Past due
Milestone
Mar 1, 2020–Sep 30, 2020
9.0.1
Release tracking ticket: #18216.
Current timeline
- 16 June 2020: Beginning of feature freeze
- Late June 2020: Cut branch
- July - August 2020: Alpha releases
- 1 September 2020: Beta release
- 25 September 2020: Final 9.0.1 release
Major highlights
See the status page.
Unstarted Issues (open and unassigned)
113
- GHC 9.0+ regression in checking program with higher-rank kinds
- Ord1 instance for Down is broken
- 8.8.1 regression: Combining -prof and -optP-traditional breaks
- LLVM Backend generates invalid assembly.
- Minor code refactoring leads to drastic performance degradation
- Fixing 64bit comparisons on 32bit x86
- ghc-option -fspecialise-aggressively could lead to wrong callstack information
- hadrian --help doesn't tell which ghc-bignum backend is the default.
- GHCi slow to start on OpenBSD
- C.UTF-8 locale not available on Centos
- writeChunk and writeChunkNonBlock are not finished?
- Improve trimming of DmdAnal results using findTypeShape.
- Windows linker validate build broken
- "lookupIdSubst" warnings when compiling haddock
- ARMv7 job fails due to RTS linker issues
- NCG: Unreachable incorrect code in `outOfLineCmmOp`
- Assertion Failure in rts/sm/MarkWeak.c, line 375
- :info (->) inconsistent: Requires -> to be parenthesized but not <->
- ranlib segfaults when building libffi on i386 Windows
- mkNewTyConRhs panic when trying to constrain newtype with Coercible
- Cannot compile with `make` and `WERROR`
- Validation issues on Darwin due to hsc2hs tests
- ghci temporary shared objects don't include -L paths in rpath
- -ddump-rn-ast and -ddump-tc-ast not working in ghci?
- TH constraint solver change will break significant portions of Hackage.
- `word2Double#` does not match `doubleFromInteger` for large values
- updateNurseryStats is racy
- Logging of "heap allocated" events is racy
- GHC 8.6.5 is quite a bit larger on aarch64 than x86
- Segmentation fault in retainer profiler
- ffi018_ghci fails when unregisterised
- ghc-8.6 fails to build on unregisterised: ‘stg_MUT_ARR_PTRS_FROZEN0_info’ undeclared (first use in this function); did you mean ‘stg_MUT_ARR_PTRS_FROZEN_DIRTY_info’?
- Performance regression in for_ alternatives from GHC 8.2.2 to newer GHCs
- Significant compilation time blowup when refactoring singletons-heavy code
- Take exhaustiveness checking into consideration when using MonadFailDesugaring
- Valid hole fits and GADT type variable names
- Test Suite Unexpected failure for ghci062(ghci-ext)
- Error message involving type families points to wrong location
- Performance regression when using the GHC API to evaluate code compared to 8.4
- Can't build `prof`-flavour with `-fauto-all`
- Error message mentions Any
- add -Wname-shadowing and -Wunused-pattern-binds to the default warnings for ghci
- Template Haskell creates System names when it shouldn't
- fixST for lazy ST is a bit wrong
- Wredundant-constraints does not work when constraint synonym is used
- Memory usage when compiling jsaddle-dom exploded in 8.4.3
- Make nofib-style anaysis for perf/compiler
- DerivingVia (and perhaps even GND) works badly with DeriveGeneric
- Internal error with PartialTypeSignatures and TH
- GHC does not warn when two conflicting packages with same name are given via a dependent package and -package-db
- executable with library and template haskell doesn't compile statically
- Performance drop 60 times on non-profiling binary
- Higher rank types in pattern synonyms
- Quantified constraints half-work with equality constraints
- The settings and behaviour of idle GC are very confusing
- Maybe ghc-pkg register should unregister packages with "incompatible" signatures
- -Wall-missed-specialisations warnings not fatal with -Werror
- GHC reports missing INLINABLE pragmas in vector and ghc-prim
- heapprof001 segfaults in prof_hc_hb way on i386
- GHC determine illegal kind for standalone deriving with Deriving via
- GHC.Exts.Heap.getClosureData doesn't return the payload for AP_STACK
- Fix performance regressions from #14737
- 2-fold memory usage regression GHC 8.2.2 -> GHC 8.4.1 compiling `mmark` package
- GHCi fails to garbage collect declaration `l = length [1..10^8]` entered at prompt
- Performance regression (~34%) in 8.2.1, poor register allocation(?) in an inner loop over an array
- Large Address space is not supported on Windows
- unfolding info as seen when building a module depends on flags in a previously-compiled module
- switch -msse2 to be on by default
- Deriving Generic of a big type takes a long time and lots of space
- users-guide: links to extensions will break on ghc-8.12
- dataToTag# doesn't take advantage of tag bits
- Deprecate and remove -Wmissing-monadfail-instances
- Document -i RTS option
- Document how to write a pattern synonym that uses a view pattern
- "unable to commit memory" is rarely a compiler bug, update the error message
- Implement the Linear Types proposal
- Refactor NewHsTypeX to DerivedCoreTy
- Check if some auto apply code is dead and remove if appropriate.
- The docs for Unsafe.Coerce should recommend using Data.Coerce instead
- Why both gen and gen_no in bdescr?
- Every implication should bump the TcLevel exactly once
- Write documentation for iserv-proxy
- Use TcLevels to decide about floating out of implications
- Deprecate and Remove Data.Semigroup.Option and Data.Monoid.First/Last
- Derived enum instances should support list fusion.
- Data.Bits.testBit should use tagToEnum#
- Migrate GHC utility parsing functions from ghc-exactprint to GHC API
- Configure per package
- Link rts symbols weakly
- Optimize dataToTag# for small constructor families.
- Add Eq1/Ord1/Read1/Show1 instances to newtypes in Data.Semigroup
- Improve the worst case performance of weak pointers
- When a type application cannot be applied to an identifier due to the absence of an explicit type signature, let the error just say so!
- Unify unlifted pointer equality primitives
- Allow associated types in Minimal pragmas
- Qualified Holes
- ghc -M requires -dep-suffix for no good reason
- Record projections with ambiguous types
- Machine accessible interface to GHCi
- [GHCi] Add an option to make tab completion case-insensitive
- Allow execution of stage2 compiler to happen later
- There should be a flag for GHCi that tells it to load as many modules as possible
- Compact Regions with Lazy Evaluation
- Make StaticPtr (more) robust to code changes and recompilation
- GHC doesn't come with dynamic object files/libraries compiled with profiling
- Inconsistency in parsing trailing commas inside import section
- Replace the atomicModifyMutVar# primop
- no way to talk about unpacking sum types / unpacking tuples
- Extend TVar/MVar to N capacity / Add primitive channel type
- Allow inline pragmas on pattern synonyms
- Simplifier ticks exhausted on mapM/traverse
- Something is fishy with byte arrays on big endian systems
- Package imports can cause linker errors
Ongoing Issues (open and assigned)
50
- Hadrian appears to link against libffi unconditionally
- Follow-ups to D5169
- The well-kinded type invariant (in TcType)
- Implement with# primop
- vector test suite uses excessive memory on GHC 8.2
- Unused parameters in AsyncWinIO.c
- Hadrian binary distributions don't package ghci executable
- Override mktemp in gcc Win32 wrapper
- Another process wait issue on Windows?
- Eta-expansion of a left-section
- hDuplicateTo001 crash in profthreaded way
- `Any` appearing in a quantified constraint
- T13676 is broken on Darwin and Windows
- Potentially problematic data race in scheduler
- runghc HelloWorld.hs is quite slow
- Provide the provenance of unification variables in error messages when possible
- SerializedCompact has a [(Ptr a, Word)] instead of a custom datatype
- ghci takes super long time to find the type of large fractional number
- Full laziness destroys opportunities for join points
- interpreter: sequence_ (replicate 100000000 (return ())) gobbles up memory
- GHC takes up huge amount of memory when compiling accelerate 1.2.0
- Runtime performance regression with binary operations on vectors
- GHC HEAD's ghc-stage1 panics on Data.Typeable.Internal
- Common Block Elimination pass doesn't eliminate common blocks
- Bad monads performance
- typePrimRep panic while compiling GHC with profiling
- Type/data family instances in kind checking
- Export Word32#, Word64# on all architectures
- Implement DWARF debugging on powerpc64
- Entered absent arg in a Repa program
- TypeFamilies painfully slow
- Compilation results are not deterministic
- Add LLVM datalayout for Darwin/ARM
- Support for Apple ARM hardware
- Do longer experiment of GHCi on multiple console hosts
- Release tracking ticket for GHC 9.0.1
- Try moving to LLVM toolchain on Windows
- Fix CI for i386 Windows
- Indirect branches followed by data are bad
- Large bump in T13701 metrics on x86_64-deb9-dwarf
- Refactor Ct and friends
- Remove SPARC NCG backend?
- Clean up TrivColorable
- Selectively reduce optimisation level in GHC build
- Stabilize AArch64 support
- Ship check-ppr in bindist and compile during testsuite run
- Bring back -fast-llvm / drop the %function -> %object mangling
- Extend -Wall with incomplete-uni-patterns and incomplete-record-updates
- Combine STG free variable traversals.
- Very poor code generation for CoreUnfold.callSiteInline
Completed Issues (closed)
130
- integerGcd sometimes returns an incorrect result
- ghc-9.0.0.20200925/compiler/stage1/build/libHSghc-9.0.0.20200925.a(Plugins.o):function ghc_GHCziDriverziPlugins_defaultPluginzux_info: error: undefined reference to 'ghc_GHCziCoreziOptziMonad_zdfxMonadCoreM_closure'
- GHC 9.0.1-alpha1 Source Distribution tarball missing Hadrian scripts
- GHC 9.0.1 Alpha1 causes cc1 error on Windows
- GHC panic (Prelude.!!: index too large) and Core Lint error (Trans coercion mis-match)
- HEAD-only Core Lint error when building JuicyPixels code with -O2
- Incorrect result with bigNatPowMod
- HEAD-only Core Lint error with bifunctors-5.5.7 (Type of case alternatives not the same as the annotation on case)
- HEAD-only panic when compiling primitive-0.7.1 without optimization (cpe_app(runRW#))
- Write release notes and migration guide for winio
- HEAD-only Core Lint error with arith-encode-1.0.2 (Type of case alternatives not the same as the annotation on case)
- GHC 8.11.0.20200527 + bytestring PR191 compile time blowup
- Windows builds broken
- HEAD-only panic (tcSpliceExpr) on typed Template Haskell example
- ASSERTion failure when compiling generics-sop with devel2 HEAD
- HEAD-only Core Lint error when compiling vector function with -O2
- GHC HEAD-only Core Lint error (Kind mis-match in inst coercion)
- Compiling pandoc-2.9.2.1 on GHC HEAD panics (applyTypeToArgs)
- Core Lint error when compiling singletons (HEAD only), possibly due to nested foralls
- "GHC.StgToCmm.Env: variable not found" panic when compiling with -fobject-code (HEAD only)
- Poly-kinded rewrite rule fails to typecheck (HEAD only)
- HEAD-only panic (coVarKindsTypesRole, non coercion variable) with -O2 and -dcore-lint
- Consistent segfault when compiled program is executed
- `Bytes` and `mkBytes` needs to be reconsidered
- Significant compilation time regression between 8.4 and HEAD due to coverage checking
- shiftL and shiftR for Integer doesn't check that the shift amount is non-negative
- `-ddump-minimal-imports` does not generate minimal `.hs-boot` imports
- GHC HEAD's error messages are no longer colorized
- Naming clash between GHC.Tuple.Unit and ()
- Performance regression in base64-bytestring/base64/base64-bytes on GHC 8.10.1
- Data.Monoid.First not removed as promised
- Links to (some) haddocks in users guide broken
- Linear types parser acccepts 'a %0001 -> b'
- Windows build fails with Cmm lint error
- Dependency interface change not detected.
- Perf metrics: Hadrian unnecessarily quotes the names of test_envs
- GHC hangs on (possibly dubious) associated type synonym default
- Fix documentation build on deb10
- Confusing `:info (->)` output on GHC HEAD
- Testsuite runs LLVM tests even if llc isn't present
- No warning for unused, un-exported closed type family
- SPECIALISE instance pragmas allow nested `forall`s/contexts
- BIGNUM_BACKEND=native does not compile: libraries/ghc-bignum/src/GHC/Num/BigNat/Native.hs:668:34: error: Variable not in scope
- Inferred type variable check is defeated by nested foralls
- DerivingVia fails to roundtrip through TH
- Fix double free errors in Winio
- pprPanic compiling `hashtables` package (GHC head)
- Division by zero in bigNatRemWord
- Crash with data type return kinds
- Core Lint error when deriving Ix/Enum instances in separate TyClGroups
- UnliftedNewtypesDifficultUnification broken due to simplified subsumption
- Unsaturated applications of runRW# are rejected
- Reordering alternatives in case expression causes exhaustive function to become non-exhaustive
- What fixity does `~` has?
- PPC NCG: Do not generate redundant section directives for exported symbols
- `mod @Int` and `rem @Int` are not strict in its first argument
- GHC crashes after using up more than 16GB of RAM ("protocol-buffers-descriptor" package)
- Impredicativity checking is fooled by a type synonym
- Field selector types bungle the order (and specificites) of type variables
- GHC Panic with levity polymorphism
- LLVM backend fails to bootstrap GHC
- interruptible file open
- Panic with overloaded quotations
- PM check time regression in 8.10-rc1
- GhcThreaded is broken and not supported by Hadrian
- Hadrian doesn't reconfigure packages upon change of Cabal Setup arguments
- TH quotes drop local fixity declarations
- saks028 is broken on Windows
- Weird error spans in patterns
- Teach Template Haskell about specificity
- hDuplicateTo001 fragile under CI
- OccurAnal should not break the linter assumptions
- set/getAllocationCounter is broken in GHCi
- Using oneshot kqueue() on macOS
- Both `Ptr a` in SerializedCompact are inaccurate because of the `a`
- Calling hs_try_putmvar from an unsafe foreign call can cause the RTS to hang
- Cut an STM release
- GHC.Conc.Windows has a surprising queue
- GHC and iserv cannot agree on what an Integer is; insanity ensues
- Exponential compilation time with RWST & ReaderT stack with `-02`
- Inconsistent precedence of ~
- Warning about __sync_fetch_and_nand semantics from gcc
- :break command should recognize also nonexported top level symbols in qualified IDs
- bang patterns with infix ops
- Update Cabal to final 3.4.0.0 tag
- Consider removing ConDeclGADTPrefixPs
- Inconsistent "pipeline" module names in 8.12
- Drop 32-bit Windows support
- Restore console CP on program exit
- Change std handles to utf8 on startup of winio
- Bump ghc-prim to 0.7.0
- Bump LLVM requirement to 10.0
- Avoid unrepresentable ForallVis/Inferred state in HsForAllTy
- Make TTG extension constructor fields strict to avoid noExtCon boilerplate
- Add singleton to Data.List.NonEmpty in libraries/base
- Implement GHC Proposal 287: Simplify subsumption
- Bump text submodule
- Parse Haddock comments outside of Parser.y
- Split up coercionKind
- Create Wiki page summarizing state of CAF rework
- Break up glasgow_exts.rst
- Instantiate eagerly in terms
- compiler/rename/RnExpr.hs uses mkVarOccUnique
- Add list singleton function
- Update the "disjointX" variants to use IntMap.disjoint
- Add missing Ix instances for tuples up to size 15
- Implement the "Explicit specificity in type variable binders" proposal
- Split up glasgow_exts.rst
- ghc-bignum is missing a few functions
- Implement -XQualifiedDo
- IsList instance for ZipList
- show breakpoint locations of breakpoints which were ignored during :force
- Discrepancy between parsed AST and source code vis-à-vis linear arrows in GADTs
- winio: xchg semantics that got merged not the same as winio expected
- Code is levity-polymorphic, but CodeQ and TExpQ aren't
- Visible dependent quantification permitted in standalone deriving declarations
- Document the behavior of ScopedTypeVariables vis-à-vis instance heads more precisely
- Consider past assignments in the linear register allocator.
- GADT constructors and the forall-or-nothing rule
- Define a Quote IO instance to parallel the existing Quasi IO instance
- Rework the RULES story for `elem`.
- -ddump-deriv doesn't display promoted data constructors correctly
- Sufficiently higher-rank record selectors fail to typecheck
- Hadrian should disable GHC's colour output when using --no-colour
- Master is broken using hadrian and 8.10
- Changelog/@since information for hGetContents'/getContents'/readFile' is inaccurate
- Code generated by GND/DerivingVia should use InstanceSigs
- Refactor DeriveFunctor to handle rank-n field types better
- liftData should use Quote, not Q
- Using type families in data types' return kinds leads to GHC panics and Core Lint errors
