numbers are unchecked for overflow, underflow, and other sad occurrences. (note, however, that some architectures trap floatingpoint overflow and lossofprecision and report a floatingpoint exception, probably terminating the program)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 fnoomityields
(see f*: platformindependent 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 nontermination 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 nontermination is reported like this:
ghc: panic! (the 'impossible' happened) (GHC version 8.2.1 for x86_64unknownlinux): Simplifier ticks exhausted When trying UnfoldingDone x_alB To increase the limit, use fsimpltickfactor=N (default 100)
with the panic being reported no matter how high a fsimpltickfactor
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 unfixed. There is more background in Secrets of the GHC inliner.
On 32bit x86 platforms when using the native code generator, the fexcessprecision
option is always on. This means that floatingpoint calculations are nondeterministic, because depending on how the program is compiled (optimisation settings, for example), certain calculations might be done at 80bit precision instead of the intended 32bit or 64bit precision. Floatingpoint 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 Platformspecific Flags), which generates code to use the SSE2 instruction set instead of the x87 instruction set. SSE2 code uses the correct precision for all floatingpoint 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 nonobvious changes in evaluation ordering which may hide exceptions, even with fpedanticbottoms
(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 fnostatehack
results in the exception one would expect.
Programs compiled with fdefertypeerrors
may fail a bit more eagerly than one might expect. For instance,
{# OPTIONS_GHC fdefertypeerrors #} main = do putStrLn "Hi there." putStrLn True
Will emit no output, despite the fact that the illtyped term appears after the welltyped putStrLn "Hi there."
. See Trac #11197.
*
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. 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 defaulttype 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.
© 2002–2007 The University Court of the University of Glasgow. All rights reserved.
Licensed under the Glasgow Haskell Compiler License.
https://downloads.haskell.org/~ghc/8.2.1/docs/html/users_guide/bugs.html