fumieval/Haskell-breaking-changes

This document describes a list of upcoming/proposed changes to Haskell core/de facto libraries.

Ongoing: the change is already rolled out.
Upcoming: the change is made upstream, and will be released in near future.
Planned: the change has been approved by the committee. Implementation is not necessaryly available

Ongoing: Word8#

Summary: Word{8,16,32} are now defined in terms of the corresponding primitive types instead of Word#.
Reason: Small word types used to store Word# inside, making record types containing them space-inefficient.
Since: GHC 9.2
Breakage: major
Proposal: https://ghc-proposals.readthedocs.io/en/latest/proposals/0074-small-primitives.html
Merge request: https://gitlab.haskell.org/ghc/ghc/-/merge_requests/4390
How to fix: insert word*ToWord* (https://hackage.haskell.org/package/ghc-prim-0.8.0/docs/GHC-Prim.html#v:word8ToWord-35-) conditionally (cf. Fix build on GHC 9.2 by josephcsible · Pull Request #354 · haskell-crypto/cryptonite) using CPP

Summary: Control.Monad.Trans.List is gone
Reason: ListT is a monad only if the underlying monad is commutative; therefore it is not a valid monad transformer.
ETA: GHC 9.4 / transformers-0.6
Expected breakage: medium
Tracking issue: https://hub.darcs.net/ross/transformers/issue/75
Patch: https://hub.darcs.net/ross/transformers/patch/7c809b6f9db019ed761c971d10c86fa004f60d89
How to fix: remove instance declarations for transformer's ListT (deprecated). Users should switch to lawful alternatives such as pipes.

Summary: Control.Monad.Trans.Error is removed in favour of Control.Monad.Trans.Except.
Reason: ErrorT imposes an Error constraint, making it difficult to use as a general-purpose Either transformer.
ETA: GHC 9.4 / transformers-0.6
Expected breakage: medium
Tracking issue: N/A
Patch: https://hub.darcs.net/ross/transformers/patch/7c809b6f9db019ed761c971d10c86fa004f60d89
How to fix: remove instance declarations for ErrorT and switch to ExceptT

Summary: Foldable/Traversable-polymorphic functions in Data.List get specialised to []
Reason: There's a custom importing Data.List unqualified, making additions like singleton and uncons controversial. Making the list API monomorphic and encouraging to import it qualified make it more consistent with other container modules.
ETA: GHC 9.6 / base-4.18?
Proposal: https://groups.google.com/g/haskell-core-libraries/c/q3zHLmzBa5E/m/OrHHKaJNAQAJ?pli=1
Tracking issue: #20025
Merge request: !5304 https://gitlab.haskell.org/ghc/ghc/-/merge_requests/5107
Expected breakage: significant
Relevant warnings: -Wcompat-unqualified-imports (implied by -Wcompat)
How to fix: import Data.List qualified or explicitly import functions that don't conflict with Foldable/Traversable variants

Summary: 'forall' will become a keyword at the term level
Reason: Making Type-level and term-level syntax more consistent is a way forward to introduce dependent types
Proposal: Visible 'forall' in types of terms
ETA: GHC 9.10?
Tracking issue: N/A
Merge request: N/A
Expected breakage: minor (authors of the affected packages were notified)
Relevant warnings: -Wforall-identifier (introduced and enabled by default in GHC 9.4)
How to fix: Rename identifiers

Summary: return gets removed from the Monad class and becomes a top-level function defined as return = pure
Reason: Now that Monad is a subclass of Applicative, return is completely redundant (it doesn't make sense if return is not pure).
Proposal: GHC wiki: monad-of-no-return, related: Enable -Wnoncanonical-monad-instances and -Wnoncanonical-monoid-instances by default
ETA: Unknown
Tracking issue: N/A
Merge request: N/A
Expected breakage: minor (canonical definition of return will be ignored)
Relevant warnings: -Wnoncanonical-monad-instances (enabled by default as of GHC 9.2)
How to fix: Remove definitions of return (compatible up to GHC 7.10)

Summary: mappend gets removed from the Monoid class and becomes a top-level function defined as mappend = (<>)
Reason: Since Semigroup is a superclass of Monoid, mappend does not have to be defined.
Proposal: GHC wiki: semigroup-monoid, related: Enable -Wnoncanonical-monad-instances and -Wnoncanonical-monoid-instances by default
ETA: Unknown
Tracking issue: N/A
Merge request: N/A
Expected breakage: major (without an update to the warning)
Relevant warnings: -Wnoncanonical-monoid-instances (enabled by default as of GHC 9.2)
How to fix: Remove definitions of mappend (compatible up to GHC 8.4)
TODO: make -Wnoncanonical-monoid-instances warn manual declarations of mappend

Summary: (/=) gets removed from the Eq class and becomes a top-level function defined as a /= b = not (a == b)
Reason: The only valid semantics of x /= y is not (x == y); unlike compare vs (<=), there's no room for any performance improvements.
Proposal: haskell/core-libraries-committee#3
ETA: Unknown
Tracking issue: N/A
Merge request: N/A
Expected breakage: 131 packages according to the proposal
How to fix: Remove definitions of (/=) (backward-compatible)

Summary: * will no longer be parsed as a kind of value types (Type).
Reason: Special treatment to * complicates the syntax and prevents the use of it as a type operator.
Proposal: ghc-proposals/ghc-proposals#143
ETA: Unknown
Tracking issue: N/A
Merge request: N/A
Expected breakage: medium
Relevant warnings: -Wstar-is-type
How to fix: Replace * with Type (from Data.Kind)