slpj tickets

This page collects tickets that Simon PJ is interested in, so that he remembers them

Status pages

hs-boot files: Commentary/Compiler/TyingTheKnot
Trees That Grow (TTG)
Language features
- ArrowNotation
- ApplicativeDo
- Exceptions: exceptions
- Pattern-synonym status page
- Overloaded record fields
- StaticPointers
- Backpack
Types:
- Type-indexed type representations: plan and status
- OIverloaded record fields meta-ticket: #18598
- TypeInType: DependentHaskell, esp #12919 (closed) (outright bug)
- TypeApplications: visible type application
- Custom type errors
- TypeErrorMessages
- Deferred type and scope errors
- Typed holes
- PartialTypeSignatures
- Impredicative types
- Functional dependencies
- Roles, Coercible and coerce
- Injective type families
- UnliftedDataTypes
- UnpackedSumTypes, #9214 (closed), #14826
- TypeNats: type-level literals
- TemplateHaskell
- LevityPolymorphism
- Type families
- #16148, #11715, #11621: Constraint vs *. #16148 is about improving inference for tuples and foralls.
- Generic deriving
- Deriving generally
- Typeable
- QuantifiedConstraints
- Type-checker plugins
Optimisation
The new pattern-match overlap/exhaustiveness checker
Code generation
- Optimising closure representation #14461
- Pointer tagging

Compile-time performance

Compile time perf page

#18541, #18535 low hanging perf fruit
#11375 (closed): type synonyms slower than type families
#8095: very slow constraint solving; perhaps discard large constraints using UnivCo
#5642: slow constraint solving
#9979, #8814, #8835 (closed): attoparsec regressed 7.8.4 to HEAD
#8852 (closed), #8980 (closed), #8941 (closed) (possibly), #9803 (closed), #8960 (closed), #7898 (closed), #7068 (closed), #7944 (closed), #5550 (closed), #8836 (closed): SpecConstr blowup
#10289 (closed): 2.5k static HashSet takes too much memory to compile
#10228 (closed): compile-time regression from 7.8.4 to 7.10.1
#7428: Non-linear compile time: addFingerprint??
#2346: desugaring let-bindings
Use wildcards for dead variables in interface files.

Code-size blowup when using deriving; see tickets with keyword deriving-perf

#7450, #7258: deriving Read generates gigantic code. Better now, but still not linear.
#9583 (closed), #9630 (closed): code blowup in Generics/Binary
#9669, #9557, #8731: slow compilation with lots of deriving clauses
#10858 (closed): Ord instances
#1544: Read1 instances

Run-time performance

Run time perf page

#15167: Adding superclass Monad m => makes the program 100x slower.
The notorious state hack, and replicateM in particular
- #1168 has a list of related tickets
- #11677
- #11365 (closed)
- #6166 (closed): apparently something with NOINLINE is getting inlined
- #9388 has ideas and preliminary work on how to limit the scope of the hack
- Search for "replicateM" to find other tickets
- #7411: state hack changes exception semantics
- #2284 (closed), #1168, #7561 (closed), #9349: the state-hack "optimisation" causes much re-computation

INLINE problems / ticks exhausted

#13027 (closed): let/app invariant
#12603 (closed)
#12747
#12790 (closed): too much inlining
#8613 (closed), #9070 (closed), #8319, #7521 (closed), #10459 (closed), #10584, #10565 (closed), #12274: simplifier ticks exhausted (there are others).

Urgent or in-flight

#12088: phase ordering for type-in-type
#11371: bogus in-scope set for substitution
#10844, #10845 (closed), #10846 (closed): implicit call stacks
#7262 (closed), #10333, #7672 (closed), #10083 (closed): hs-boot niggles
#10114: AnyK and kind generalisation in type synonyms
#10068: runtime reflection API for packages, modules, names; SrcLoc etc.
#9960 (closed), #9805: TrieMap stuff
#10016: UNPACK support for existentials
#10803: type signature sections (Lennart)
#8634, #9267, #9227 (closed): relaxing functional dependency rules -XDysFunctionalDependencies
#9123 (closed): higher order roles
#9352: allow State# in FFI types.
#9725 (closed): kind equalities (Richard)
#9717, #9729, #10420 (closed): orphan modules

Type families, roles, `Coercible`, `Typeable`, kind polymorphism

Refactoring/optimisation

#11598 (closed), #8095: drop coercions
#11739: coercion axioms applied to types, not coercions
#11735, #11598 (closed): coercionKind
#11196: resurrect FunTy
Make CoVar a proper Var
#1965: make existentials have a really cheap implementation

Other tickets

#12369 (closed): data families with non-* kind (Kmett)
#11622 (closed): parens around kind signatures in types
#1311 (closed): newtypes over unboxed types
#11203 (closed), #11453 (closed): Kind inference with SigTvs is wrong
#11080: open data kinds
#7102: type instance overlap in GHCi is unsound
#9858 (closed), #10343 (closed): Typeable and kind polymorphism
#10347 (closed): wrong "unused data constructor" warning for a newtype (using Coercible)
#10184,#10185: incompleteness in Coercible solver
#8165 (closed): GND for classes with associated types
#10056 (closed): more systematic treatment of (~) in parser and renamer
#9918: closed type families, apartness, and overlapping instances (inconsistency)
#9118: Eta reduction for Coercible (not expressible in Core, yet)
#6018 (closed), #10227, #10832, #4259: injective type families
#8161 (closed): associated type more specific than class decl
#8177: role signature for type families
#9547: kinding for empty constraint tuples

Data types and pattern matching

#12159 naughty record selectors
#11253: duplicate warnings with new pattern checker
#10183 (closed): warning for redundant constraints, and pattern-match overlap warnings
#9113 (closed): pattern match overlap/exhaustiveness checked in Typed TH

Type inference

#12201 (closed): overlapping instances and instance declarations
#10179: displaying kinds
#10089: better warnings for unused data types (involves instances)
#10027: export behaviour for data families
#9223 (closed): error message mentioning untouchables
#9198: slow typechecker (big types!)
#9210, #10675: functional dependencies and overlapping instances
#8673: more generous GADT record selectors
#8550 (closed): Don't build recursive type-function dictionaries
#8281: UnliftedFFITypes etc
#8441: family instances in hs-boot files
#9450: need more eager checking of compatibility with hs-boot files
#7259: Eta-expansion of products in System FC
#7204: a class to controll FFI marshalling; and newtype deriving for classes with ATs
#7842: polymorphism in recursive do-blocks
#8171: extended default rules
#7828, #5267, #5777, #5333 (closed), #344: bugs in arrows
#7862 (closed): overlap beween givens and instances

Template Haskell

Demand analysis and CSE

#10918: float in used-once things into recursive bindings
#7782 (closed): late demand analysis
#10069 (closed): bad reboxing in data types with many fields
#8655: analysis for quick-to-evaluate thunks
#6070, #5949 (closed), #5775 (closed), #4267, #5302, #6087: Demand analysis
#5075, #3138: CPR for sum types (mostly done, needs finishing)
#10678 (closed), #2289, #2387, #1600, #1885: nested CPR analysis
#7596, #5996 (closed), #149, #2940 (closed), #947, #701, #5344, #9441 (closed): CSE opportunities; see MoreCSE
#6040, #5945 (closed), #3458, #1216: lambda-lift functions with a few args, just before code gen
#1171: strictness and exceptions (long, open ticket)

Arity

#10181: unsound eta-reduction
#11029: eta expansion causes slow-down
#11146: manual eta expansion is a big win
#10260 (closed), #9020, #10319: not enough eta expansion
#7542: optimisation of eta expansion/reduction
#5809 (closed): arity analysis could be better
#3924 (closed): weak arity leads to weak strictness
#3698 (closed): arity analysis again
#3697 (closed): class method selectors
#3034: divInt and arity
#2915 (closed): arity too small
#2823 (closed): another arity expansion bug (related to dictionaries)
#2440 (closed): bad code with type families; I believe this is also arity-related
#1547 (closed): profiling and arity
#5587 (closed), #7364 (closed): eta-expansion can imcrease termination. These tickets are closed because we don't know how to fix them properly, not becuase they are really fixed.

Inlining

Inlining wiki page

#13011, #11240 (closed), #8833 (closed), #3872, #5400, #5448, #5722, #7057, #7369 (closed), #9235: Russell's paradox; Y combinator; contravariant data type loop in simplifier inliner
#4833: finding the right loop breaker
#3781, #3755: inlining for local functions: discount for scrutinising free vars
#3765: CONLIKE things and case expressions (need two unfoldings)
#3073 and Commentary/Compiler/DesugaringInstances: better desugaring for instances
#3123 (closed): feature request: recursive inlining and peeling
#4960: better inlining tests

Optimisation

#11475: Lint should check for inexhaustive cases expressions
#9476 (closed): late lambda lifting
#7374, #10417, #10418, #7287, #7398, #10528 (closed), #11688 (closed), #10595, #7141, #5973 (closed), #2271 (closed), #1434 (closed): rules not firing; e.g. class methods, newtype constructors
#8457 (summarises), #917, #1945 (closed), #3273 (closed), #4276 (closed), #5729 (closed) (closed as dups, but not fixed): full laziness sometimes makes things worse.
#10528 (closed), #10595: RULES for class methods
#10535 (closed), #7367, #7206, #7309: float-out and excessive sharing
#9370, #8635: respecting mixed -O and -O0 in ghc --make, especially concerning cross-module inlining
#9279: local wrapper remains in final program
#9246: Too much case-of-case leads to bad code
#9136 (closed): better constant folding
#9041: bad code from native code generator (NCG)
#6056 (closed): conflict between the w/w unfolding and the one from INLINEABLE
#8472 (closed): top level unlifted string literals
#8335 (closed): more gc entry points
#8336: sinking pass does not sink
#8327: dead code (in heap checks) not eliminated by C--
#8317, #8326, #9661, #12231, #2289: don't push heapchecks into case branches too aggressively. This is closely related to work Jan Stolarek did on unboxed booleans #6135 (closed); and perhaps also to #10124, #10137 (closed)
#8308 (closed): resurrect ticky histograms
#8321 (closed): branch prediction for stack/heap checks for LLVM
#8279: alignment and its effect on performance
#1498: heap checks in recursive functions
#7378: (a) identical-alts when the RHSs are bigger; (b) nested comparisions (cf #2132)
#7511: inlining can make programs allocate MORE!!
#7307: low hanging fruit for shrinking code size for string constants and top-level indirections
#5954: performance regression. Duplication of primops?
#6047: GHC retains unnecessary binding
#2439: Strict dictionaries; see also map/coerce rule #9792
#5916 (closed), #10678 (closed): runST isn't free
#5522 (closed), #6092: liberate-case runs out of memory
#3990 (closed), #7647: UNPACK for data families, and with phantom types, needs data con wrapper refactoring; see also #9655 if you are in this area.
#5928: running Specialise more than once

Numerics, floating point, and primops

#10555: division primops on LHS of a RULE
#9407 (closed): 64-bit floating point behaves differently with and without -O
#5780: aggressive-primop attempt that didn't quite work
#9328, #10215 (closed): negative zero and case expressions
#9304 (closed): floating point woes: 32 vs 64 bit
#5615 (closed): poor code for div with powers of 2
#4101: constant folding for (**)
#3676: realToFrac conversions
#3744: comparisons against minBound and maxBound are not optimised away
#2269: Word type to Double or Float conversions
#3070: NaNs and divide-by-zero
#9251: branchless max/min
#9276: auditing for IEEE compliance

Features

#9049 (closed): explicit call stack, abstractable call site information
#5972: option to suppress record selectors

Cmm and code generation

See Edward's collection of code-generation stupidities

#10012: Cheap-to-compute values aren't pushed into case branches
#9718 (closed): avoiding TidyPgm having to predict arity and CAF-ref-ness
#9159 (closed): jump tables for dense cases
#9157 (closed): common block elimination
#8905: spilling around an eval
#8903: dead stores
#8887: double assignment
#8871: Cmm optimisation opportunity
#8585 (closed), #11372 (closed): loopification; see also Commentary/Compiler/Loopification
Summary ticket: #4258 (closed), and wiki page
#3462: allocate large objects with allocateLocal()
#2289: cheap check at start of case alternatives
#2731: avoiding unnecessary evaluation when unpacking constructors
#4121 (closed): arity and CAF info computation is horribly fragile

Optimiser

#10346, #2255, #3767, #2642, #3831, #4941, #2598: Improve SpecConstr in various ways, including for free variables, and for join points.
#5059: specialise on value arguments
#5974: casts, RULES, and parametricity
#5262: seq magic
#605: strict/unboxed enumerations
#2607: space leak: inlining defeats selector thunk optimisation
#4470: merge identical counters
#2988, #11197: better float-in
#2374: SAT and MutableByteArray Max?
#7080: inconsistent treatment of RULES and SPECIALISE
#876 (closed): make length into a good consumer. Perhaps using foldl?

GHCi

#9394 (closed): Display type/data family instance information in :info
#4017 (closed): unhelpful GHCi message

Outright bugs

#1241 (closed): Lifting the Coverage Condition for functional dependencies isn’t the Right Thing Manuel

Types and type inference

TypeFunctionsStatus
#10338: delicacy in generalisation: MonoLocalBinds isn't enough
#9587, #9607: -XAllowAmbiguousTypes debate
#9334: instance chains
#9427: SCC analysis for type and class decls
#7503: Kind polymorphism and mutual recursion
#4296: SkolemOccurs
#816: extreme delicacy in solve order, involving fundeps
#8109: as-patterns in type-family declarations
#6065: GHC suggests a type signature that it then rejects
#5320, #7296: overlap delicacy
#5224: make it cheaper to check for inconsistent type family instances
#4259: overlapping type family instances
#3490: superclasses and ambiguity
#3632: better update for records with existentials
#2641: revise what -XExtendedDefaultRules does

Better error messages

#10450: poor error message when arg is insufficiently polymorphic
#9901 (closed): f is applied to two arguments, but its type has only two
#9456: more info in "relevant bindings" message
#9244: suggest scoped type variables
#9173: inferred/expected error messages
#1330 (closed): another bad error message (Church2)
#2648: Report out of date interface files robustly
#1928: Confusing type error message (Claus makes suggestions)

Features

#6024: allow defining a kind without also getting the corresponding type
#5927 (closed): Constraints with universal quantification
#5429 (closed): docase and joinads
#5073: blockST and friends
#2895: Class aliases
#2595: record updates
#4823: strength reduction for array indexing
#4479 (closed): type directed name resolution (TDNR)
#4426 (closed): simpler rule for implicit quantification
#3701: Implicitly declared existentials: a class as a type
#3217 (closed): better flag handling for GHCi
#2600 (closed), #2110 (closed): Bind type variables and constraints in RULES
#960 (closed): Generate local info for ‘undefined’. Implicit location parameters in general
#2135: Warn when exporting a function whose type mentions a type constructor defined locally but not itself exported
#2119: Explicitly importing deprecated symbols should generate deprecation warnings
#2207: Load interfaces for GHC.* even without -O
#1231: Better deprecations

Not sure what to do about these

#7897: make the fingerprint in a TypeRep be a proper fingerprint, including the type definition all the way down (see Recompilation avoidance)
#4005: generational GC pathalogical case
#1349: strict function argument types

slpj tickets

This page collects tickets that Simon PJ is interested in, so that he remembers them

Status pages

Compile-time performance

Run-time performance

INLINE problems / ticks exhausted

Urgent or in-flight

Type families, roles, Coercible, Typeable, kind polymorphism

Data types and pattern matching

Type inference

Template Haskell

Demand analysis and CSE

Arity

Inlining

Optimisation

Numerics, floating point, and primops

Features

Cmm and code generation

Optimiser

GHCi

Outright bugs

Types and type inference

Better error messages

Features

Not sure what to do about these

Type families, roles, `Coercible`, `Typeable`, kind polymorphism