{-
(c) The University of Glasgow 2006
(c) The GRASP/AQUA Project, Glasgow University, 1992-1998


@Uniques@ are used to distinguish entities in the compiler (@Ids@,
@Classes@, etc.) from each other.  Thus, @Uniques@ are the basic
comparison key in the compiler.

If there is any single operation that needs to be fast, it is @Unique@

comparison.  Unsurprisingly, there is quite a bit of huff-and-puff
directed to that end.

Some of the other hair in this code is to be able to use a
``splittable @UniqueSupply@'' if requested/possible (not standard
Haskell).
-}

{-# LANGUAGE CPP #-}
{-# LANGUAGE BangPatterns, MagicHash #-}

module GHC.Types.Unique (
        -- * Main data types
        Unique, Uniquable(..),
        uNIQUE_BITS,

        -- ** Constructors, destructors and operations on 'Unique's
        hasKey,

        pprUniqueAlways,

        mkTag,
        mkUniqueGrimily,
        mkUniqueIntGrimily,
        getKey,
        mkUnique, unpkUnique,
        mkUniqueInt,
        eqUnique, ltUnique,
        incrUnique, stepUnique,

        newTagUnique,
        nonDetCmpUnique,
        isValidKnownKeyUnique,

        -- ** Local uniques
        -- | These are exposed exclusively for use by 'GHC.Types.Var.Env.uniqAway', which
        -- has rather peculiar needs. See Note [Local uniques].
        mkLocalUnique, minLocalUnique, maxLocalUnique,
    ) where

#include "Unique.h"

import GHC.Prelude

import GHC.Data.FastString
import GHC.Utils.Outputable
import GHC.Utils.Word64 (intToWord64, word64ToInt)

-- just for implementing a fast [0,61) -> Char function
import GHC.Exts (indexCharOffAddr#, Char(..), Int(..))

import GHC.Word         ( Word64 )
import Data.Char        ( chr, ord )

import Language.Haskell.Syntax.Module.Name

{-
************************************************************************
*                                                                      *
\subsection[Unique-type]{@Unique@ type and operations}
*                                                                      *
************************************************************************

Note [Uniques and tags]
~~~~~~~~~~~~~~~~~~~~~~~~
A `Unique` in GHC is a 64 bit value composed of two pieces:
* A "tag", of width `UNIQUE_TAG_BITS`, in the high order bits
* A number, of width `uNIQUE_BITS`, which fills up the remainder of the Word64

The tag is typically an ASCII character.  It is typically used to make it easier
to distinguish uniques constructed by different parts of the compiler.
There is a (potentially incomplete) list of unique tags used given in
GHC.Builtin.Uniques. See Note [Uniques for wired-in prelude things and known tags]

`mkUnique` constructs a `Unique` from its pieces
  mkUnique :: Char -> Word64 -> Unique

-}

-- | Unique identifier.
--
-- The type of unique identifiers that are used in many places in GHC
-- for fast ordering and equality tests. You should generate these with
-- the functions from the 'UniqSupply' module
--
-- These are sometimes also referred to as \"keys\" in comments in GHC.
newtype Unique = MkUnique Word64

{-# INLINE uNIQUE_BITS #-}
uNIQUE_BITS :: Int
uNIQUE_BITS :: Int
uNIQUE_BITS = Int
64 Int -> Int -> Int
forall a. Num a => a -> a -> a
- UNIQUE_TAG_BITS

{-
Now come the functions which construct uniques from their pieces, and vice versa.
The stuff about unique *supplies* is handled further down this module.
-}

unpkUnique      :: Unique -> (Char, Word64)        -- The reverse

mkUniqueGrimily :: Word64 -> Unique                -- A trap-door for UniqSupply
getKey          :: Unique -> Word64                -- for Var

incrUnique   :: Unique -> Unique
stepUnique   :: Unique -> Word64 -> Unique
newTagUnique :: Unique -> Char -> Unique

mkUniqueGrimily :: Word64 -> Unique
mkUniqueGrimily = Word64 -> Unique
MkUnique

{-# INLINE getKey #-}
getKey :: Unique -> Word64
getKey (MkUnique Word64
x) = Word64
x

incrUnique :: Unique -> Unique
incrUnique (MkUnique Word64
i) = Word64 -> Unique
MkUnique (Word64
i Word64 -> Word64 -> Word64
forall a. Num a => a -> a -> a
+ Word64
1)
stepUnique :: Unique -> Word64 -> Unique
stepUnique (MkUnique Word64
i) Word64
n = Word64 -> Unique
MkUnique (Word64
i Word64 -> Word64 -> Word64
forall a. Num a => a -> a -> a
+ Word64
n)

mkLocalUnique :: Word64 -> Unique
mkLocalUnique :: Word64 -> Unique
mkLocalUnique Word64
i = Char -> Word64 -> Unique
mkUnique Char
'X' Word64
i

minLocalUnique :: Unique
minLocalUnique :: Unique
minLocalUnique = Word64 -> Unique
mkLocalUnique Word64
0

maxLocalUnique :: Unique
maxLocalUnique :: Unique
maxLocalUnique = Word64 -> Unique
mkLocalUnique Word64
uniqueMask

-- newTagUnique changes the "domain" of a unique to a different char
newTagUnique :: Unique -> Char -> Unique
newTagUnique Unique
u Char
c = Char -> Word64 -> Unique
mkUnique Char
c Word64
i where (Char
_,Word64
i) = Unique -> (Char, Word64)
unpkUnique Unique
u

-- | Bitmask that has zeros for the tag bits and ones for the rest.
uniqueMask :: Word64
uniqueMask :: Word64
uniqueMask = (Word64
1 Word64 -> Int -> Word64
forall a. Bits a => a -> Int -> a
`shiftL` Int
uNIQUE_BITS) Word64 -> Word64 -> Word64
forall a. Num a => a -> a -> a
- Word64
1

-- | Put the character in the highest bits of the Word64.
-- This may truncate the character to UNIQUE_TAG_BITS.
-- This function is used in @`mkSplitUniqSupply`@ so that it can
-- precompute and share the tag part of the uniques it generates.
mkTag :: Char -> Word64
mkTag :: Char -> Word64
mkTag Char
c = HasCallStack => Int -> Word64
Int -> Word64
intToWord64 (Char -> Int
ord Char
c) Word64 -> Int -> Word64
forall a. Bits a => a -> Int -> a
`shiftL` Int
uNIQUE_BITS

-- pop the Char in the top 8 bits of the Unique(Supply)

-- No 64-bit bugs here, as long as we have at least 32 bits. --JSM

-- and as long as the Char fits in 8 bits, which we assume anyway!

mkUnique :: Char -> Word64 -> Unique       -- Builds a unique from pieces
-- EXPORTED and used only in GHC.Builtin.Uniques
mkUnique :: Char -> Word64 -> Unique
mkUnique Char
c Word64
i
  = Word64 -> Unique
MkUnique (Word64
tag Word64 -> Word64 -> Word64
forall a. Bits a => a -> a -> a
.|. Word64
bits)
  where
    tag :: Word64
tag  = Char -> Word64
mkTag Char
c
    bits :: Word64
bits = Word64
i Word64 -> Word64 -> Word64
forall a. Bits a => a -> a -> a
.&. Word64
uniqueMask

mkUniqueInt :: Char -> Int -> Unique
mkUniqueInt :: Char -> Int -> Unique
mkUniqueInt Char
c Int
i = Char -> Word64 -> Unique
mkUnique Char
c (HasCallStack => Int -> Word64
Int -> Word64
intToWord64 Int
i)

mkUniqueIntGrimily :: Int -> Unique
mkUniqueIntGrimily :: Int -> Unique
mkUniqueIntGrimily = Word64 -> Unique
MkUnique (Word64 -> Unique) -> (Int -> Word64) -> Int -> Unique
forall b c a. (b -> c) -> (a -> b) -> a -> c
. HasCallStack => Int -> Word64
Int -> Word64
intToWord64

unpkUnique :: Unique -> (Char, Word64)
unpkUnique (MkUnique Word64
u)
  = let
        -- The potentially truncating use of fromIntegral here is safe
        -- because the argument is just the tag bits after shifting.
        tag :: Char
tag = Int -> Char
chr (HasCallStack => Word64 -> Int
Word64 -> Int
word64ToInt (Word64
u Word64 -> Int -> Word64
forall a. Bits a => a -> Int -> a
`shiftR` Int
uNIQUE_BITS))
        i :: Word64
i   = Word64
u Word64 -> Word64 -> Word64
forall a. Bits a => a -> a -> a
.&. Word64
uniqueMask
    in
    (Char
tag, Word64
i)

-- | The interface file symbol-table encoding assumes that known-key uniques fit
-- in 30-bits; verify this.
--
-- See Note [Symbol table representation of names] in "GHC.Iface.Binary" for details.
isValidKnownKeyUnique :: Unique -> Bool
isValidKnownKeyUnique :: Unique -> Bool
isValidKnownKeyUnique Unique
u =
    case Unique -> (Char, Word64)
unpkUnique Unique
u of
      (Char
c, Word64
x) -> Char -> Int
ord Char
c Int -> Int -> Bool
forall a. Ord a => a -> a -> Bool
< Int
0xff Bool -> Bool -> Bool
&& Word64
x Word64 -> Word64 -> Bool
forall a. Ord a => a -> a -> Bool
<= (Word64
1 Word64 -> Int -> Word64
forall a. Bits a => a -> Int -> a
`shiftL` Int
22)

{-
************************************************************************
*                                                                      *
\subsection[Uniquable-class]{The @Uniquable@ class}
*                                                                      *
************************************************************************
-}

-- | Class of things that we can obtain a 'Unique' from
class Uniquable a where
    getUnique :: a -> Unique

hasKey          :: Uniquable a => a -> Unique -> Bool
a
x hasKey :: forall a. Uniquable a => a -> Unique -> Bool
`hasKey` Unique
k    = a -> Unique
forall a. Uniquable a => a -> Unique
getUnique a
x Unique -> Unique -> Bool
forall a. Eq a => a -> a -> Bool
== Unique
k

instance Uniquable FastString where
 getUnique :: FastString -> Unique
getUnique FastString
fs = Int -> Unique
mkUniqueIntGrimily (FastString -> Int
uniqueOfFS FastString
fs)

instance Uniquable Int where
  getUnique :: Int -> Unique
getUnique Int
i = Int -> Unique
mkUniqueIntGrimily Int
i

instance Uniquable ModuleName where
  getUnique :: ModuleName -> Unique
getUnique (ModuleName FastString
nm) = FastString -> Unique
forall a. Uniquable a => a -> Unique
getUnique FastString
nm


{-
************************************************************************
*                                                                      *
\subsection[Unique-instances]{Instance declarations for @Unique@}
*                                                                      *
************************************************************************

And the whole point (besides uniqueness) is fast equality.  We don't
use `deriving' because we want {\em precise} control of ordering
(equality on @Uniques@ is v common).
-}

-- Note [Unique Determinism]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~
-- The order of allocated @Uniques@ is not stable across rebuilds.
-- The main reason for that is that typechecking interface files pulls
-- @Uniques@ from @UniqSupply@ and the interface file for the module being
-- currently compiled can, but doesn't have to exist.
--
-- It gets more complicated if you take into account that the interface
-- files are loaded lazily and that building multiple files at once has to
-- work for any subset of interface files present. When you add parallelism
-- this makes @Uniques@ hopelessly random.
--
-- As such, to get deterministic builds, the order of the allocated
-- @Uniques@ should not affect the final result.
-- see also wiki/deterministic-builds
--
-- Note [Unique Determinism and code generation]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-- The goal of the deterministic builds (wiki/deterministic-builds, #4012)
-- is to get ABI compatible binaries given the same inputs and environment.
-- The motivation behind that is that if the ABI doesn't change the
-- binaries can be safely reused.
-- Note that this is weaker than bit-for-bit identical binaries and getting
-- bit-for-bit identical binaries is not a goal for now.
-- This means that we don't care about nondeterminism that happens after
-- the interface files are created, in particular we don't care about
-- register allocation and code generation.
-- To track progress on bit-for-bit determinism see #12262.

eqUnique :: Unique -> Unique -> Bool
eqUnique :: Unique -> Unique -> Bool
eqUnique (MkUnique Word64
u1) (MkUnique Word64
u2) = Word64
u1 Word64 -> Word64 -> Bool
forall a. Eq a => a -> a -> Bool
== Word64
u2

ltUnique :: Unique -> Unique -> Bool
ltUnique :: Unique -> Unique -> Bool
ltUnique (MkUnique Word64
u1) (MkUnique Word64
u2) = Word64
u1 Word64 -> Word64 -> Bool
forall a. Ord a => a -> a -> Bool
< Word64
u2

-- Provided here to make it explicit at the call-site that it can
-- introduce non-determinism.
-- See Note [Unique Determinism]
-- See Note [No Ord for Unique]
nonDetCmpUnique :: Unique -> Unique -> Ordering
nonDetCmpUnique :: Unique -> Unique -> Ordering
nonDetCmpUnique (MkUnique Word64
u1) (MkUnique Word64
u2)
  = if Word64
u1 Word64 -> Word64 -> Bool
forall a. Eq a => a -> a -> Bool
== Word64
u2 then Ordering
EQ else if Word64
u1 Word64 -> Word64 -> Bool
forall a. Ord a => a -> a -> Bool
< Word64
u2 then Ordering
LT else Ordering
GT

{-
Note [No Ord for Unique]
~~~~~~~~~~~~~~~~~~~~~~~~~~
As explained in Note [Unique Determinism] the relative order of Uniques
is nondeterministic. To prevent from accidental use the Ord Unique
instance has been removed.
This makes it easier to maintain deterministic builds, but comes with some
drawbacks.
The biggest drawback is that Maps keyed by Uniques can't directly be used.
The alternatives are:

  1) Use UniqFM or UniqDFM, see Note [Deterministic UniqFM] to decide which
  2) Create a newtype wrapper based on Unique ordering where nondeterminism
     is controlled. See GHC.Unit.Module.Env.ModuleEnv
  3) Change the algorithm to use nonDetCmpUnique and document why it's still
     deterministic
  4) Use TrieMap as done in GHC.Cmm.CommonBlockElim.groupByLabel
-}

instance Eq Unique where
    Unique
a == :: Unique -> Unique -> Bool
== Unique
b = Unique -> Unique -> Bool
eqUnique Unique
a Unique
b
    Unique
a /= :: Unique -> Unique -> Bool
/= Unique
b = Bool -> Bool
not (Unique -> Unique -> Bool
eqUnique Unique
a Unique
b)

instance Uniquable Unique where
    getUnique :: Unique -> Unique
getUnique Unique
u = Unique
u

-- We do sometimes make strings with @Uniques@ in them:

showUnique :: Unique -> String
showUnique :: Unique -> String
showUnique Unique
uniq
  = case Unique -> (Char, Word64)
unpkUnique Unique
uniq of
      (Char
tag, Word64
u) -> Char
tag Char -> String -> String
forall a. a -> [a] -> [a]
: Word64 -> String
w64ToBase62 Word64
u

pprUniqueAlways :: IsLine doc => Unique -> doc
-- The "always" means regardless of -dsuppress-uniques
-- It replaces the old pprUnique to remind callers that
-- they should consider whether they want to consult
-- Opt_SuppressUniques
pprUniqueAlways :: forall doc. IsLine doc => Unique -> doc
pprUniqueAlways Unique
u
  = String -> doc
forall doc. IsLine doc => String -> doc
text (Unique -> String
showUnique Unique
u)
{-# SPECIALIZE pprUniqueAlways :: Unique -> SDoc #-}
{-# SPECIALIZE pprUniqueAlways :: Unique -> HLine #-} -- see Note [SPECIALIZE to HDoc] in GHC.Utils.Outputable

instance Outputable Unique where
    ppr :: Unique -> SDoc
ppr = Unique -> SDoc
forall doc. IsLine doc => Unique -> doc
pprUniqueAlways

instance Show Unique where
    show :: Unique -> String
show Unique
uniq = Unique -> String
showUnique Unique
uniq

{-
************************************************************************
*                                                                      *
\subsection[Utils-base62]{Base-62 numbers}
*                                                                      *
************************************************************************

A character-stingy way to read/write numbers (notably Uniques).
The ``62-its'' are \tr{[0-9a-zA-Z]}.
Code stolen from Lennart.
-}

w64ToBase62 :: Word64 -> String
w64ToBase62 :: Word64 -> String
w64ToBase62 Word64
n_ = Word64 -> String -> String
go Word64
n_ String
""
  where
    -- The potentially truncating uses of fromIntegral here are safe
    -- because the argument is guaranteed to be less than 62 in both cases.
    go :: Word64 -> String -> String
go Word64
n String
cs | Word64
n Word64 -> Word64 -> Bool
forall a. Ord a => a -> a -> Bool
< Word64
62
            = let !c :: Char
c = Int -> Char
chooseChar62 (HasCallStack => Word64 -> Int
Word64 -> Int
word64ToInt Word64
n) in Char
c Char -> String -> String
forall a. a -> [a] -> [a]
: String
cs
            | Bool
otherwise
            = Word64 -> String -> String
go Word64
q (Char
c Char -> String -> String
forall a. a -> [a] -> [a]
: String
cs) where (!Word64
q, Word64
r) = Word64 -> Word64 -> (Word64, Word64)
forall a. Integral a => a -> a -> (a, a)
quotRem Word64
n Word64
62
                                  !c :: Char
c = Int -> Char
chooseChar62 (HasCallStack => Word64 -> Int
Word64 -> Int
word64ToInt Word64
r)

    chooseChar62 :: Int -> Char
    {-# INLINE chooseChar62 #-}
    chooseChar62 :: Int -> Char
chooseChar62 (I# Int#
n) = Char# -> Char
C# (Addr# -> Int# -> Char#
indexCharOffAddr# Addr#
chars62 Int#
n)
    chars62 :: Addr#
chars62 = Addr#
"0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ"#