HORIZON HASKELLDocslts/ghc-9.10.x248f8f02026-10-05Search names, modules, packages, or :: a typeCtrl K

GHC 9.10.3 · lts/ghc-9.10.x · 248f8f0 · 2026-10-05

Modulemegaparsec-9.7.0Haskell2010

Text.Megaparsec.Stream

Megaparsec's input stream facilities.

You probably do not want to import this module directly because Text.Megaparsec re-exports it anyway.

  • 2 types
  • 3 classes
  • Packagemegaparsec-9.7.0
  • Exports5
  • LanguageHaskell2010
  • LicenceBSD-2-Clause
  • SourceStream.hs
classclass (Ord (Token s), Ord (Tokens s)) => Stream s where
#

Type class for inputs that can be consumed by the library.

Note that the Stream instances for Text and ByteString (strict and lazy) default to "input sharing" (see ShareInput, NoShareInput). We plan to move away from input sharing in a future major release; if you want to retain the current behaviour and are concerned with maximum performance you should consider using the ShareInput wrapper explicitly.

Note: before the version 9.0.0 the class included the methods from VisualStream and TraversableStream.

Associated types

  • type family Token s

    Type of token in the stream.

  • type family Tokens s

    Type of “chunk” of the stream.

Methods

  • tokenToChunk :: Proxy s -> Token s -> Tokens s

    Lift a single token to chunk of the stream. The default implementation is:

    tokenToChunk pxy = tokensToChunk pxy . pure

    However for some types of stream there may be a more efficient way to lift.

  • tokensToChunk :: Proxy s -> [Token s] -> Tokens s

    The first method that establishes isomorphism between list of tokens and chunk of the stream. Valid implementation should satisfy:

    chunkToTokens pxy (tokensToChunk pxy ts) == ts
  • chunkToTokens :: Proxy s -> Tokens s -> [Token s]

    The second method that establishes isomorphism between list of tokens and chunk of the stream. Valid implementation should satisfy:

    tokensToChunk pxy (chunkToTokens pxy chunk) == chunk
  • chunkLength :: Proxy s -> Tokens s -> Int

    Return length of a chunk of the stream.

  • chunkEmpty :: Proxy s -> Tokens s -> Bool

    Check if a chunk of the stream is empty. The default implementation is in terms of the more general chunkLength:

    chunkEmpty pxy ts = chunkLength pxy ts <= 0

    However for many streams there may be a more efficient implementation.

  • take1_ :: s -> Maybe (Token s, s)

    Extract a single token form the stream. Return Nothing if the stream is empty.

  • takeN_ :: Int -> s -> Maybe (Tokens s, s)

    takeN_ n s should try to extract a chunk of length n, or if the stream is too short, the rest of the stream. Valid implementation should follow the rules:

    • If the requested length n is 0 (or less), Nothing should never be returned, instead Just ("", s) should be returned, where "" stands for the empty chunk, and s is the original stream (second argument).

    • If the requested length is greater than 0 and the stream is empty, Nothing should be returned indicating end of input.

    • In other cases, take chunk of length n (or shorter if the stream is not long enough) from the input stream and return the chunk along with the rest of the stream.

  • takeWhile_ :: (Token s -> Bool) -> s -> (Tokens s, s)

    Extract chunk of the stream taking tokens while the supplied predicate returns True. Return the chunk and the rest of the stream.

    For many types of streams, the method allows for significant performance improvements, although it is not strictly necessary from conceptual point of view.

Instances14Stream, …
newtypenewtype ShareInput a
#

This wrapper selects the input-sharing Stream implementation for Text (Text) and ByteString (ByteString). By input sharing we mean that our parsers will use slices whenever possible to avoid having to copy parts of the input. See also the documentation of split.

Note that using slices is in general faster than copying; on the other hand it also has the potential for causing surprising memory leaks: if any slice of the input survives in the output, holding on to the output will force the entire input Text/ByteString to stay in memory! Even when using lazy Text/ByteString we will hold on to whole chunks at a time leading to to significantly worse memory residency in some cases.

See NoShareInput for a somewhat slower implementation that avoids this memory leak scenario.

Constructors

Instances12Stream, Token, Tokens, …
newtypenewtype NoShareInput a
#

This wrapper selects the no-input-sharing Stream implementation for Text (Text) and ByteString (ByteString). This means that our parsers will create independent copies rather than using slices of the input. See also the documentation of copy.

More importantly, any parser output will be independent of the input, and holding on to parts of the output will never prevent the input from being garbage collected.

For maximum performance you might consider using ShareInput instead, but beware of its pitfalls!

Constructors

Instances12Stream, Token, Tokens, …
classclass Stream s => VisualStream s where
#

Type class for inputs that can also be used for debugging.

Methods

  • showTokens :: Proxy s -> NonEmpty (Token s) -> String

    Pretty-print non-empty stream of tokens. This function is also used to print single tokens (represented as singleton lists).

  • tokensLength :: Proxy s -> NonEmpty (Token s) -> Int

    Return the number of characters that a non-empty stream of tokens spans. The default implementation is sufficient if every token spans exactly 1 character.

Instances5VisualStream
classclass Stream s => TraversableStream s where
#

Type class for inputs that can also be used for error reporting.

Methods

  • reachOffset :: Int -> PosState s -> (Maybe String, PosState s)

    Given an offset o and initial PosState, adjust the state in such a way that it starts at the offset.

    Return two values (in order):

    • Maybe String representing the line on which the given offset o is located. It can be omitted (i.e. Nothing); in that case error reporting functions will not show offending lines. If returned, the line should satisfy a number of conditions that are described below.

    • The updated PosState which can be in turn used to locate another offset o' given that o' >= o.

    The String representing the offending line in input stream should satisfy the following:

    • It should adequately represent location of token at the offset of interest, that is, character at sourceColumn of the returned SourcePos should correspond to the token at the offset o.

    • It should not include the newline at the end.

    • It should not be empty, if the line happens to be empty, it should be replaced with the string "<empty line>".

    • Tab characters should be replaced by appropriate number of spaces, which is determined by the pstateTabWidth field of PosState.

    Note: type signature of the function was changed in the version 9.0.0.

  • reachOffsetNoLine :: Int -> PosState s -> PosState s

    A version of reachOffset that may be faster because it doesn't need to fetch the line at which the given offset in located.

    The default implementation is this:

    reachOffsetNoLine o pst =
  snd (reachOffset o pst)

    Note: type signature of the function was changed in the version 8.0.0.

Instances5TraversableStream