3. Using GHCi

GHCi [1] is GHC’s interactive environment that includes an interactive debugger (see The GHCi Debugger).

GHCi can

  • interactively evaluate Haskell expressions
  • interpret Haskell programs
  • load GHC-compiled modules.

At the moment GHCi supports most of GHC’s language extensions.

[1]The “i” stands for “Interactive”

3.1. Introduction to GHCi

Let’s start with an example GHCi session. You can fire up GHCi with the command ghci:

$ ghci
GHCi, version 8.y.z: https://www.haskell.org/ghc/  :? for help
ghci>

There may be a short pause while GHCi loads the prelude and standard libraries, after which the prompt is shown. As the banner says, you can type :? to see the list of commands available, and a half line description of each of them. We’ll explain most of these commands as we go along, and there is complete documentation for all the commands in GHCi commands.

Haskell expressions can be typed at the prompt:

ghci> 1+2
3
ghci> let x = 42 in x / 9
4.666666666666667
ghci>

GHCi interprets the whole line as an expression to evaluate. The expression may not span several lines - as soon as you press enter, GHCi will attempt to evaluate it.

In Haskell, a let expression is followed by in. However, in GHCi, since the expression can also be interpreted in the IO monad, a let binding with no accompanying in statement can be signalled by an empty line, as in the above example.

Since GHC 8.0.1, you can bind values and functions to names without let statement:

ghci> x = 42
ghci> x
42
ghci>

3.2. Loading source files

Suppose we have the following Haskell source code, which we place in a file Main.hs:

main = print (fac 20)

fac 0 = 1
fac n = n * fac (n-1)

You can save Main.hs anywhere you like, but if you save it somewhere other than the current directory [3] then we will need to change to the right directory in GHCi:

ghci> :cd dir

where ⟨dir⟩ is the directory (or folder) in which you saved Main.hs.

To load a Haskell source file into GHCi, use the :load command:

ghci> :load Main
Compiling Main             ( Main.hs, interpreted )
Ok, modules loaded: Main.
*ghci>

GHCi has loaded the Main module, and the prompt has changed to *ghci> to indicate that the current context for expressions typed at the prompt is the Main module we just loaded (we’ll explain what the * means later in What’s really in scope at the prompt?). So we can now type expressions involving the functions from Main.hs:

*ghci> fac 17
355687428096000

Loading a multi-module program is just as straightforward; just give the name of the “topmost” module to the :load command (hint: :load can be abbreviated to :l). The topmost module will normally be Main, but it doesn’t have to be. GHCi will discover which modules are required, directly or indirectly, by the topmost module, and load them all in dependency order.

[3]If you started up GHCi from the command line then GHCi’s current directory is the same as the current directory of the shell from which it was started. If you started GHCi from the “Start” menu in Windows, then the current directory is probably something like C:\Documents and Settings\user name.
-fshow-loaded-modules
Default:off
Since:8.2.2

Typically GHCi will show only the number of modules that it loaded after a :load command. With this flag, GHC will also list the loaded modules’ names. This was the default behavior prior to GHC 8.2.1 and can be useful for some tooling users.

3.2.1. Modules vs. filenames

Question: How does GHC find the filename which contains module ⟨M⟩? Answer: it looks for the file M.hs, or M.lhs. This means that for most modules, the module name must match the filename. If it doesn’t, GHCi won’t be able to find it.

There is one exception to this general rule: when you load a program with :load, or specify it when you invoke ghci, you can give a filename rather than a module name. This filename is loaded if it exists, and it may contain any module you like. This is particularly convenient if you have several Main modules in the same directory and you can’t call them all Main.hs.

The search path for finding source files is specified with the -i option on the GHCi command line, like so:

ghci -idir1:...:dirn

or it can be set using the :set command from within GHCi (see Setting GHC command-line options in GHCi) [4]

One consequence of the way that GHCi follows dependencies to find modules to load is that every module must have a source file. The only exception to the rule is modules that come from a package, including the Prelude and standard libraries such as IO and Complex. If you attempt to load a module for which GHCi can’t find a source file, even if there are object and interface files for the module, you’ll get an error message.

[4]Note that in GHCi, and --make mode, the -i option is used to specify the search path for source files, whereas in standard batch-compilation mode the -i option is used to specify the search path for interface files, see The search path.

3.2.2. Making changes and recompilation

If you make some changes to the source code and want GHCi to recompile the program, give the :reload command. The program will be recompiled as necessary, with GHCi doing its best to avoid actually recompiling modules if their external dependencies haven’t changed. This is the same mechanism we use to avoid re-compiling modules in the batch compilation setting (see The recompilation checker).

3.3. Loading compiled code

When you load a Haskell source module into GHCi, it is normally converted to byte-code and run using the interpreter. However, interpreted code can also run alongside compiled code in GHCi; indeed, normally when GHCi starts, it loads up a compiled copy of the base package, which contains the Prelude.

Why should we want to run compiled code? Well, compiled code is roughly 10x faster than interpreted code, but takes about 2x longer to produce (perhaps longer if optimisation is on). So it pays to compile the parts of a program that aren’t changing very often, and use the interpreter for the code being actively developed.

When loading up source modules with :load, GHCi normally looks for any corresponding compiled object files, and will use one in preference to interpreting the source if possible. For example, suppose we have a 4-module program consisting of modules A, B, C, and D. Modules B and C both import D only, and A imports both B and C:

  A
 / \
B   C
 \ /
  D

We can compile D, then load the whole program, like this:

ghci> :! ghc -c -dynamic D.hs
ghci> :load A
Compiling B                ( B.hs, interpreted )
Compiling C                ( C.hs, interpreted )
Compiling A                ( A.hs, interpreted )
Ok, modules loaded: A, B, C, D (D.o).
*ghci>

In the messages from the compiler, we see that there is no line for D. This is because it isn’t necessary to compile D, because the source and everything it depends on is unchanged since the last compilation.

Note the -dynamic flag to GHC: GHCi uses dynamically-linked object code (if you are on a platform that supports it), and so in order to use compiled code with GHCi it must be compiled for dynamic linking.

At any time you can use the command :show modules to get a list of the modules currently loaded into GHCi:

*ghci> :show modules
D                ( D.hs, D.o )
C                ( C.hs, interpreted )
B                ( B.hs, interpreted )
A                ( A.hs, interpreted )
*ghci>

If we now modify the source of D (or pretend to: using the Unix command touch on the source file is handy for this), the compiler will no longer be able to use the object file, because it might be out of date:

*ghci> :! touch D.hs
*ghci> :reload
Compiling D                ( D.hs, interpreted )
Ok, modules loaded: A, B, C, D.
*ghci>

Note that module D was compiled, but in this instance because its source hadn’t really changed, its interface remained the same, and the recompilation checker determined that A, B and C didn’t need to be recompiled.

So let’s try compiling one of the other modules:

*ghci> :! ghc -c C.hs
*ghci> :load A
Compiling D                ( D.hs, interpreted )
Compiling B                ( B.hs, interpreted )
Compiling C                ( C.hs, interpreted )
Compiling A                ( A.hs, interpreted )
Ok, modules loaded: A, B, C, D.

We didn’t get the compiled version of C! What happened? Well, in GHCi a compiled module may only depend on other compiled modules, and in this case C depends on D, which doesn’t have an object file, so GHCi also rejected C’s object file. Ok, so let’s also compile D:

*ghci> :! ghc -c D.hs
*ghci> :reload
Ok, modules loaded: A, B, C, D.

Nothing happened! Here’s another lesson: newly compiled modules aren’t picked up by :reload, only :load:

*ghci> :load A
Compiling B                ( B.hs, interpreted )
Compiling A                ( A.hs, interpreted )
Ok, modules loaded: A, B, C (C.o), D (D.o).

The automatic loading of object files can sometimes lead to confusion, because non-exported top-level definitions of a module are only available for use in expressions at the prompt when the module is interpreted (see What’s really in scope at the prompt?). For this reason, you might sometimes want to force GHCi to load a module using the interpreter. This can be done by prefixing a * to the module name or filename when using :load, for example

ghci> :load *A
Compiling A                ( A.hs, interpreted )
*ghci>

When the * is used, GHCi ignores any pre-compiled object code and interprets the module. If you have already loaded a number of modules as object code and decide that you wanted to interpret one of them, instead of re-loading the whole set you can use :add *M to specify that you want M to be interpreted (note that this might cause other modules to be interpreted too, because compiled modules cannot depend on interpreted ones).

To always compile everything to object code and never use the interpreter, use the -fobject-code option (see Compiling to object code inside GHCi).

Hint

Since GHCi will only use a compiled object file if it can be sure that the compiled version is up-to-date, a good technique when working on a large program is to occasionally run ghc --make to compile the whole project (say before you go for lunch :-), then continue working in the interpreter. As you modify code, the changed modules will be interpreted, but the rest of the project will remain compiled.

3.4. Interactive evaluation at the prompt

When you type an expression at the prompt, GHCi immediately evaluates and prints the result:

ghci> reverse "hello"
"olleh"
ghci> 5+5
10

3.4.1. I/O actions at the prompt

GHCi does more than simple expression evaluation at the prompt. If you enter an expression of type IO a for some a, then GHCi executes it as an IO-computation.

ghci> "hello"
"hello"
ghci> putStrLn "hello"
hello

This works even if the type of the expression is more general, provided it can be instantiated to IO a. For example

ghci> return True
True

Furthermore, GHCi will print the result of the I/O action if (and only if):

  • The result type is an instance of Show.
  • The result type is not ().

For example, remembering that putStrLn :: String -> IO ():

ghci> putStrLn "hello"
hello
ghci> do { putStrLn "hello"; return "yes" }
hello
"yes"

3.4.2. Using do notation at the prompt

GHCi actually accepts statements rather than just expressions at the prompt. This means you can bind values and functions to names, and use them in future expressions or statements.

The syntax of a statement accepted at the GHCi prompt is exactly the same as the syntax of a statement in a Haskell do expression. However, there’s no monad overloading here: statements typed at the prompt must be in the IO monad.

ghci> x <- return 42
ghci> print x
42
ghci>

The statement x <- return 42 means “execute return 42 in the IO monad, and bind the result to x”. We can then use x in future statements, for example to print it as we did above.

-fprint-bind-result

If -fprint-bind-result is set then GHCi will print the result of a statement if and only if:

  • The statement is not a binding, or it is a monadic binding (p <- e) that binds exactly one variable.
  • The variable’s type is not polymorphic, is not (), and is an instance of Show.

Of course, you can also bind normal non-IO expressions using the let-statement:

ghci> let x = 42
ghci> x
42
ghci>

Another important difference between the two types of binding is that the monadic bind (p <- e) is strict (it evaluates e), whereas with the let form, the expression isn’t evaluated immediately:

ghci> let x = error "help!"
ghci> print x
*** Exception: help!
ghci>

Note that let bindings do not automatically print the value bound, unlike monadic bindings.

You can also define functions at the prompt:

ghci> add a b = a + b
ghci> add 1 2
3
ghci>

However, this quickly gets tedious when defining functions with multiple clauses, or groups of mutually recursive functions, because the complete definition has to be given on a single line, using explicit semicolons instead of layout:

ghci> f op n [] = n ; f op n (h:t) = h `op` f op n t
ghci> f (+) 0 [1..3]
6
ghci>
:{
:}

Begin or end a multi-line GHCi command block.

To alleviate this issue, GHCi commands can be split over multiple lines, by wrapping them in :{ and :} (each on a single line of its own):

ghci> :{
ghci| g op n [] = n
ghci| g op n (h:t) = h `op` g op n t
ghci| :}
ghci> g (*) 1 [1..3]
6

Such multiline commands can be used with any GHCi command, and note that the layout rule is in effect. The main purpose of multiline commands is not to replace module loading but to make definitions in .ghci-files (see The .ghci and .haskeline files) more readable and maintainable.

Any exceptions raised during the evaluation or execution of the statement are caught and printed by the GHCi command line interface (for more information on exceptions, see the module Control.Exception in the libraries documentation.)

Every new binding shadows any existing bindings of the same name, including entities that are in scope in the current module context.

Warning

Temporary bindings introduced at the prompt only last until the next :load, :reload, :add or :unadd command, at which time they will be simply lost. However, they do survive a change of context with :module: the temporary bindings just move to the new location.

Hint

To get a list of the bindings currently in scope, use the :show bindings command:

ghci> :show bindings
x :: Int
ghci>

Hint

If you turn on the +t option, GHCi will show the type of each variable bound by a statement. For example:

ghci> :set +t
ghci> let (x:xs) = [1..]
x :: Integer
xs :: [Integer]

3.4.3. Multiline input

Apart from the :{ ... :} syntax for multi-line input mentioned above, GHCi also has a multiline mode, enabled by :set +m, :set +m in which GHCi detects automatically when the current statement is unfinished and allows further lines to be added. A multi-line input is terminated with an empty line. For example:

ghci> :set +m
ghci> let x = 42
ghci|

Further bindings can be added to this let statement, so GHCi indicates that the next line continues the previous one by changing the prompt. Note that layout is in effect, so to add more bindings to this let we have to line them up:

ghci> :set +m
ghci> let x = 42
ghci|     y = 3
ghci|
ghci>

Explicit braces and semicolons can be used instead of layout:

ghci> do {
ghci| putStrLn "hello"
ghci| ;putStrLn "world"
ghci| }
hello
world
ghci>

Note that after the closing brace, GHCi knows that the current statement is finished, so no empty line is required.

Multiline mode is useful when entering monadic do statements:

ghci> flip evalStateT 0 $ do
ghci| i <- get
ghci| lift $ do
ghci|   putStrLn "Hello World!"
ghci|   print i
ghci|
"Hello World!"
0
ghci>

During a multiline interaction, the user can interrupt and return to the top-level prompt.

ghci> do
ghci| putStrLn "Hello, World!"
ghci| ^C
ghci>

3.4.4. Type, class and other declarations

At the GHCi prompt you can also enter any top-level Haskell declaration, including data, type, newtype, class, instance, deriving, and foreign declarations. For example:

ghci> data T = A | B | C deriving (Eq, Ord, Show, Enum)
ghci> [A ..]
[A,B,C]
ghci> :i T
data T = A | B | C      -- Defined at <interactive>:2:6
instance Enum T -- Defined at <interactive>:2:45
instance Eq T -- Defined at <interactive>:2:30
instance Ord T -- Defined at <interactive>:2:34
instance Show T -- Defined at <interactive>:2:39

As with ordinary variable bindings, later definitions shadow earlier ones, so you can re-enter a declaration to fix a problem with it or extend it. But there’s a gotcha: when a new type declaration shadows an older one, there might be other declarations that refer to the old type. The thing to remember is that the old type still exists, and these other declarations still refer to the old type. However, while the old and the new type have the same name, GHCi will treat them as distinct. For example:

ghci> data T = A | B
ghci> let f A = True; f B = False
ghci> data T = A | B | C
ghci> f A

<interactive>:2:3:
    Couldn't match expected type `main::Interactive.T'
                with actual type `T'
    In the first argument of `f', namely `A'
    In the expression: f A
    In an equation for `it': it = f A
ghci>

The old, shadowed, version of T is displayed as main::Interactive.T by GHCi in an attempt to distinguish it from the new T, which is displayed as simply T.

Class and type-family instance declarations are simply added to the list of available instances, with one exception. Since you might want to re-define one, a class instance replaces any earlier instance with an identical head. You aren’t allowed to re-define a type family instance, since it might not be type safe to do so. Instead, re-define the whole type-family. (See Type families.) For example:

ghci> type family T a b
ghci> type instance T a b = a
ghci> let uc :: a -> T a b; uc = id

ghci> type instance T a b = b

<interactive>:3:15: error:
    Conflicting family instance declarations:
      T a b = a -- Defined at <interactive>:3:15
      T a b = b -- Defined at <interactive>:5:15

-- Darn! We have to re-declare T.

ghci> type family T a b
-- This is a brand-new T, unrelated to the old one
ghci> type instance T a b = b
ghci> uc 'a' :: Int

<interactive>:8:1: error:
    • Couldn't match type ‘Char’ with ‘Int’
      Expected type: Int
        Actual type: Ghci1.T Char b0
    • In the expression: uc 'a' :: Int
      In an equation for ‘it’: it = uc 'a' :: Int

3.4.5. What’s really in scope at the prompt?

When you type an expression at the prompt, what identifiers and types are in scope? GHCi provides a flexible way to control exactly how the context for an expression is constructed:

The command :show imports will show a summary of which modules contribute to the top-level scope.

Hint

GHCi will tab-complete names that are in scope; for example, if you run GHCi and type J<tab> then GHCi will expand it to Just.

3.4.5.1. The effect of :load on what is in scope

The :load, :reload, :add and :unadd commands (Loading source files and Loading compiled code) affect the top-level scope. Let’s start with the simple cases; when you start GHCi the prompt looks like this:

ghci>

By default, this means that everything from the module Prelude is currently in scope. Should the prompt be set to %s> in the .ghci configuration file, we would be seeing Prelude> displayed. However, it is not the default mechanism due to the large space the prompt can take if more imports are done.

The syntax in the prompt *module indicates that it is the full top-level scope of ⟨module⟩ that is contributing to the scope for expressions typed at the prompt. Without the *, just the exports of the module are visible.

Note

For technical reasons, GHCi can only support the *-form for modules that are interpreted. Compiled modules and package modules can only contribute their exports to the current scope. To ensure that GHCi loads the interpreted version of a module, add the * when loading the module, e.g. :load *M.

In general, after a :load command, an automatic import is added to the scope for the most recently loaded “target” module, in a *-form if possible. For example, if you say :load foo.hs bar.hs and bar.hs contains module Bar, then the scope will be set to *Bar if Bar is interpreted, or if Bar is compiled it will be set to Prelude and Bar (GHCi automatically adds Prelude if it isn’t present and there aren’t any *-form modules). These automatically-added imports can be seen with :show imports:

ghci> :load hello.hs
[1 of 1] Compiling Main             ( hello.hs, interpreted )
Ok, modules loaded: Main.
*ghci> :show imports
:module +*Main -- added automatically
*ghci>

and the automatically-added import is replaced the next time you use :load, :reload, :add or :unadd. It can also be removed by :module as with normal imports.

3.4.5.2. Controlling what is in scope with import

We are not limited to a single module: GHCi can combine scopes from multiple modules, in any mixture of * and non-* forms. GHCi combines the scopes from all of these modules to form the scope that is in effect at the prompt.

To add modules to the scope, use ordinary Haskell import syntax:

ghci> import System.IO
ghci> hPutStrLn stdout "hello\n"
hello

The full Haskell import syntax is supported, including hiding and as clauses. The prompt shows the modules that are currently imported, but it omits details about hiding, as, and so on. To see the full story, use :show imports:

ghci> import System.IO
ghci> import Data.Map as Map
ghci Map> :show imports
import Prelude -- implicit
import System.IO
import Data.Map as Map

Note that the Prelude import is marked as implicit. It can be overridden with an explicit Prelude import, just like in a Haskell module.

With multiple modules in scope, especially multiple *-form modules, it is likely that name clashes will occur. Haskell specifies that name clashes are only reported when an ambiguous identifier is used, and GHCi behaves in the same way for expressions typed at the prompt.

3.4.5.3. Controlling what is in scope with the :module command

Another way to manipulate the scope is to use the :module command, whose syntax is this:

:module +|- *mod1 ... *modn

Using the + form of the module commands adds modules to the current scope, and - removes them. Without either + or -, the current scope is replaced by the set of modules specified. Note that if you use this form and leave out Prelude, an implicit Prelude import will be added automatically.

The :module command provides a way to do two things that cannot be done with ordinary import declarations:

  • :module supports the * modifier on modules, which opens the full top-level scope of a module, rather than just its exports.
  • Imports can be removed from the context, using the syntax :module -M. The import syntax is cumulative (as in a Haskell module), so this is the only way to subtract from the scope.

3.4.5.4. Qualified names

-fimplicit-import-qualified
Default:on

To make life slightly easier, the GHCi prompt also behaves as if there is an implicit import qualified declaration for every module in every package, and every module currently loaded into GHCi. This behaviour can be disabled with the -fno-implicit-import-qualified flag.

3.4.5.5. :module and :load

It might seem that :module/import and :load/:add/:reload do similar things: you can use both to bring a module into scope. However, there is a very important difference. GHCi is concerned with two sets of modules:

  • The set of modules that are currently loaded. This set is modified by :load, :reload, :add and :unadd, and can be shown with :show modules.
  • The set of modules that are currently in scope at the prompt. This set is modified by import and :module, and it is also modified automatically after :load, :reload, :add and :unadd, as described above. The set of modules in scope can be shown with :show imports.

You can add a module to the scope (via :module or import) only if either (a) it is loaded, or (b) it is a module from a package that GHCi knows about. Using :module or import to try bring into scope a non-loaded module may result in the message module M is not loaded.

3.4.5.6. Shadowing and the Ghci1 module name

Bindings on the prompt can shadow earlier bindings:

ghci> let foo = True
ghci> let foo = False
ghci> :show bindings
foo :: Bool = False

But the shadowed thing still exists, and may show up again later, for example in a type signature:

ghci> data T = A | B deriving Eq
ghci> let a = A
ghci> data T = ANewType
ghci> :t a
a :: Ghci1.T

Now the type of a is printed using the fully qualified name of T, using the module name Ghci1 (and Ghci2 for the next set of bindings, and so on). You can use these qualified names as well:

ghci> a == Ghci1.A
True
ghci> let a = False -- shadowing a
ghci> Ghci2.a == Ghci1.A
True

The command :show bindings only shows bindings that are not shadowed. Bindings that define multiple names, such as a type constructor and its data constructors, are shown if any defined name is still available without the need for qualification.

3.4.6. The it variable

Whenever an expression (or a non-binding statement, to be precise) is typed at the prompt, GHCi implicitly binds its value to the variable it. For example:

ghci> 1+2
3
ghci> it * 2
6

What actually happens is that GHCi typechecks the expression, and if it doesn’t have an IO type, then it transforms it as follows: an expression e turns into

let it = e;
print it

which is then run as an IO-action.

Hence, the original expression must have a type which is an instance of the Show class, or GHCi will complain:

ghci> id

<interactive>:1:0:
    No instance for (Show (a -> a))
      arising from use of `print' at <interactive>:1:0-1
    Possible fix: add an instance declaration for (Show (a -> a))
    In the expression: print it
    In a 'do' expression: print it

The error message contains some clues as to the transformation happening internally.

If the expression was instead of type IO a for some a, then it will be bound to the result of the IO computation, which is of type a. eg.:

ghci> Data.Time.getZonedTime
2017-04-10 12:34:56.93213581 UTC
ghci> print it
2017-04-10 12:34:56.93213581 UTC

The corresponding translation for an IO-typed e is

it <- e

Note that it is shadowed by the new value each time you evaluate a new expression, and the old value of it is lost.

In order to stop the value it being bound on each command, the flag -fno-it can be set. The it variable can be the source of space leaks due to how shadowed declarations are handled by GHCi (see Type, class and other declarations).

-fno-it

When this flag is set, the variable it will no longer be set to the result of the previously evaluated expression.

3.4.7. Type defaulting in GHCi

ExtendedDefaultRules
Since:6.8.1

Allow defaulting to take place for more than just numeric classes.

Consider this GHCi session:

ghci> reverse []

What should GHCi do? Strictly speaking, the program is ambiguous. show (reverse []) (which is what GHCi computes here) has type Show a => String and how that displays depends on the type a. For example:

ghci> reverse ([] :: String)
""
ghci> reverse ([] :: [Int])
[]

However, it is tiresome for the user to have to specify the type, so GHCi extends Haskell’s type-defaulting rules (Section 4.3.4 of the Haskell 2010 Report) as follows. The standard rules take each group of constraints (C1 a, C2 a, ..., Cn a) for each type variable a, and defaults the type variable if

  1. The type variable a appears in no other constraints
  2. All the classes Ci are standard.
  3. At least one of the classes Ci is numeric.

At the GHCi prompt, or with GHC if the ExtendedDefaultRules flag is given, the types are instead resolved with the following method:

Find all the unsolved constraints. Then:

  • Find those that are of form (C a) where a is a type variable, and partition those constraints into groups that share a common type variable a.
  • Keep only the groups in which at least one of the classes is an interactive class (defined below).
  • Now, for each remaining group G, try each type ty from the default-type list in turn; if setting a = ty would allow the constraints in G to be completely solved. If so, default a to ty.
  • The unit type () and the list type [] are added to the start of the standard list of types which are tried when doing type defaulting.

Note that any multi-parameter constraints (D a b) or (D [a] Int) do not participate in the process (either to help or to hinder); but they must of course be soluble once the defaulting process is complete.

The last point means that, for example, this program:

main :: IO ()
main = print def

instance Num ()

def :: (Num a, Enum a) => a
def = toEnum 0

prints () rather than 0 as the type is defaulted to () rather than Integer.

The motivation for the change is that it means IO a actions default to IO (), which in turn means that ghci won’t try to print a result when running them. This is particularly important for printf, which has an instance that returns IO a. However, it is only able to return undefined (the reason for the instance having this type is so that printf doesn’t require extensions to the class system), so if the type defaults to Integer then ghci gives an error when running a printf.

See also I/O actions at the prompt for how the monad of a computational expression defaults to IO if possible.

3.4.7.1. Interactive classes

The interactive classes (only relevant when ExtendedDefaultRules is in effect) are: any numeric class, Show, Eq, Ord, Foldable or Traversable.

As long as a type variable is constrained by one of these classes, defaulting will occur, as outlined above.

3.4.7.2. Extended rules around default declarations

Since the rules for defaulting are relaxed under ExtendedDefaultRules, the rules for default declarations are also relaxed. According to Section 4.3.4 of the Haskell 2010 Report, a default declaration looks like default (t1, ..., tn) where, for each ti, Num ti must hold. This is relaxed to say that for each ti, there must exist an interactive class C such that C ti holds. This means that type constructors can be allowed in these lists. For example, the following works if you wish your Foldable constraints to default to Maybe but your Num constraints to still default to Integer or Double:

default (Maybe, Integer, Double)

3.4.8. Using a custom interactive printing function

Since GHC 7.6.1, GHCi prints the result of expressions typed at the prompt using the function System.IO.print. Its type signature is Show a => a -> IO (), and it works by converting the value to String using show.

This is not ideal in certain cases, like when the output is long, or contains strings with non-ascii characters.

The -interactive-print ⟨name⟩ flag allows to specify any function of type C a => a -> IO (), for some constraint C, as the function for printing evaluated expressions. The function can reside in any loaded module or any registered package, but only when it resides in a registered package will it survive a :cd, :add, :unadd, :load, :reload or, :set.

-interactive-print ⟨name⟩

Set the function used by GHCi to print evaluation results. Given name must be of type C a => a -> IO ().

As an example, suppose we have following special printing module:

module SpecPrinter where
import System.IO

sprint a = putStrLn $ show a ++ "!"

The sprint function adds an exclamation mark at the end of any printed value. Running GHCi with the command:

ghci -interactive-print=SpecPrinter.sprint SpecPrinter

will start an interactive session where values with be printed using sprint:

*SpecPrinter> [1,2,3]
[1,2,3]!
*SpecPrinter> 42
42!

A custom pretty printing function can be used, for example, to format tree-like and nested structures in a more readable way.

The -interactive-print ⟨name⟩ flag can also be used when running GHC in -e mode:

% ghc -e "[1,2,3]" -interactive-print=SpecPrinter.sprint SpecPrinter
[1,2,3]!

3.4.9. Stack Traces in GHCi

[ This is an experimental feature enabled by the new -fexternal-interpreter flag that was introduced in GHC 8.0.1. It is currently not supported on Windows.]

GHCi can use the profiling system to collect stack trace information when running interpreted code. To gain access to stack traces, start GHCi like this:

ghci -fexternal-interpreter -prof

This runs the interpreted code in a separate process (see Running the interpreter in a separate process) and runs it in profiling mode to collect call stack information. Note that because we’re running the interpreted code in profiling mode, all packages that you use must be compiled for profiling. The -prof flag to GHCi only works in conjunction with -fexternal-interpreter.

There are three ways to get access to the current call stack.

  • error and undefined automatically attach the current stack to the error message. This often complements the HasCallStack stack (see HasCallStack), so both call stacks are shown.
  • Debug.Trace.traceStack is a version of Debug.Trace.trace that also prints the current call stack.
  • Functions in the module GHC.Stack can be used to get the current stack and render it.

You don’t need to use -fprof-auto for interpreted modules, annotations are automatically added at a granularity fine enough to distinguish individual call sites. However, you won’t see any call stack information for compiled code unless it was compiled with -fprof-auto or has explicit SCC annotations (see Inserting cost centres by hand).

3.5. The GHCi Debugger

GHCi contains a simple imperative-style debugger in which you can stop a running computation in order to examine the values of variables. The debugger is integrated into GHCi, and is turned on by default: no flags are required to enable the debugging facilities. There is one major restriction: breakpoints and single-stepping are only available in interpreted modules; compiled code is invisible to the debugger [5].

The debugger provides the following:

  • The ability to set a breakpoint on a function definition or expression in the program. When the function is called, or the expression evaluated, GHCi suspends execution and returns to the prompt, where you can inspect the values of local variables before continuing with the execution.
  • Execution can be single-stepped: the evaluator will suspend execution approximately after every reduction, allowing local variables to be inspected. This is equivalent to setting a breakpoint at every point in the program.
  • Execution can take place in tracing mode, in which the evaluator remembers each evaluation step as it happens, but doesn’t suspend execution until an actual breakpoint is reached. When this happens, the history of evaluation steps can be inspected.
  • Exceptions (e.g. pattern matching failure and error) can be treated as breakpoints, to help locate the source of an exception in the program.

There is currently no support for obtaining a “stack trace”, but the tracing and history features provide a useful second-best, which will often be enough to establish the context of an error. For instance, it is possible to break automatically when an exception is thrown, even if it is thrown from within compiled code (see Debugging exceptions).

-fbreak-points
Default:enabled for GHCi

This flag’s purpose is to allow disabling breakpoint insertion with the reverse form.

3.5.1. Breakpoints and inspecting variables

Let’s use quicksort as a running example. Here’s the code:

qsort [] = []
qsort (a:as) = qsort left ++ [a] ++ qsort right
  where (left,right) = (filter (<=a) as, filter (>a) as)

main = print (qsort [8, 4, 0, 3, 1, 23, 11, 18])

First, load the module into GHCi:

ghci> :l qsort.hs
[1 of 1] Compiling Main             ( qsort.hs, interpreted )
Ok, modules loaded: Main.
*ghci>

Now, let’s set a breakpoint on the right-hand-side of the second equation of qsort:

*ghci> :break 2
Breakpoint 0 activated at qsort.hs:2:15-46
*ghci>

The command :break 2 sets a breakpoint on line 2 of the most recently-loaded module, in this case qsort.hs. Specifically, it picks the leftmost complete subexpression on that line on which to set the breakpoint, which in this case is the expression (qsort left ++ [a] ++ qsort right).

Now, we run the program:

*ghci> main
Stopped at qsort.hs:2:15-46
_result :: [a]
a :: a
left :: [a]
right :: [a]
[qsort.hs:2:15-46] *ghci>

Execution has stopped at the breakpoint. The prompt has changed to indicate that we are currently stopped at a breakpoint, and the location: [qsort.hs:2:15-46]. To further clarify the location, we can use the :list command:

[qsort.hs:2:15-46] *ghci> :list
1  qsort [] = []
2  qsort (a:as) = qsort left ++ [a] ++ qsort right
3    where (left,right) = (filter (<=a) as, filter (>a) as)

The :list command lists the source code around the current breakpoint. If your output device supports it, then GHCi will highlight the active subexpression in bold.

GHCi has provided bindings for the free variables [6] of the expression on which the breakpoint was placed (a, left, right), and additionally a binding for the result of the expression (_result). These variables are just like other variables that you might define in GHCi; you can use them in expressions that you type at the prompt, you can ask for their types with :type, and so on. There is one important difference though: these variables may only have partial types. For example, if we try to display the value of left:

[qsort.hs:2:15-46] *ghci> left

<interactive>:1:0:
    Ambiguous type variable `a' in the constraint:
      `Show a' arising from a use of `print' at <interactive>:1:0-3
    Cannot resolve unknown runtime types: a
    Use :print or :force to determine these types

This is because qsort is a polymorphic function, and because GHCi does not carry type information at runtime, it cannot determine the runtime types of free variables that involve type variables. Hence, when you ask to display left at the prompt, GHCi can’t figure out which instance of Show to use, so it emits the type error above.

Fortunately, the debugger includes a generic printing command, :print, which can inspect the actual runtime value of a variable and attempt to reconstruct its type. If we try it on left:

[qsort.hs:2:15-46] *ghci> :set -fprint-evld-with-show
[qsort.hs:2:15-46] *ghci> :print left
left = (_t1::[a])

This isn’t particularly enlightening. What happened is that left is bound to an unevaluated computation (a suspension, or thunk), and :print does not force any evaluation. The idea is that :print can be used to inspect values at a breakpoint without any unfortunate side effects. It won’t force any evaluation, which could cause the program to give a different answer than it would normally, and hence it won’t cause any exceptions to be raised, infinite loops, or further breakpoints to be triggered (see Nested breakpoints). Rather than forcing thunks, :print binds each thunk to a fresh variable beginning with an underscore, in this case _t1.

-fprint-evld-with-show

The flag -fprint-evld-with-show instructs :print to reuse available Show instances when possible. This happens only when the contents of the variable being inspected are completely evaluated.

If we aren’t concerned about preserving the evaluatedness of a variable, we can use :force instead of :print. The :force command behaves exactly like :print, except that it forces the evaluation of any thunks it encounters:

[qsort.hs:2:15-46] *ghci> :force left
left = [4,0,3,1]

Now, since :force has inspected the runtime value of left, it has reconstructed its type. We can see the results of this type reconstruction:

[qsort.hs:2:15-46] *ghci> :show bindings
_result :: [Integer]
a :: Integer
left :: [Integer]
right :: [Integer]
_t1 :: [Integer]

Not only do we now know the type of left, but all the other partial types have also been resolved. So we can ask for the value of a, for example:

[qsort.hs:2:15-46] *ghci> a
8

You might find it useful to use Haskell’s seq function to evaluate individual thunks rather than evaluating the whole expression with :force. For example:

[qsort.hs:2:15-46] *ghci> :print right
right = (_t1::[Integer])
[qsort.hs:2:15-46] *ghci> seq _t1 ()
()
[qsort.hs:2:15-46] *ghci> :print right
right = 23 : (_t2::[Integer])

We evaluated only the _t1 thunk, revealing the head of the list, and the tail is another thunk now bound to _t2. The seq function is a little inconvenient to use here, so you might want to use :def to make a nicer interface (left as an exercise for the reader!).

Finally, we can continue the current execution:

[qsort.hs:2:15-46] *ghci> :continue
Stopped at qsort.hs:2:15-46
_result :: [a]
a :: a
left :: [a]
right :: [a]
[qsort.hs:2:15-46] *ghci>

The execution continued at the point it previously stopped, and has now stopped at the breakpoint for a second time.

3.5.1.1. Setting breakpoints

Breakpoints can be set in various ways. Perhaps the easiest way to set a breakpoint is to name a top-level function:

:break identifier

Where ⟨identifier⟩ names any top-level function in an interpreted module currently loaded into GHCi (qualified names may be used). The breakpoint will be set on the body of the function, when it is fully applied. If the function has several patterns, then a breakpoint will be set on each of them.

By using qualified names, one can set breakpoints on all functions (top-level and nested) in every loaded and interpreted module:

:break [ModQual.]topLevelIdent[.nestedIdent]...[.nestedIdent]

⟨ModQual⟩ is optional and is either the effective name of a module or the local alias of a qualified import statement.

⟨topLevelIdent⟩ is the name of a top level function in the module referenced by ⟨ModQual⟩.

⟨nestedIdent⟩ is optional and the name of a function nested in a let or where clause inside the previously mentioned function ⟨nestedIdent⟩ or ⟨topLevelIdent⟩.

If ⟨ModQual⟩ is a module name, then ⟨topLevelIdent⟩ can be any top level identifier in this module. If ⟨ModQual⟩ is missing or a local alias of a qualified import, then ⟨topLevelIdent⟩ must be in scope.

Breakpoints can be set on arbitrarily deeply nested functions, but the whole chain of nested function names must be specified.

Consider the function foo in a module Main:

foo s = 'a' : add s
    where add = (++"z")

The breakpoint on the function add can be set with one of the following commands:

:break Main.foo.add
:break foo.add

Breakpoints can also be set by line (and optionally column) number:

:break line
:break line column
:break module line
:break module line column

When a breakpoint is set on a particular line, GHCi sets the breakpoint on the leftmost subexpression that begins and ends on that line. If two complete subexpressions start at the same column, the longest one is picked. If there is no complete subexpression on the line, then the leftmost expression starting on the line is picked, and failing that the rightmost expression that partially or completely covers the line.

When a breakpoint is set on a particular line and column, GHCi picks the smallest subexpression that encloses that location on which to set the breakpoint. Note: GHC considers the TAB character to have a width of 1, wherever it occurs; in other words it counts characters, rather than columns. This matches what some editors do, and doesn’t match others. The best advice is to avoid tab characters in your source code altogether (see -Wtabs in Warnings and sanity-checking).

If the module is omitted, then the most recently-loaded module is used.

Not all subexpressions are potential breakpoint locations. Single variables are typically not considered to be breakpoint locations (unless the variable is the right-hand-side of a function definition, lambda, or case alternative). The rule of thumb is that all redexes are breakpoint locations, together with the bodies of functions, lambdas, case alternatives and binding statements. There is normally no breakpoint on a let expression, but there will always be a breakpoint on its body, because we are usually interested in inspecting the values of the variables bound by the let.

3.5.1.2. Managing breakpoints

The list of breakpoints currently defined can be displayed using :show breaks:

*ghci> :show breaks
[0] Main qsort.hs:1:11-12 enabled
[1] Main qsort.hs:2:15-46 enabled

To disable one or several defined breakpoint, use the :disable command with one or several blank separated numbers given in the output from :show breaks:. To disable all breakpoints at once, use :disable *.

*ghci> :disable 0
*ghci> :show breaks
[0] Main qsort.hs:1:11-12 disabled
[1] Main qsort.hs:2:15-46 enabled

Disabled breakpoints can be (re-)enabled with the :enable command. The parameters of the :disable and :enable commands are identical.

To delete a breakpoint, use the :delete command with the number given in the output from :show breaks:

*ghci> :delete 0
*ghci> :show breaks
[1] Main qsort.hs:2:15-46 disabled

To delete all breakpoints at once, use :delete *.

3.5.2. Single-stepping

Single-stepping is a great way to visualise the execution of your program, and it is also a useful tool for identifying the source of a bug. GHCi offers two variants of stepping. Use :step to enable all the breakpoints in the program, and execute until the next breakpoint is reached. Use :steplocal to limit the set of enabled breakpoints to those in the current top level function. Similarly, use :stepmodule to single step only on breakpoints contained in the current module. For example:

*ghci> :step main
Stopped at qsort.hs:5:7-47
_result :: IO ()

The command :step expr begins the evaluation of ⟨expr⟩ in single-stepping mode. If ⟨expr⟩ is omitted, then it single-steps from the current breakpoint. :steplocal and :stepmodule commands work similarly.

The :list command is particularly useful when single-stepping, to see where you currently are:

[qsort.hs:5:7-47] *ghci> :list
4
5  main = print (qsort [8, 4, 0, 3, 1, 23, 11, 18])
6
[qsort.hs:5:7-47] *ghci>

In fact, GHCi provides a way to run a command when a breakpoint is hit, so we can make it automatically do :list:

[qsort.hs:5:7-47] *ghci> :set stop :list
[qsort.hs:5:7-47] *ghci> :step
Stopped at qsort.hs:5:14-46
_result :: [Integer]
4
5  main = print (qsort [8, 4, 0, 3, 1, 23, 11, 18])
6
[qsort.hs:5:14-46] *ghci>

3.5.3. Nested breakpoints

When GHCi is stopped at a breakpoint, and an expression entered at the prompt triggers a second breakpoint, the new breakpoint becomes the “current” one, and the old one is saved on a stack. An arbitrary number of breakpoint contexts can be built up in this way. For example:

[qsort.hs:2:15-46] *ghci> :st qsort [1,3]
Stopped at qsort.hs:(1,0)-(3,55)
_result :: [a]
... [qsort.hs:(1,0)-(3,55)] *ghci>

While stopped at the breakpoint on line 2 that we set earlier, we started a new evaluation with :step qsort [1,3]. This new evaluation stopped after one step (at the definition of qsort). The prompt has changed, now prefixed with ..., to indicate that there are saved breakpoints beyond the current one. To see the stack of contexts, use :show context:

... [qsort.hs:(1,0)-(3,55)] *ghci> :show context
--> main
  Stopped at qsort.hs:2:15-46
--> qsort [1,3]
  Stopped at qsort.hs:(1,0)-(3,55)
... [qsort.hs:(1,0)-(3,55)] *ghci>

To abandon the current evaluation, use :abandon:

... [qsort.hs:(1,0)-(3,55)] *ghci> :abandon
[qsort.hs:2:15-46] *ghci> :abandon
*ghci>

3.5.4. The _result variable

When stopped at a breakpoint or single-step, GHCi binds the variable _result to the value of the currently active expression. The value of _result is presumably not available yet, because we stopped its evaluation, but it can be forced: if the type is known and showable, then just entering _result at the prompt will show it. However, there’s one caveat to doing this: evaluating _result will be likely to trigger further breakpoints, starting with the breakpoint we are currently stopped at (if we stopped at a real breakpoint, rather than due to :step). So it will probably be necessary to issue a :continue immediately when evaluating _result. Alternatively, you can use :force which ignores breakpoints.

3.5.5. Tracing and history

A question that we often want to ask when debugging a program is “how did I get here?”. Traditional imperative debuggers usually provide some kind of stack-tracing feature that lets you see the stack of active function calls (sometimes called the “lexical call stack”), describing a path through the code to the current location. Unfortunately this is hard to provide in Haskell, because execution proceeds on a demand-driven basis, rather than a depth-first basis as in strict languages. The “stack“ in GHC’s execution engine bears little resemblance to the lexical call stack. Ideally GHCi would maintain a separate lexical call stack in addition to the dynamic call stack, and in fact this is exactly what our profiling system does (Profiling), and what some other Haskell debuggers do. For the time being, however, GHCi doesn’t maintain a lexical call stack (there are some technical challenges to be overcome). Instead, we provide a way to backtrack from a breakpoint to previous evaluation steps: essentially this is like single-stepping backwards, and should in many cases provide enough information to answer the “how did I get here?” question.

To use tracing, evaluate an expression with the :trace command. For example, if we set a breakpoint on the base case of qsort:

*ghci> :list qsort
1  qsort [] = []
2  qsort (a:as) = qsort left ++ [a] ++ qsort right
3    where (left,right) = (filter (<=a) as, filter (>a) as)
4
*ghci> :b 1
Breakpoint 1 activated at qsort.hs:1:11-12
*ghci>

and then run a small qsort with tracing:

*ghci> :trace qsort [3,2,1]
Stopped at qsort.hs:1:11-12
_result :: [a]
[qsort.hs:1:11-12] *ghci>

We can now inspect the history of evaluation steps:

[qsort.hs:1:11-12] *ghci> :hist
-1  : qsort.hs:3:24-38
-2  : qsort.hs:3:23-55
-3  : qsort.hs:(1,0)-(3,55)
-4  : qsort.hs:2:15-24
-5  : qsort.hs:2:15-46
-6  : qsort.hs:3:24-38
-7  : qsort.hs:3:23-55
-8  : qsort.hs:(1,0)-(3,55)
-9  : qsort.hs:2:15-24
-10 : qsort.hs:2:15-46
-11 : qsort.hs:3:24-38
-12 : qsort.hs:3:23-55
-13 : qsort.hs:(1,0)-(3,55)
-14 : qsort.hs:2:15-24
-15 : qsort.hs:2:15-46
-16 : qsort.hs:(1,0)-(3,55)
<end of history>

To examine one of the steps in the history, use :back:

[qsort.hs:1:11-12] *ghci> :back
Logged breakpoint at qsort.hs:3:24-38
_result :: [a]
as :: [a]
a :: a
[-1: qsort.hs:3:24-38] *ghci>

Note that the local variables at each step in the history have been preserved, and can be examined as usual. Also note that the prompt has changed to indicate that we’re currently examining the first step in the history: -1. The command :forward can be used to traverse forward in the history.

The :trace command can be used with or without an expression. When used without an expression, tracing begins from the current breakpoint, just like :step.

The history is only available when using :trace; the reason for this is we found that logging each breakpoint in the history cuts performance by a factor of 2 or more.

-fghci-hist-size=⟨n⟩
Default:50

Modify the depth of the evaluation history tracked by GHCi.

3.5.6. Debugging exceptions

Another common question that comes up when debugging is “where did this exception come from?”. Exceptions such as those raised by error or head [] have no context information attached to them. Finding which particular call to head in your program resulted in the error can be a painstaking process, usually involving Debug.Trace.trace, or compiling with profiling and using Debug.Trace.traceStack or +RTS -xc (see -xc).

The GHCi debugger offers a way to hopefully shed some light on these errors quickly and without modifying or recompiling the source code. One way would be to set a breakpoint on the location in the source code that throws the exception, and then use :trace and :history to establish the context. However, head is in a library and we can’t set a breakpoint on it directly. For this reason, GHCi provides the flags -fbreak-on-exception which causes the evaluator to stop when an exception is thrown, and -fbreak-on-error, which works similarly but stops only on uncaught exceptions. When stopping at an exception, GHCi will act just as it does when a breakpoint is hit, with the deviation that it will not show you any source code location. Due to this, these commands are only really useful in conjunction with :trace, in order to log the steps leading up to the exception. For example:

*ghci> :set -fbreak-on-exception
*ghci> :trace qsort ("abc" ++ undefined)
“Stopped at <exception thrown>
_exception :: e
[<exception thrown>] *ghci> :hist
-1  : qsort.hs:3:24-38
-2  : qsort.hs:3:23-55
-3  : qsort.hs:(1,0)-(3,55)
-4  : qsort.hs:2:15-24
-5  : qsort.hs:2:15-46
-6  : qsort.hs:(1,0)-(3,55)
<end of history>
[<exception thrown>] *ghci> :back
Logged breakpoint at qsort.hs:3:24-38
_result :: [a]
as :: [a]
a :: a
[-1: qsort.hs:3:24-38] *ghci> :force as
*** Exception: Prelude.undefined
[-1: qsort.hs:3:24-38] *ghci> :print as
as = 'b' : 'c' : (_t1::[Char])

The exception itself is bound to a new variable, _exception.

Breaking on exceptions is particularly useful for finding out what your program was doing when it was in an infinite loop. Just hit Control-C, and examine the history to find out what was going on.

-fbreak-on-exception

Causes GHCi to halt evaluation and return to the interactive prompt in the event of an exception. -fbreak-on-exception breaks on all exceptions.

-fbreak-on-error

Causes GHCi to halt evaluation and return to the interactive prompt in the event of an exception. -fbreak-on-error breaks on only those exceptions which would otherwise be uncaught.

3.5.7. Example: inspecting functions

It is possible to use the debugger to examine function values. When we are at a breakpoint and a function is in scope, the debugger cannot show you the source code for it; however, it is possible to get some information by applying it to some arguments and observing the result.

The process is slightly complicated when the binding is polymorphic. We show the process by means of an example. To keep things simple, we will use the well known map function:

import Prelude hiding (map)

map :: (a->b) -> [a] -> [b]
map f [] = []
map f (x:xs) = f x : map f xs

We set a breakpoint on map, and call it.

*ghci> :break 5
Breakpoint 0 activated at  map.hs:5:15-28
*ghci> map Just [1..5]
Stopped at map.hs:(4,0)-(5,12)
_result :: [b]
x :: a
f :: a -> b
xs :: [a]

GHCi tells us that, among other bindings, f is in scope. However, its type is not fully known yet, and thus it is not possible to apply it to any arguments. Nevertheless, observe that the type of its first argument is the same as the type of x, and its result type is shared with _result.

As we demonstrated earlier (Breakpoints and inspecting variables), the debugger has some intelligence built-in to update the type of f whenever the types of x or _result are discovered. So what we do in this scenario is force x a bit, in order to recover both its type and the argument part of f.

*ghci> seq x ()
*ghci> :print x
x = 1

We can check now that as expected, the type of x has been reconstructed, and with it the type of f has been too:

*ghci> :t x
x :: Integer
*ghci> :t f
f :: Integer -> b

From here, we can apply f to any argument of type Integer and observe the results.

*ghci> let b = f 10
*ghci> :t b
b :: b
*ghci> b
<interactive>:1:0:
    Ambiguous type variable `b' in the constraint:
      `Show b' arising from a use of `print' at <interactive>:1:0
*ghci> :p b
b = (_t2::a)
*ghci> seq b ()
()
*ghci> :t b
b :: a
*ghci> :p b
b = Just 10
*ghci> :t b
b :: Maybe Integer
*ghci> :t f
f :: Integer -> Maybe Integer
*ghci> f 20
Just 20
*ghci> map f [1..5]
[Just 1, Just 2, Just 3, Just 4, Just 5]

In the first application of f, we had to do some more type reconstruction in order to recover the result type of f. But after that, we are free to use f normally.

3.5.8. Limitations

  • When stopped at a breakpoint, if you try to evaluate a variable that is already under evaluation, the second evaluation will hang. The reason is that GHC knows the variable is under evaluation, so the new evaluation just waits for the result before continuing, but of course this isn’t going to happen because the first evaluation is stopped at a breakpoint. Control-C can interrupt the hung evaluation and return to the prompt.

    The most common way this can happen is when you’re evaluating a CAF (e.g. main), stop at a breakpoint, and ask for the value of the CAF at the prompt again.

  • Implicit parameters (see Implicit parameters) are only available at the scope of a breakpoint if there is an explicit type signature.

3.6. Invoking GHCi

GHCi is invoked with the command ghci or ghc --interactive. One or more modules or filenames can also be specified on the command line; this instructs GHCi to load the specified modules or filenames (and all the modules they depend on), just as if you had said :load modules at the GHCi prompt (see GHCi commands). For example, to start GHCi and load the program whose topmost module is in the file Main.hs, we could say:

$ ghci Main.hs

Most of the command-line options accepted by GHC (see Using GHC) also make sense in interactive mode. The ones that don’t make sense are mostly obvious.

-flocal-ghci-history

By default, GHCi keeps global history in $XDG_DATA_HOME/ghc/ghci_history or %APPDATA%/<app>/ghci_history, but you can use current directory, e.g.:

$ ghci -flocal-ghci-history

It will create .ghci-history in current folder where GHCi is launched.

-fghci-leak-check

(Debugging only) When loading new modules with :load, check that any previously loaded modules have been correctly garbage collected. Emits messages if a leak is detected.

3.6.1. Packages

Most packages (see Using Packages) are available without needing to specify any extra flags at all: they will be automatically loaded the first time they are needed.

For hidden packages, however, you need to request the package be loaded by using the -package ⟨pkg⟩ flag:

$ ghci -package readline
GHCi, version 8.y.z: https://www.haskell.org/ghc/  :? for help
Loading package base ... linking ... done.
Loading package readline-1.0 ... linking ... done.
ghci>

The following command works to load new packages into a running GHCi:

ghci> :set -package name

But note that doing this will cause all currently loaded modules to be unloaded, and you’ll be dumped back into the Prelude.

3.6.2. Extra libraries

Extra libraries may be specified on the command line using the normal -llib option. (The term library here refers to libraries of foreign object code; for using libraries of Haskell source code, see Modules vs. filenames.) For example, to load the “m” library:

$ ghci -lm

On systems with .so-style shared libraries, the actual library loaded will the liblib.so. GHCi searches the following places for libraries, in this order:

  • Paths specified using the -L ⟨dir⟩ command-line option,
  • The standard library search path for your system loader, which on some systems may be overridden by setting the LD_LIBRARY_PATH environment variable.
  • The linker standard library search can also be overridden on some systems using the LIBRARY_PATH environment variable. Because of some implementation detail on Windows, setting LIBRARY_PATH will also extend the system loader path for any library it finds. So often setting LIBRARY_PATH is enough.

On systems with .dll-style shared libraries, the actual library loaded will be lib.dll, liblib.dll. GHCi also has full support for import libraries, either Microsoft style .lib, or GNU GCC style .a and .dll.a libraries. If you have an import library it is advisable to always specify the import library instead of the .dll. e.g. use -lgcc` instead of ``-llibgcc_s_seh-1. Again, GHCi will signal an error if it can’t find the library.

GHCi can also load plain object files (.o or .obj depending on your platform) or static archives (.a) from the command-line. Just add the name the object file or library to the command line. On Windows GHCi also supports the big-obj format.

Ordering of -l options matters: a library should be mentioned before the libraries it depends on (see Options affecting linking).

3.7. GHCi commands

GHCi commands all begin with “:” and consist of a single command name followed by zero or more parameters. The command name may be abbreviated, with ambiguities being resolved in favour of the more commonly used commands.

:abandon

Abandons the current evaluation (only available when stopped at a breakpoint).

:add [*]⟨module⟩

Add ⟨module⟩(s) to the current target set, and perform a reload. Normally pre-compiled code for the module will be loaded if available, or otherwise the module will be compiled to byte-code. Using the * prefix forces the module to be loaded as byte-code.

⟨module⟩ may be a file path. A “~” symbol at the beginning of ⟨module⟩ will be replaced by the contents of the environment variable HOME.

:all-types

List all types collected for expressions and (local) bindings currently loaded (while :set +c was active) with their respective source-code span, e.g.

GhciTypes> :all-types
GhciTypes.hs:(38,13)-(38,24): Maybe Id
GhciTypes.hs:(45,10)-(45,29): Outputable SpanInfo
GhciTypes.hs:(45,10)-(45,29): (Rational -> SpanInfo -> SDoc) -> Outputable SpanInfo
:back ⟨n⟩

Travel back ⟨n⟩ steps in the history. ⟨n⟩ is one if omitted. See Tracing and history for more about GHCi’s debugging facilities. See also: :trace, :history, :forward.

:break [⟨identifier⟩ | [⟨module⟩] ⟨line⟩ [⟨column⟩]]

Set a breakpoint on the specified function or line and column. See Setting breakpoints.

:browse[!] [[*] ⟨module⟩]

Displays the identifiers exported by the module ⟨module⟩, which must be either loaded into GHCi or be a member of a package. If ⟨module⟩ is omitted, the most recently-loaded module is used.

Like all other GHCi commands, the output is always displayed in the current GHCi scope (What’s really in scope at the prompt?).

There are two variants of the browse command:

  • If the * symbol is placed before the module name, then all the identifiers in scope in ⟨module⟩ (rather that just its exports) are shown.

    The *-form is only available for modules which are interpreted; for compiled modules (including modules from packages) only the non-* form of :browse is available.

  • Data constructors and class methods are usually displayed in the context of their data type or class declaration. However, if the ! symbol is appended to the command, thus :browse!, they are listed individually. The !-form also annotates the listing with comments giving possible imports for each group of entries. Here is an example:

    ghci> :browse! Data.Maybe
    -- not currently imported
    Data.Maybe.catMaybes :: [Maybe a] -> [a]
    Data.Maybe.fromJust :: Maybe a -> a
    Data.Maybe.fromMaybe :: a -> Maybe a -> a
    Data.Maybe.isJust :: Maybe a -> Bool
    Data.Maybe.isNothing :: Maybe a -> Bool
    Data.Maybe.listToMaybe :: [a] -> Maybe a
    Data.Maybe.mapMaybe :: (a -> Maybe b) -> [a] -> [b]
    Data.Maybe.maybeToList :: Maybe a -> [a]
    -- imported via Prelude
    Just :: a -> Maybe a
    data Maybe a = Nothing | Just a
    Nothing :: Maybe a
    maybe :: b -> (a -> b) -> Maybe a -> b
    

    This output shows that, in the context of the current session (ie in the scope of Prelude), the first group of items from Data.Maybe are not in scope (although they are available in fully qualified form in the GHCi session - see What’s really in scope at the prompt?), whereas the second group of items are in scope (via Prelude) and are therefore available either unqualified, or with a Prelude. qualifier.

:cd ⟨dir⟩

Changes the current working directory to ⟨dir⟩. A “~” symbol at the beginning of ⟨dir⟩ will be replaced by the contents of the environment variable HOME. See also the :show paths command for showing the current working directory.

Note: changing directories causes all currently loaded modules to be unloaded. This is because the search path is usually expressed using relative directories, and changing the search path in the middle of a session is not supported.

:cmd ⟨expr⟩

Executes ⟨expr⟩ as a computation of type IO String, and then executes the resulting string as a list of GHCi commands. Multiple commands are separated by newlines. The :cmd command is useful with :def and :set stop.

:complete ⟨type⟩ [⟨n⟩-][⟨m⟩] ⟨string-literal⟩

This command allows to request command completions from GHCi even when interacting over a pipe instead of a proper terminal and is designed for integrating GHCi’s completion with text editors and IDEs.

When called, :complete prints the ⟨n⟩th to ⟨m⟩th completion candidates for the partial input ⟨string-literal⟩ for the completion domain denoted by ⟨type⟩. Currently, only the repl domain is supported which denotes the kind of completion that would be provided interactively by GHCi at the input prompt.

If omitted, ⟨n⟩ and ⟨m⟩ default to the first or last available completion candidate respectively. If there are less candidates than requested via the range argument, ⟨n⟩ and ⟨m⟩ are implicitly capped to the number of available completion candidates.

The output of :complete begins with a header line containing three space-delimited fields:

  • An integer denoting the number l of printed completions,
  • an integer denoting the total number of completions available, and finally
  • a string literal denoting a common prefix to be added to the returned completion candidates.

The header line is followed by ⟨l⟩ lines each containing one completion candidate encoded as (quoted) string literal. Here are some example invocations showing the various cases:

ghci> :complete repl 0 ""
0 470 ""
ghci> :complete repl 5 "import For"
5 21 "import "
"Foreign"
"Foreign.C"
"Foreign.C.Error"
"Foreign.C.String"
"Foreign.C.Types"
ghci> :complete repl 5-10 "import For"
6 21 "import "
"Foreign.C.Types"
"Foreign.Concurrent"
"Foreign.ForeignPtr"
"Foreign.ForeignPtr.Safe"
"Foreign.ForeignPtr.Unsafe"
"Foreign.Marshal"
ghci> :complete repl 20- "import For"
2 21 "import "
"Foreign.StablePtr"
"Foreign.Storable"
ghci> :complete repl "map"
3 3 ""
"map"
"mapM"
"mapM_"
ghci> :complete repl 5-10 "map"
0 3 ""
:continue [⟨ignoreCount⟩]

Continue the current evaluation, when stopped at a breakpoint.

If an ⟨ignoreCount⟩ is specified, the program will ignore the current breakpoint for the next ⟨ignoreCount⟩ iterations. See command :ignore.

:def[!] ⟨name⟩ ⟨expr⟩

:def is used to define new commands, or macros, in GHCi. The command :def ⟨name⟩ ⟨expr⟩ defines a new GHCi command :name, implemented by the Haskell expression ⟨expr⟩, which must have type String -> IO String. When :name args is typed at the prompt, GHCi will run the expression (name args), take the resulting String, and feed it back into GHCi as a new sequence of commands. Separate commands in the result must be separated by “\n”.

That’s all a little confusing, so here’s a few examples. To start with, here’s a new GHCi command which doesn’t take any arguments or produce any results, it just outputs the current date and time:

ghci> let date _ = Data.Time.getZonedTime >>= print >> return ""
ghci> :def date date
ghci> :date
2017-04-10 12:34:56.93213581 UTC

Here’s an example of a command that takes an argument. It’s a re-implementation of :cd:

ghci> let mycd d = System.Directory.setCurrentDirectory d >> return ""
ghci> :def mycd mycd
ghci> :mycd ..

Or I could define a simple way to invoke “ghc --make Main” in the current directory:

ghci> :def make (\_ -> return ":! ghc --make Main")

We can define a command that reads GHCi input from a file. This might be useful for creating a set of bindings that we want to repeatedly load into the GHCi session:

ghci> :def . readFile
ghci> :. cmds.ghci

Notice that we named the command :., by analogy with the “.” Unix shell command that does the same thing.

Typing :def on its own lists the currently-defined macros. Attempting to redefine an existing command name results in an error unless the :def! form is used, in which case the old command with that name is silently overwritten. However for builtin commands the old command can still be used by preceding the command name with a double colon (eg ::load). It’s not possible to redefine the commands :{, :} and :!.

:delete * | ⟨num⟩ ...

Delete one or more breakpoints by number (use :show breaks to see the number of each breakpoint). The * form deletes all the breakpoints.

:disable * | ⟨num⟩ ...

Disable one or more breakpoints by number (use :show breaks to see the number and state of each breakpoint). The * form disables all the breakpoints.

:doc ⟨name⟩

(Experimental: This command will likely change significantly in GHC 8.8.)

Displays the documentation for the given name. Currently the command is restricted to displaying the documentation directly on the declaration in question, ignoring documentation for arguments, constructors etc.

:edit ⟨file⟩

Opens an editor to edit the file ⟨file⟩, or the most recently loaded module if ⟨file⟩ is omitted. If there were errors during the last loading, the cursor will be positioned at the line of the first error. The editor to invoke is taken from the VISUAL or EDITOR environment variables, or a default editor on your system if neither is not set. You can change the editor using :set editor.

VISUAL
Hidden:
:enable * | ⟨num⟩ ...

Enable one or more disabled breakpoints by number (use :show breaks to see the number and state of each breakpoint). The * form enables all the disabled breakpoints. Enabling a break point will reset its ignore count to 0. (See :ignore)

:force ⟨identifier⟩ ...

Prints the value of ⟨identifier⟩ in the same way as :print. Unlike :print, :force evaluates each thunk that it encounters while traversing the value. This may cause exceptions or infinite loops, or further breakpoints (which are ignored, but displayed).

:forward ⟨n⟩

Move forward ⟨n⟩ steps in the history. ⟨n⟩ is one if omitted. See Tracing and history for more about GHCi’s debugging facilities. See also: :trace, :history, :back.

:help
:?

Displays a list of the available commands.

:

Repeat the previous command.

:history [num]

Display the history of evaluation steps. With a number, displays that many steps (default: 20). For use with :trace; see Tracing and history. To set the number of history entries stored by GHCi, use the -fghci-hist-size=⟨n⟩ flag.

:info[!] ⟨name⟩

Displays information about the given name(s). For example, if ⟨name⟩ is a class, then the class methods and their types will be printed; if ⟨name⟩ is a type constructor, then its definition will be printed; if ⟨name⟩ is a function, then its type will be printed. If ⟨name⟩ has been loaded from a source file, then GHCi will also display the location of its definition in the source.

For types and classes, GHCi also summarises instances that mention them. To avoid showing irrelevant information, an instance is shown only if (a) its head mentions ⟨name⟩, and (b) all the other things mentioned in the instance are in scope (either qualified or otherwise) as a result of a :load or :module commands.

The command :info! works in a similar fashion but it removes restriction (b), showing all instances that are in scope and mention ⟨name⟩ in their head.

:instances ⟨type⟩

Displays all the class instances available to the argument ⟨type⟩. The command will match ⟨type⟩ with the first parameter of every instance and then check that all constraints are satisfiable.

When combined with PartialTypeSignatures, a user can insert wildcards into a query and learn the constraints required of each wildcard for ⟨type⟩ match with an instance.

The output is a listing of all matching instances, simplified and instantiated as much as possible.

For example:

> :instances Maybe (Maybe Int)
instance Eq (Maybe (Maybe Int)) -- Defined in ‘GHC.Maybe’
instance Ord (Maybe (Maybe Int)) -- Defined in ‘GHC.Maybe’
instance Show (Maybe (Maybe Int)) -- Defined in ‘GHC.Show’
instance Read (Maybe (Maybe Int)) -- Defined in ‘GHC.Read’

> :set -XPartialTypeSignatures -fno-warn-partial-type-signatures

> :instances Maybe _
instance Eq _ => Eq (Maybe _) -- Defined in ‘GHC.Maybe’
instance Semigroup _ => Monoid (Maybe _) -- Defined in ‘GHC.Base’
instance Ord _ => Ord (Maybe _) -- Defined in ‘GHC.Maybe’
instance Semigroup _ => Semigroup (Maybe _) -- Defined in ‘GHC.Base’
instance Show _ => Show (Maybe _) -- Defined in ‘GHC.Show’
instance Read _ => Read (Maybe _) -- Defined in ‘GHC.Read’

Only instances which could potentially be used will be displayed in the results. Instances which require unsatisfiable constraints such as TypeError will not be included. In the following example, the instance for A is not shown because it cannot be used.

ghci>:set -XDataKinds -XUndecidableInstances
ghci>import GHC.TypeLits
ghci>class A a
ghci>instance (TypeError (Text "Not possible")) => A Bool
ghci>:instances Bool
instance Eq Bool -- Defined in ‘GHC.Classes’
instance Ord Bool -- Defined in ‘GHC.Classes’
instance Enum Bool -- Defined in ‘GHC.Enum’
instance Show Bool -- Defined in ‘GHC.Show’
instance Read Bool -- Defined in ‘GHC.Read’
instance Bounded Bool -- Defined in ‘GHC.Enum’
:issafe [⟨module⟩]

Displays Safe Haskell information about the given module (or the current module if omitted). This includes the trust type of the module and its containing package.

:ignore ⟨break⟩ ⟨ignoreCount⟩

Set the ignore count of the breakpoint with number ⟨break⟩ to ⟨ignoreCount⟩.

The next ⟨ignoreCount⟩ times the program hits the breakpoint ⟨break⟩, this breakpoint is ignored and the program doesn’t stop. Every time the breakpoint is ignored, the ignore count is decremented by 1. When the ignore count is zero, the program again stops at the break point.

You can also specify an ⟨ignoreCount⟩ on a :continue command when you resume execution of your program.

:kind[!] ⟨type⟩

Infers and prints the kind of ⟨type⟩. The latter can be an arbitrary type expression, including a partial application of a type constructor, such as Either Int. In fact, :kind even allows you to write a partial application of a type synonym (usually disallowed), so that this works:

ghci> type T a b = (a,b,a)
ghci> :k T Int Bool
T Int Bool :: *
ghci> :k T
T :: * -> * -> *
ghci> :k T Int
T Int :: * -> *

Free type variables are also implicitly quantified, same as if you wrote :k forall a. [a] so this also works:

ghci> :k [a]
[a] :: *

If you specify the optional “!”, GHC will in addition normalise the type by expanding out type synonyms and evaluating type-function applications, and display the normalised result.

:list ⟨identifier⟩

Lists the source code around the definition of ⟨identifier⟩ or the current breakpoint if not given. This requires that the identifier be defined in an interpreted module. If your output device supports it, then GHCi will highlight the active subexpression in bold.

:list [⟨module⟩] ⟨line⟩

Lists the source code around the given line number of ⟨module⟩. This requires that the module be interpreted. If your output device supports it, then GHCi will highlight the active subexpression in bold.

:load[!] [*]⟨module⟩

Recursively loads the specified ⟨module⟩s, and all the modules they depend on. Here, each ⟨module⟩ must be a module name or filename, but may not be the name of a module in a package.

All previously loaded modules, except package modules, are forgotten. The new set of modules is known as the target set. Note that :load can be used without any arguments to unload all the currently loaded modules and bindings.

Normally pre-compiled code for a module will be loaded if available, or otherwise the module will be compiled to byte-code. Using the * prefix forces a module to be loaded as byte-code.

Adding the optional “!” turns type errors into warnings while loading. This allows to use the portions of the module that are correct, even if there are type errors in some definitions. Effectively, the -fdefer-type-errors flag is set before loading and unset after loading if the flag has not already been set before. See Deferring type errors to runtime for further motivation and details.

After a :load command, the current context is set to:

  • ⟨module⟩, if it was loaded successfully, or
  • the most recently successfully loaded module, if any other modules were loaded as a result of the current :load, or
  • Prelude otherwise.
:loc-at ⟨module⟩ ⟨line⟩ ⟨col⟩ ⟨end-line⟩ ⟨end-col⟩ [⟨name⟩]

Tries to find the definition site of the name at the given source-code span, e.g.:

X> :loc-at X.hs 6 14 6 16 mu
X.hs:(8,7)-(8,9)

This command is useful when integrating GHCi with text editors and IDEs for providing a goto-definition facility.

The :loc-at command requires :set +c to be set.

:main ⟨arg1⟩ ... ⟨argn⟩

When a program is compiled and executed, it can use the getArgs IO action to access the command-line arguments. However, we cannot simply pass the arguments to main while we are testing in ghci, as main doesn’t take its arguments directly.

Instead, we can use the :main command. This runs whatever main is in scope, with any arguments being treated the same as command-line arguments, e.g.:

ghci> main = System.Environment.getArgs >>= print
ghci> :main foo bar
["foo","bar"]

We can also quote arguments which contains characters like spaces, and they are treated like Haskell strings, or we can just use Haskell list syntax:

ghci> :main foo "bar baz"
["foo","bar baz"]
ghci> :main ["foo", "bar baz"]
["foo","bar baz"]

Finally, other IO actions can be called, either with the -main-is flag or the :run command:

ghci> foo = putStrLn "foo" >> System.Environment.getArgs >>= print
ghci> bar = putStrLn "bar" >> System.Environment.getArgs >>= print
ghci> :set -main-is foo
ghci> :main foo "bar baz"
foo
["foo","bar baz"]
ghci> :run bar ["foo", "bar baz"]
bar
["foo","bar baz"]
:module +|- [*]⟨mod1⟩ ...
import ⟨mod⟩

Sets or modifies the current context for statements typed at the prompt. The form import mod is equivalent to :module +mod. See What’s really in scope at the prompt? for more details.

:print ⟨names⟩

Prints a value without forcing its evaluation. :print may be used on values whose types are unknown or partially known, which might be the case for local variables with polymorphic types at a breakpoint. While inspecting the runtime value, :print attempts to reconstruct the type of the value, and will elaborate the type in GHCi’s environment if possible. If any unevaluated components (thunks) are encountered, then :print binds a fresh variable with a name beginning with _t to each thunk. See Breakpoints and inspecting variables for more information. See also the :sprint command, which works like :print but does not bind new variables.

:quit

Quits GHCi. You can also quit by typing Control-D at the prompt.

:reload[!]

Attempts to reload the current target set (see :load) if any of the modules in the set, or any dependent module, has changed. Note that this may entail loading new modules, or dropping modules which are no longer indirectly required by the target.

Adding the optional “!” turns type errors into warnings while loading. This allows to use the portions of the module that are correct, even if there are type errors in some definitions. Effectively, the -fdefer-type-errors flag is set before loading and unset after loading if the flag has not already been set before. See Deferring type errors to runtime for further motivation and details.

:run

See :main.

:script [⟨n⟩] ⟨filename⟩

Executes the lines of a file as a series of GHCi commands. The syntax for file-name arguments respects shell quoting rules, i.e., file names containing spaces can be enclosed in double quotes or with spaces escaped with a backslash. This command is compatible with multiline statements as set by :set +m

:set [⟨option⟩ ...]

Sets various options. See The :set and :seti commands for a list of available options and Interactive-mode options for a list of GHCi-specific flags. The :set command by itself shows which options are currently set. It also lists the current dynamic flag settings, with GHCi-specific flags listed separately.

:set args ⟨arg⟩

Sets the list of arguments which are returned when the program calls System.Environment.getArgs.

:set editor ⟨cmd⟩

Sets the command used by :edit to ⟨cmd⟩.

:set local-config ⟨source|ignore⟩

If ignore, ./.ghci files will be ignored (sourcing untrusted local scripts is a security risk). The default is source. Set this directive in your user .ghci script, i.e. before the local script would be sourced.

Even when set to ignore, a local script will still be processed if given by -ghci-script on the command line, or sourced via :script.

:set prog ⟨prog⟩

Sets the string to be returned when the program calls System.Environment.getProgName.

:set prompt ⟨prompt⟩

Sets the string to be used as the prompt in GHCi. Inside ⟨prompt⟩, the next sequences are replaced:

  • %s by the names of the modules currently in scope.
  • %l by the line number (as referenced in compiler messages) of the current prompt.
  • %d by the date in “Weekday Month Date” format (e.g., “Tue May 26”) .
  • %t by the current time in 24-hour HH:MM:SS format.
  • %T by the current time in 12-hour HH:MM:SS format.
  • %@ by the current time in 12-hour am/pm format.
  • %A by the current time in 24-hour HH:MM format.
  • %u by the username of the current user.
  • %w by the current working directory.
  • %o by the operating system.
  • %a by the machine architecture.
  • %N by the compiler name.
  • %V by the compiler version.
  • %call(cmd [args]) by the result of calling cmd args.
  • %% by %.

If ⟨prompt⟩ starts with " then it is parsed as a Haskell String; otherwise it is treated as a literal string.

:set prompt-cont ⟨prompt⟩

Sets the string to be used as the continuation prompt (used when using the :{ command) in GHCi.

:set prompt-function ⟨prompt-function⟩

Sets the function to be used for the prompt displaying in GHCi. The function should be of the type [String] -> Int -> IO String. This function is called each time the prompt is being made. The first argument stands for the names of the modules currently in scope(the name of the “topmost” module will begin with a *; see What’s really in scope at the prompt? for more information). The second arguments is the line number (as referenced in compiler messages) of the current prompt.

:set prompt-cont-function ⟨prompt-function⟩

Sets the function to be used for the continuation prompt (used when using the :{ command) displaying in GHCi.

:set stop ⟨num⟩ ⟨cmd⟩

Set a command to be executed when a breakpoint is hit, or a new item in the history is selected. The most common use of :set stop is to display the source code at the current location, e.g. :set stop :list.

If a number is given before the command, then the commands are run when the specified breakpoint (only) is hit. This can be quite useful: for example, :set stop 1 :continue effectively disables breakpoint 1, by running :continue whenever it is hit In this case GHCi will still emit a message to say the breakpoint was hit. If you don’t want such a message, you can use the :disable command. What’s more, with cunning use of :def and :cmd you can use :set stop to implement conditional breakpoints:

*ghci> :def cond \expr -> return (":cmd if (" ++ expr ++ ") then return \"\" else return \":continue\"")
*ghci> :set stop 0 :cond (x < 3)

To ignore breakpoints for a specified number of iterations use the :ignore or the ⟨ignoreCount⟩ parameter of the :continue command.

:seti [⟨option⟩ ...]

Like :set, but options set with :seti affect only expressions and commands typed at the prompt, and not modules loaded with :load (in contrast, options set with :set apply everywhere). See Setting options for interactive evaluation only.

Without any arguments, displays the current set of options that are applied to expressions and commands typed at the prompt.

:show bindings

Show the bindings made at the prompt and their types.

:show breaks

List the active breakpoints.

:show context

List the active evaluations that are stopped at breakpoints.

:show imports

Show the imports that are currently in force, as created by import and :module commands.

:show modules

Show the list of modules currently loaded.

:show packages

Show the currently active package flags, as well as the list of packages currently loaded.

:show paths

Show the current working directory (as set via :cd command), as well as the list of directories searched for source files (as set by the -i option).

:show language

Show the currently active language flags for source files.

:showi language

Show the currently active language flags for expressions typed at the prompt (see also :seti).

:show targets

Show list of currently loaded modules. This set of loaded modules can be modified by :load, :reload, :add and :unadd.

:show [args|prog|prompt|editor|stop]

Displays the specified setting (see :set).

:sprint ⟨expr⟩

Prints a value without forcing its evaluation. :sprint is similar to :print, with the difference that unevaluated subterms are not bound to new variables, they are simply denoted by _.

:step [⟨expr⟩]

Enable all breakpoints and begin evaluating an expression in single-stepping mode. In this mode evaluation will be stopped after every reduction, allowing local variables to be inspected. If ⟨expr⟩ is not given, evaluation will resume at the last breakpoint. See Single-stepping.

:steplocal

Enable only breakpoints in the current top-level binding and resume evaluation at the last breakpoint. Continuation with :steplocal is not possible if this last breakpoint was hit by an error (-fbreak-on-error) or an exception (-fbreak-on-exception).

:stepmodule

Enable only breakpoints in the current module and resume evaluation at the last breakpoint.

:trace ⟨expr⟩

Evaluates the given expression (or from the last breakpoint if no expression is given), and additionally logs the evaluation steps for later inspection using :history. See Tracing and history.

:type ⟨expression⟩

Infers and prints the type of ⟨expression⟩, solving constraints and reducing type families as much as possible. For polymorphic types, it does not instantiate any forall quantified variables.

*X> :type length
length :: Foldable t => t a -> Int

Type family reduction is skipped if the function is not fully instantiated, as this has been observed to give more intuitive results. You may want to use :info if you are not applying any arguments, as that will return the original type of the function.

:type +d ⟨expression⟩

Infers and prints the type of ⟨expression⟩, instantiating all the forall quantifiers, solving constraints, defaulting, and generalising. In this mode, if the inferred type is constrained by any interactive class (Num, Show, Eq, Ord, Foldable, or Traversable), the constrained type variable(s) are defaulted according to the rules described under ExtendedDefaultRules. This mode is quite useful when the inferred type is quite general (such as for foldr) and it may be helpful to see a more concrete instantiation.

*X> :type +d length
length :: [a] -> Int
:type-at ⟨path⟩ ⟨line⟩ ⟨col⟩ ⟨end-line⟩ ⟨end-col⟩ [⟨name⟩]

Reports the inferred type at the given span/position in the module, e.g.:

*X> :type-at X.hs 6 6 6 7 f
Int -> Int

This command is useful when integrating GHCi with text editors and IDEs for providing a show-type-under-point facility.

The first parameter (path) must be a file path and not a module name. The type of this path is dependent on how the module was loaded into GHCi: If the module was loaded by name, then the path name calculated by GHCi as described in Modules vs. filenames must be used. If the module was loaded with an absolute or a relative path, then the same path must be specified.

The last string parameter is useful for when the span is out of date, i.e. the file changed and the code has moved. In which case :type-at falls back to a general :type like lookup.

The :type-at command requires :set +c to be set.

:unadd ⟨module⟩

Removes ⟨module⟩(s) from the current target set, and perform a reload (see :add above).

:undef ⟨name⟩

Undefines the user-defined command ⟨name⟩ (see :def above).

:unset ⟨option⟩

Unsets certain options. See The :set and :seti commands for a list of available options.

:uses ⟨module⟩ ⟨line⟩ ⟨col⟩ ⟨end-line⟩ ⟨end-col⟩ [⟨name⟩]

Reports all module-local uses of the thing at the given position in the module, e.g.:

:uses GhciFind.hs 53 66 53 70 name
GhciFind.hs:(46,25)-(46,29)
GhciFind.hs:(47,37)-(47,41)
GhciFind.hs:(53,66)-(53,70)
GhciFind.hs:(57,62)-(57,66)

This command is useful for highlighting and navigating all uses of an identifier in editors and IDEs.

The :uses command requires :set +c to be set.

:: ⟨builtin-command⟩

Executes the GHCi built-in command (e.g. ::type 3). That is, look up on the list of builtin commands, excluding defined macros. See also: :def.

:! ⟨command⟩

Executes the shell command ⟨command⟩.

3.8. The :set and :seti commands

The :set command sets two types of options: GHCi options, which begin with “+”, and “command-line” options, which begin with “-“.

Note

At the moment, the :set command doesn’t support any kind of quoting in its arguments: quotes will not be removed and cannot be used to group words together. For example, :set -DFOO='BAR BAZ' will not do what you expect.

3.8.1. GHCi options

GHCi options may be set using :set and unset using :unset.

The available GHCi options are:

:set +c

Collect type and location information after loading modules. The commands :all-types, :loc-at, :type-at, and :uses require +c to be active.

:set +m

Enable parsing of multiline commands. A multiline command is prompted for when the current input line contains open layout contexts (see Multiline input).

:set +r

Normally, any evaluation of top-level expressions (otherwise known as CAFs or Constant Applicative Forms) in loaded modules is retained between evaluations. Turning on +r causes all evaluation of top-level expressions to be discarded after each evaluation (they are still retained during a single evaluation).

This option may help if the evaluated top-level expressions are consuming large amounts of space, or if you need repeatable performance measurements.

:set +s

Display some stats after evaluating each expression, including the elapsed time and number of bytes allocated. NOTE: the allocation figure is only accurate to the size of the storage manager’s allocation area, because it is calculated at every GC. Hence, you might see values of zero if no GC has occurred.

:set +t

Display the type of each variable bound after a statement is entered at the prompt. If the statement is a single expression, then the only variable binding will be for the variable it.

3.8.2. Setting GHC command-line options in GHCi

Normal GHC command-line options may also be set using :set. For example, to turn on -Wmissing-signatures, you would say:

ghci> :set -Wmissing-signatures

GHCi will also accept any file-header pragmas it finds, such as {-# OPTIONS_GHC ... #-} and {-# LANGUAGE ... #-} (see Pragmas). For example, instead of using :set to enable -Wmissing-signatures, you could instead write:

ghci> {-# OPTIONS_GHC -Wmissing-signatures #-}

Any GHC command-line option that is designated as dynamic (see the table in Flag reference), may be set using :set. To unset an option, you can set the reverse option:

ghci> :set -Wno-incomplete-patterns -XNoMultiParamTypeClasses

Flag reference lists the reverse for each option where applicable.

Certain static options (-package ⟨pkg⟩, -I⟨dir⟩, -i⟨dir⟩[:⟨dir⟩]*, and -l ⟨lib⟩ in particular) will also work, but some may not take effect until the next reload.

3.8.3. Setting options for interactive evaluation only

GHCi actually maintains two sets of options:

  • The loading options apply when loading modules
  • The interactive options apply when evaluating expressions and commands typed at the GHCi prompt.

The :set command modifies both, but there is also a :seti command (for “set interactive”) that affects only the interactive options set.

It is often useful to change the interactive options, without having that option apply to loaded modules too. For example

:seti -XMonoLocalBinds

It would be undesirable if MonoLocalBinds were to apply to loaded modules too: that might cause a compilation error, but more commonly it will cause extra recompilation, because GHC will think that it needs to recompile the module because the flags have changed.

If you are setting language options in your .ghci file, it is good practice to use :seti rather than :set, unless you really do want them to apply to all modules you load in GHCi.

The two sets of options can be inspected using the :set and :seti commands respectively, with no arguments. For example, in a clean GHCi session we might see something like this:

ghci> :seti
base language is: GHC2021
with the following modifiers:
  -XExtendedDefaultRules
  -XNoMonomorphismRestriction
GHCi-specific dynamic flag settings:
other dynamic, non-language, flag settings:
  -fexternal-dynamic-refs
  -fignore-optim-changes
  -fignore-hpc-changes
  -fimplicit-import-qualified
warning settings:

The two sets of options are initialised as follows. First, both sets of options are initialised as described in The .ghci and .haskeline files. Then the interactive options are modified as follows:

3.9. The .ghci and .haskeline files

3.9.1. The .ghci files

When it starts, unless the -ignore-dot-ghci flag is given, GHCi reads and executes commands from the following files, in this order, if they exist:

  1. ghcappdata/ghci.conf, where ⟨ghcappdata⟩ depends on your system, but is usually something like $HOME/.ghc or $XDG_CONFIG_HOME/ghc on Unix or C:\Users{username}\AppData\Roaming\ghc on Windows.
  2. ./.ghci

The ghci.conf file is most useful for turning on favourite options (e.g. :set +s), and defining useful macros.

Note

When setting language options in this file it is usually desirable to use :seti rather than :set (see Setting options for interactive evaluation only).

Placing a .ghci file in a directory with a Haskell project is a useful way to set certain project-wide options so you don’t have to type them every time you start GHCi: eg. if your project uses multi-parameter type classes, scoped type variables, and CPP, and has source files in three subdirectories A, B and C, you might put the following lines in .ghci:

:set -XMultiParamTypeClasses -XScopedTypeVariables -cpp
:set -iA:B:C

(Note that strictly speaking the -i flag is a static one, but in fact it works to set it using :set like this. The changes won’t take effect until the next :load, though.)

Warning

Sourcing untrusted ./.ghci files is a security risk. They can contain arbitrary commands that will be executed as the user. Use :set local-config to inhibit the processing of ./.ghci files.

Once you have a library of GHCi macros, you may want to source them from separate files, or you may want to source your .ghci file into your running GHCi session while debugging it

:def source readFile

With this macro defined in your .ghci file, you can use :source file to read GHCi commands from file. You can find (and contribute!-) other suggestions for .ghci files on this Haskell wiki page: GHC/GHCi

Additionally, any files specified with -ghci-script flags will be read after the standard files, allowing the use of custom .ghci files.

Two command-line options control whether the startup files files are read:

-ignore-dot-ghci

Don’t read either ./.ghci or the other startup files when starting up.

-ghci-script

Read a specific file after the usual startup files. May be specified repeatedly for multiple inputs. -ignore-dot-ghci does not apply to these files.

When defining GHCi macros, there is some important behavior you should be aware of when names may conflict with built-in commands, especially regarding tab completion.

For example, consider if you had a macro named :time and in the shell, typed :t 3 — what should happen? The current algorithm we use for completing commands is:

  1. First, look up an exact match on the name from the defined macros.
  2. Look for the exact match on the name in the built-in command list.
  3. Do a prefix lookup on the list of built-in commands - if a built-in command matches, but a macro is defined with the same name as the built-in defined, pick the macro.
  4. Do a prefix lookup on the list of built-in commands.
  5. Do a prefix lookup on the list of defined macros.

Here are some examples:

  1. You have a macro :time and enter :t 3

    You get :type 3

  2. You have a macro :type and enter :t 3

    You get :type 3 with your defined macro, not the builtin.

  3. You have a macro :time and a macro :type, and enter :t 3

    You get :type 3 with your defined macro.

When giving priority to built-in commands, you can use :: ⟨builtin-command⟩, like ::type 3.

3.9.2. The .haskeline file

GHCi uses Haskeline under the hood. You can configure it to, among other things, prune duplicates from GHCi history. See: Haskeline user preferences.

3.10. Compiling to object code inside GHCi

By default, GHCi compiles Haskell source code into byte-code that is interpreted by the runtime system. GHCi can also compile Haskell code to object code: to turn on this feature, use the -fobject-code flag either on the command line or with :set (the option -fbyte-code restores byte-code compilation again). Compiling to object code takes longer, but typically the code will execute 10-20 times faster than byte-code.

Compiling to object code inside GHCi is particularly useful if you are developing a compiled application, because the :reload command typically runs much faster than restarting GHC with --make from the command-line, because all the interface files are already cached in memory.

There are disadvantages to compiling to object-code: you can’t set breakpoints in object-code modules, for example. Only the exports of an object-code module will be visible in GHCi, rather than all top-level bindings as in interpreted modules.

3.11. Running the interpreter in a separate process

Normally GHCi runs the interpreted code in the same process as GHC itself, on top of the same RTS and sharing the same heap. However, if the flag -fexternal-interpreter is given, then GHC will spawn a separate process for running interpreted code, and communicate with it using messages over a pipe.

-fexternal-interpreter
Since:8.0.1

Run interpreted code (for GHCi, Template Haskell, Quasi-quoting, or Annotations) in a separate process. The interpreter will run in profiling mode if -prof is in effect, and in dynamically-linked mode if -dynamic is in effect.

There are a couple of caveats that will hopefully be removed in the future: this option is currently not implemented on Windows (it is a no-op), and the external interpreter does not support the GHCi debugger, so breakpoints and single-stepping don’t work with -fexternal-interpreter.

See also the -pgmi ⟨cmd⟩ (Replacing the program for one or more phases) and -opti ⟨option⟩ (Forcing options to a particular phase) flags.

Why might we want to do this? The main reason is that the RTS running the interpreted code can be a different flavour (profiling or dynamically-linked) from GHC itself. So for example:

This feature is experimental in GHC 8.0.x, but it may become the default in future releases.

3.12. Running the interpreter on a different host

When using the flag -fexternal-interpreter GHC will spawn and communicate with the separate process using pipes. There are scenarios (e.g. when cross compiling) where it is favourable to have the communication happen over the network. GHC provides two utilities for this, which can be found in the utils directory.

  • remote-iserv needs to be built with the cross compiler to be executed on the remote host. Or in the case of using it on the same host the stage2 compiler will do as well.
  • iserv-proxy needs to be built on the build machine by the build compiler.

After starting remote-iserv ⟨tmp_dir⟩ ⟨port⟩ on the target and providing it with a temporary folder (where it will copy the necessary libraries to load to) and port it will listen for the proxy to connect.

Providing -pgmi ⟨/path/to/iserv-proxy⟩ and -opti ⟨slave-ip⟩ -opti ⟨slave-port⟩ [-opti -v] in addition to -fexternal-interpreter will then make ghc go through the proxy instead.

There are some limitations when using this. File and process IO will be executed on the target. As such packages like git-embed, file-embed and others might not behave as expected if the target and host do not share the same filesystem.

3.13. Building GHCi libraries

When invoked in the static way, GHCi will use the GHC RTS’s static runtime linker to load object files for imported modules when available. However, when these modules are built with -split-sections this linking can be quite expensive. To reduce this cost, package managers and build systems may opt to produce a pre-linked GHCi object using the --merge-objs mode. This merges the per-module objects into a single object, collapsing function sections into a single text section which can be efficiently loaded by the runtime linker.

3.14. FAQ and Things To Watch Out For

The interpreter can’t load modules with foreign export declarations!
Unfortunately not. We haven’t implemented it yet. Please compile any offending modules by hand before loading them into GHCi.

-O is ineffective in GHCi!

Before GHC 9.8, optimizations were considered too unstable to be used with the bytecode interpreter. This restriction has been lifted, but is still regarded as experimental and guarded by -funoptimized-core-for-interpreter, which is enabled by default. In order to use optimizations, run:

ghci -fno-unoptimized-core-for-interpreter -O
Concurrent threads don’t carry on running when GHCi is waiting for input.
This should work, as long as your GHCi was built with the -threaded switch, which is the default. Consult whoever supplied your GHCi installation.
After using getContents, I can’t use stdin, until I do :load or :reload

This is the defined behaviour of getContents: it puts the stdin Handle in a state known as semi-closed, wherein any further I/O operations on it are forbidden. Because I/O state is retained between computations, the semi-closed state persists until the next :load or :reload command.

You can make stdin reset itself after every evaluation by giving GHCi the command :set +r. This works because stdin is just a top-level expression that can be reverted to its unevaluated state in the same way as any other top-level expression (CAF).

I can’t use Control-C to interrupt computations in GHCi on Windows.
See Running GHCi on Windows.
The default buffering mode is different in GHCi to GHC.

In GHC, the stdout handle is line-buffered by default. However, in GHCi we turn off the buffering on stdout, because this is normally what you want in an interpreter: output appears as it is generated.

If you want line-buffered behaviour, as in GHC, you can start your program thus:

main = do { hSetBuffering stdout LineBuffering; ... }
[5]Note that packages only contain compiled code, so debugging a package requires finding its source and loading that directly.
[6]We originally provided bindings for all variables in scope, rather than just the free variables of the expression, but found that this affected performance considerably, hence the current restriction to just the free variables.