March 2018 - ghc-tickets

[GHC] #13208: Do two-phase inlining in simpleOptPgm
by GHC 06 Mar '19

06 Mar '19

#13208: Do two-phase inlining in simpleOptPgm -------------------------------------+------------------------------------- Reporter: lukemaurer | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- As of #12988, having `simpleOptPgm` leave β-redexes lying around is painful because it requires a special loophole in Core Lint to allow a jump inside a β-redex. We'd rather use a two-phase approach, mirroring the simplifier's `preInlineUnconditionally` and `postInlineUnconditionally`, so we can aggressively eliminate β-redexes without fear of exponential blowup. Consider: {{{ joinrec j1 x = e1 in join j2 x y = jump j1 e2 in jump j2 e3 e4 }}} Since `j2` is only used once, it gets inlined. But `simpleOptPgm` only performs inlining //after// a binding is simplified, so we end up here: {{{ joinrec j1 x = e1' in (\x y -> jump j1 e2') e3' e4' }}} Since `e2'` was already simplified, performing the β-reduction here risks exponential blowup for the same reason it can happen in the simplifier (see the Secrets paper; `perf/compiler/T783` is a live example, increasing in allocs by over 80% if we re-simplify here). We //could// just `let`-bind `e3'` and `e4'` (carefully avoiding capturing free occurrences of `x` in `e4'`), but not if one them is actually a type argument. Well, okay, we //could// introduce a type `let`, but doing that here means the desugarer can now return a type `let` and we're not prepared for that. (Specifically, under certain circumstances, the simplifier never runs in between the desugarer and Float Out, and the latter breaks in the presence of type `let`.) So for the moment, we leave the β-redex there, but this needs a special rule just for β-redexes because normally you can't have a jump under a lambda (or an application, for that matter). In the long term, we'd prefer to take the simplifier's approach instead: If we want to inline something because it only occurs once (even though it may be big), we add it to the substitution //before// simplifying it so that we only simplify it once. This means the substitution has to carry lexical closures around, though, just like the simplifier does (see `SimplSR`'s `ContEx` disjunct), so it's not trivial. The `simpleOptPgm` β-redexes are the only reason for the special rule in Core Lint (see Note [Beta redexes] in `CoreLint.hs`), so once this is done we can be rid of that. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13208> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 9

[GHC] #14030: Implement the "Derive Lift instances for data types in template-haskell" proposal
by GHC 05 Mar '19

05 Mar '19

#14030: Implement the "Derive Lift instances for data types in template-haskell" proposal -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: (none) Type: task | Status: new Priority: normal | Milestone: 8.4.1 Component: Template | Version: 8.3 Haskell | Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Back in September 2015, I [https://mail.haskell.org/pipermail/libraries/2015-September/026117.html proposed] using the `DeriveLift` extension to, well, derive `Lift` instance for data types in the `template-haskell` library. The proposal was well received, but I was unable to implement it at the time due to `DeriveLift`'s newness (having only been introduced in GHC 8.0). Now that GHC 8.0 is the oldest version of GHC that we support bootstrapping with, this is no longer an obstacle. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14030> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 9

[GHC] #12596: can't find interface-file declaration
by GHC 05 Mar '19

05 Mar '19

#12596: can't find interface-file declaration -------------------------------------+------------------------------------- Reporter: mwotton | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- full repro at https://github.com/mwotton/liftwoes/issues/1 ``` /home/mark/projects/liftwoes/src/Lib.hs:14:11: error: • Can't find interface-file declaration for variable Data.Text.Internal.pack Probable cause: bug in .hi-boot file, or inconsistent .hi file Use -ddump-if-trace to get an idea of which file caused the error • In the first argument of ‘(:)’, namely ‘Data.Text.Internal.pack ((:) 'A' [])’ In the first argument of ‘HS.fromList’, namely ``` {{{#!hs {-# LANGUAGE OverloadedStrings #-} {-# LANGUAGE TemplateHaskell #-} module Lib where import Data.Data import qualified Data.Set as HS import qualified Data.Text as T import qualified Data.Text.IO as T import Data.Time -- import Instances.TH.Lift import Instances import Language.Haskell.TH.Syntax table = $(do r <- runIO (HS.fromList . T.lines <$> T.readFile "/usr/share/dict/words") [|r|] ) someFunc = do print $ HS.member "foo" table }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12596> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 7

[GHC] #10892: ApplicativeDo should use *> and <*
by GHC 05 Mar '19

05 Mar '19

#10892: ApplicativeDo should use *> and <* -------------------------------------+------------------------------------- Reporter: simonmar | Owner: Type: task | Status: new Priority: normal | Milestone: 8.0.1 Component: Compiler | Version: 7.11 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Revisions: | -------------------------------------+------------------------------------- From @ekmett: Would it be possible to tweak the generation to use `(<*)` or `(*>)` where appropriate when the result isn't being used? For many Applicatives this can be a massive asymptotic win in terms of sharing and/or computational cost. When desugaring using (<*) you'd just omit any handling of the unused result instead. {{{ (\x y -> ...) <$> foo <* bar <*> baz }}} corresponds to {{{ do x <- foo bar y <- baz return ... }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10892> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 30

[GHC] #13633: testwsdeque failure on POWER8
by GHC 04 Mar '19

04 Mar '19

#13633: testwsdeque failure on POWER8 ---------------------------------------+---------------------------------- Reporter: trommler | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Runtime System | Version: 8.1 Keywords: SMP | Operating System: Linux Architecture: powerpc64 | Type of failure: Runtime crash Test Case: rts/testwsdeque | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: ---------------------------------------+---------------------------------- On a POWER8 `testwsdeque` fails occasionally under load: {{{ Wrong exit code for testwsdeque(threaded1)(expected 0 , actual 134 ) Stderr ( testwsdeque ): internal error: FAIL: 277808608 3 13 (GHC version 8.3.20170430 for powerpc64le_unknown_linux) }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13633> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 9

[GHC] #14037: Fix fusion for GHC's utility functions
by GHC 04 Mar '19

04 Mar '19

#14037: Fix fusion for GHC's utility functions -------------------------------------+------------------------------------- Reporter: goldfire | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.3 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Compile-time Unknown/Multiple | performance bug Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- In a patchset I'm working on, I encountered a smallish "bytes allocated" regression in `perf/compiler/T5631`. After some sleuthing, I discovered that the regression is caused by using a `zipWith3M` where previously there was a `zipWithM`. Interestingly, it was not the zipped function that caused the problem, but the `zipWith3M` itself. Further sleuthing found that `zipWithM` is inlined into a combination of `sequenceA` and `zipWith`, both of which have been taught to work nicely with fusion. I thought of doing the same to GHC's `zipWith3M`, but I discovered that `base`'s `zipWith3` has no rules or inlining pragmas, suggesting that it does not fuse. I was quite surprised to find this, thinking that all functions as basic as `zipWith3` in `base` would play nicely with fusion. (Perhaps it does fuse -- I didn't actually check.) I then checked out GHC's `MonadUtils`, which defines a bunch of functions, some of which must be hammered on, and only one of which has any kind of performance- enhancing pragma. (The one is `zipWithAndUnzipM`, whose `INLINABLE` pragma was added only after much pain suffered by your dear author.) So: this ticket is a request to sweep through `MonadUtils` and `Util` and perhaps `ListSetOps`, looking for performance enhancements via fusion, tighter definitions, and such. Also, get `zipWith3` (and the other `zipWith`s) to fuse in `base`. I figure this is all some fairly low-hanging fruit for performance improvements in GHC. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14037> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 17

[GHC] #14671: Make the Template Haskell class Lift work over typed expressions
by GHC 04 Mar '19

04 Mar '19

#14671: Make the Template Haskell class Lift work over typed expressions -------------------------------------+------------------------------------- Reporter: simonpj | Owner: (none) Type: feature | Status: new request | Priority: normal | Milestone: Component: Template | Version: 8.2.2 Haskell | Keywords: | Operating System: Unknown/Multiple TemplateHaskell | Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- At the moment we have the `Lift` class in TH: {{{ class Lift t where lift :: t -> Q Exp }}} But that loses type information. What we want is this: {{{ Lift t where typedLift :: t -> Q (TExp t) lift :: Lift t => t -> Q Exp lift = fmap unType . typedLift }}} Otherwise, when you want the typed version, people resort to unsafe-coerce operations, which are entirely unnecessary. See [https://mail.haskell.org/pipermail/libraries/2018-January/028409.html this thread on the libraries list]. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14671> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 4

[GHC] #8657: -fregs-graph still has a limit on spill slots
by GHC 04 Mar '19

04 Mar '19

#8657: -fregs-graph still has a limit on spill slots -----------------------------------+--------------------------------------- Reporter: simonmar | Owner: Type: bug | Status: new Priority: normal | Milestone: 7.8.1 Component: Compiler (NCG) | Version: 7.7 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Compile-time crash Unknown/Multiple | Test Case: Difficulty: Unknown | Blocking: Blocked By: | Related Tickets: | -----------------------------------+--------------------------------------- The limit on spill slots was removed for the linear register allocator in 7.8, but not for `-fregs-graph`. SHA-1 fails to compile with `-fregs-graph` in 7.8. Related: #7679 (`-fregs-graph` generates poor code with the new codegen). -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8657> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 18

[GHC] #12919: Equality not used for substitution
by GHC 04 Mar '19

04 Mar '19

#12919: Equality not used for substitution -------------------------------------+------------------------------------- Reporter: int-index | Owner: Type: bug | Status: new Priority: high | Milestone: Component: Compiler | Version: 8.0.1 (Type checker) | Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: GHC rejects Unknown/Multiple | valid program Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- This code {{{ {-# LANGUAGE TypeInType, TypeFamilies, GADTs, ConstraintKinds #-} import Data.Kind data N = Z data V :: N -> Type where VZ :: V Z type family VC (n :: N) :: Type where VC Z = Type type family VF (xs :: V n) (f :: VC n) :: Type where VF VZ f = f data Dict c where Dict :: c => Dict c prop :: xs ~ VZ => Dict (VF xs f ~ f) prop = Dict }}} fails with this error: {{{ Op.hs:20:8: error: • Couldn't match type ‘f’ with ‘VF 'VZ f’ arising from a use of ‘Dict’ ‘f’ is a rigid type variable bound by the type signature for: prop :: forall (xs :: V 'Z) f. xs ~ 'VZ => Dict VF xs f ~ f at Op.hs:19:9 • In the expression: Dict In an equation for ‘prop’: prop = Dict • Relevant bindings include prop :: Dict VF xs f ~ f (bound at Op.hs:20:1) }}} However, if I substitute `xs` with `VZ` (by hand) in the type of `prop`, it compiles. Can be reproduced with GHC 8.0.1 but not HEAD. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12919> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 25

[GHC] #14729: normaliseType is not well-kinded
by GHC 04 Mar '19

04 Mar '19

#14729: normaliseType is not well-kinded -------------------------------------+------------------------------------- Reporter: goldfire | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.2 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Suppose we have {{{ Proxy :: forall k. k -> Type type family F a type instance F Int = Bool }}} and we're trying to normalise the target {{{ Proxy (F Int) x }}} Clearly, for this target to be well kinded, we need `x :: F Int`. Currently, the output of `normaliseType` is `Proxy Bool x`, which is just plain wrong. I need to implement logic like in `flatten_args` on [https://github.com/goldfirere/ghc my branch] in `normaliseType` to fix this. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14729> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 15