GHCJS uses the same representation for Addr# and StablePtr#, but not for
Int#.

In old GHC:

    I8# :: Int# -> Int8
    indexInt8OffAddr# :: Addr# -> Int# -> Int#
    ...

In upcoming GHC 9.2:

    I8# :: Int8# -> Int8
    indexInt8OffAddr# :: Addr# -> Int# -> Int8#
    ...

Both of those work with the last array release, where the only
requirement is merely that the constructor and primops agree, i.e.:

    exists alpha.
    I8# :: alpha -> Int8
    indexInt8OffAddr# :: Addr# -> Int# -> alpha

In current GHC HEAD, we had

    I8# :: Int8# -> Int8
    indexInt8OffAddr# :: Addr# -> Int# -> Int#

and it was only because `Int8# /= Int#` that we needed the commit being
reverted. we are about to fix the primops to match the constructors in
accordance with the final 9.2 design, and so we don't need that commit
anymore.

This reverts commit c7a696e3.

Release v0.5.4.0

Better late than never. Fixes #3.

As noted in #3, these were apparently released without ever making it
into a commit. Thankfully changes are pretty harmless.

Alec Theriault authored 6 years ago and Ryan Scott committed 6 years ago

Summary:
This matches exactly the 'Read (Array i e)' instance defined in base.
Note that the same thing is being done for 'Show (UArray i e)'.

Reviewers: RyanGlScott, bgamari

GHC Trac Issues: #11335

Differential Revision: https://phabricator.haskell.org/D5156

Chaitanya Koparkar authored 7 years ago and Ben Gamari committed 7 years ago

Summary:
The old Travis script still had targets to test against GHC < 8.0,
and wasn't using Cabal `new-build` among other things.

Reviewers: bgamari, RyanGlScott

Reviewed By: RyanGlScott

Differential Revision: https://phabricator.haskell.org/D4681

See https://ghc.haskell.org/trac/ghc/ticket/15018.

David Feuer authored 7 years ago and Ben Gamari committed 7 years ago

Summary: These have shifted out of the support window.

Reviewers: bgamari, hvr, RyanGlScott

Reviewed By: RyanGlScott

Subscribers: RyanGlScott

Differential Revision: https://phabricator.haskell.org/D4499

The fact that this was previously hidden appears to be an oversight and
I've needed it at least once.

The recent overflow check made the code stop working w/ prior versions
of GHC. But there's little benefit in support much older GHCs, so it's
easier to just cut the support window.

The ghc-8.2 and master branches disagree on the order of the instances.
Normalise this difference away.

Previously it failed with,

-T229: Data.Array.Base.safe_scale: Overflow; scale: 4, n: 4611686018427387904
-CallStack (from HasCallStack):
-  error, called at libraries/array/Data/Array/Base.hs ...
+T229: Ix{Int}.index: Index (17) out of range ((0,-1))

No overflow check is necessary in the case of Bool arrays since the array size
is actually smaller than the number of elements it contains. Moreover, we can
use setByteArray# to initialize the array.

As akio points out, the fix to #229 which I authored previously was blatantly
wrong.

Fixes GHC #4505.