June 2018 - ghc-tickets

[GHC] #14909: Change default armhf target to a newer architecture
by GHC 21 Oct '18

21 Oct '18

#14909: Change default armhf target to a newer architecture -------------------------------------+------------------------------------- Reporter: Phyx- | Owner: (none) Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.5 (CodeGen) | Keywords: | Operating System: Unknown/Multiple Architecture: arm | Type of failure: None/Unknown Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Currently the default [https://github.com/ghc/ghc/blob/60b06456ddef08bd8a8a47497a6cbefbb5e359fb /llvm-targets#L4 llvm-targets] file specifies the default `arm-unknown- linux-gnueabihf` target to be using `-mcpu=arm1176jzf-s`. Which is an `Armv6` architecture. This has the problem that it's generating memory barriers using `CP15` instructions that are deprecated in `Armv7` and removed in `Armv8`. Because of this there's no way to run the binaries being produced by GHC HQ on Armv8 CPUs without having kernel support for `CP15 Barrier emulation.` which is slow. We're also not taking advantage of new Armv8 instructions this way. The debian folks have a bug report about this as well https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=864847 This means that GHC cannot work on most boards people have/can easily buy today. I think the default target should be an `Armv7` architecture at the very least. But realistically `Armv8` would probably be ok as well as I doubt many people use GHC to compile on older cores. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14909> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 4

[GHC] #14461: Reuse free variable lists through nested closures
by GHC 21 Oct '18

21 Oct '18

#14461: Reuse free variable lists through nested closures -------------------------------------+------------------------------------- Reporter: tdammers | Owner: (none) Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: 7258 Differential Rev(s): | Wiki Page: NestedClosures -------------------------------------+------------------------------------- Consider [wiki:NestedClosures] and [ticket:7258]; essentially, deeply nested closures exhibit quadratic compiler performance due to the fact that when allocating registers, each nesting level will have the compiler unpack the entire parent closure and then re-pack the variables into the child closure. To solve this, check if the parent closure can be carried along wholesale, and pull variables from there so that the repackaging can be bypassed. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14461> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 20

[GHC] #14971: Use appropriatly sized comparison instruction for small values.
by GHC 20 Oct '18

20 Oct '18

#14971: Use appropriatly sized comparison instruction for small values. -------------------------------------+------------------------------------- Reporter: AndreasK | Owner: (none) Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: (CodeGen) | 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: -------------------------------------+------------------------------------- GHC currently defaults all comparisons originating from Cmm switch statements to 64bit on x64. This incurs a small overhead in instruction size. Fixing this manually gave a speedup of ~1,5% in microbenchmarks. In detail we generate Cmm of the form: {{{ _s8Dg::P64 = R1; _c8EF::P64 = _s8Dg::P64 & 7; switch [1 .. 2] _c8EF::P64 { case 1 : goto c8Ey; case 2 : goto c8EC; } }}} Which results in assembly like: {{{ andl $7,%ebx cmpq $1,%rbx }}} It's obvious that c8EF fits into a byte, but is sized up to 64 bits. Changing this would enable us to use cmpl instead of cmpq and save us a byte on each comparison. While this isn't major in my microbenchmarks it resultet in a speedup of ~1,5% for such constructs in inner loops. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14971> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 3

[GHC] #15258: Implement CMOV support.
by GHC 20 Oct '18

20 Oct '18

#15258: Implement CMOV support. -------------------------------------+------------------------------------- Reporter: AndreasK | Owner: (none) Type: task | Status: new Priority: normal | Milestone: 8.6.1 Component: Compiler | Version: 8.4.3 Keywords: CodeGen | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15258> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 2

[GHC] #14830: Use test instead of cmp for comparison against zero.
by GHC 20 Oct '18

20 Oct '18

#14830: Use test instead of cmp for comparison against zero. -------------------------------------+------------------------------------- Reporter: AndreasK | Owner: (none) Type: task | Status: new Priority: low | Milestone: Component: Compiler | Version: (NCG) | 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: -------------------------------------+------------------------------------- For comparisons `test r, r` and `cmp r, $0` are interchangeable at least on x86. However using test leads to slightly smaller code. (Saves a byte per comparison). A good starting point is stmtToInstrs in compiler/nativeGen/X86/CodeGen starting with how CmmCondBranch is compiled to Instructions. From there it should be not too hard to track down the required changes. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14830> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 3

[GHC] #15234: WARNING in hptSomeThingsBelowUs when using a source plugin
by GHC 19 Oct '18

19 Oct '18

#15234: WARNING in hptSomeThingsBelowUs when using a source plugin -------------------------------------+------------------------------------- Reporter: mpickering | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: 8.6.1 Component: Compiler | Version: 8.4.3 Keywords: | Operating System: Unknown/Multiple SourcePlugins, plugins | Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- When using a source plugin with a stage2 compiler built with the devel2 flavour, there are a fair few warnings about hptSomeThingsBelowUs. {{{ WARNING in hptSomeThingsBelowUs missing module CoercePlugin Probable cause: out-of-date interface files }}} I don't know if this just started happening or has been going on for a while. Is this warning cause for concern? Everything appears to work as expected despite this hiccup. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15234> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 12

[GHC] #15328: cpphs: internal error: evacuate(static): strange closure type 8440
by GHC 18 Oct '18

18 Oct '18

#15328: cpphs: internal error: evacuate(static): strange closure type 8440 -------------------------------------+------------------------------------- Reporter: cnmne | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: 8.6.1 Component: Package | Version: 8.4.3 system | Keywords: cabal, cpphs, | Operating System: MacOS X strange closure | Type of failure: Building GHC Architecture: x86 | failed Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Attempting to install Agda via Cabal fails. I've installed cpphs through Cabal and {{{~/Library/Haskell/bin}}} is in my path. {{{~/.cabal/config}}} is default, but in {{{/Library/Frameworks/GHC.framework/Versions/8.4.3-x86_64/usr/lib/ghc-8.4.3 /settings}}} I have changed only {{{("C compiler supports -no- pie","NO")}}} (from "YES") because that was giving me errors earlier. The log states: "Please report this as a GHC bug..." so here I am :] -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15328> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 6

[GHC] #15112: ghc 8.4.2 on OS X: clang: warning: argument unused during compilation: '-nopie'
by GHC 18 Oct '18

18 Oct '18

#15112: ghc 8.4.2 on OS X: clang: warning: argument unused during compilation: '-nopie' -------------------------------------+------------------------------------- Reporter: elaforge | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: 8.6.1 Component: Compiler | Version: 8.4.2 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: -------------------------------------+------------------------------------- {{{ % cat Nopie.hs module Nopie where main :: IO () main = putStrLn "hi" % ghc -fforce-recomp -c -fhpc Nopie.hs clang: warning: argument unused during compilation: '-nopie' [-Wunused- command-line-argument] clang: warning: argument unused during compilation: '-nopie' [-Wunused- command-line-argument] }}} This happens for every source file when -fhpc is given, but also happens on links even when -fhpc isn't given. With -v, I see a call to gcc with -no-pie. When I run the same compile with ghc 8.4.1, I see the same call to gcc, only without that -no-pie flag, so this is new. It seems to be a harmless warning, but really clutters up the output. Incidentally, it looks like the particular spelling of -no-pie was added to LLVM a year ago due to ghc: https://reviews.llvm.org/D35462 Not sure if that's relevant. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15112> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 17

[GHC] #11307: Regresssion: parsing type operators
by GHC 17 Oct '18

17 Oct '18

#11307: Regresssion: parsing type operators -------------------------------------+------------------------------------- Reporter: alanz | Owner: Type: bug | 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 Rev(s): | Wiki Page: -------------------------------------+------------------------------------- GHC 7.10.3 will parse (but not type-check) the following {{{#!hs {-# LANGUAGE TypeOperators #-} type family (r1 :++: r2); infixr 5 :++: type instance r :++: Nil = r type instance r1 :++: r2 :> a = (r1 :++: r2) :> a }}} Current master (c7830bdb14e0a85a78617314156d101155fdf7aa) fails with {{{#!hs /tmp/Foo.hs:5:15: error: Malformed head of type or class declaration: r1 :++: r2 :> a }}} @simonpj comment on ghc-devs mailing list > `| type instance r1 :++: r2 :> a = (r1 :++: r2) :> a` > > What do you expect this to mean? I suppose you could hope that GHC will > unravel the fixity of :++: and :>, to determine whether you are giving an > instance for :++: or for :>? > > That sounds reasonable, but it's not trivial. > > Currently the LHS of a 'type instance' decl is ultimately a TyFamEqn, and > it looks like this: > > {{{#!hs > data TyFamEqn name pats > = TyFamEqn > { tfe_tycon :: Located name > , tfe_pats :: pats > , tfe_rhs :: LHsType name } > }}} > > So we've already decided (in the parser) what the type-function name is. > But we can't do that in this case, because the parser doesn't understand > fixity. > > To deal with this we'd need to change to > > {{{#!hs > data TyFamEqn name pats > = TyFamEqn > { tfe_lhs :: LHSType name > , tfe_rhs :: LHsType name } > }}} > > so that the LHS was just a type. Now the renamer can re-jiggle its > fixities in the usual way, and only in the type checker will we need to > decide exactly which type it's an instance of. Easy! > > This might be a good change to make! It's a bit like in ClsInstDecl, you'll > see that the instance cid_poly_ty is just a LHsSigType, i.e. not decomposed > into which class it is. > -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11307> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 6

[GHC] #15278: Add -Werror=compat, enable it in the testsuite
by GHC 15 Oct '18

15 Oct '18

#15278: Add -Werror=compat, enable it in the testsuite -------------------------------------+------------------------------------- Reporter: int-index | Owner: (none) Type: task | Status: new Priority: normal | Milestone: 8.8.1 Component: Compiler | Version: 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: -------------------------------------+------------------------------------- = Proposed change Add a flag `-Werror=compat` to GHC which would have the effect of `-Werror=x -Werror=y ...` where `x, y, ...` are warnings from `minusWcompatOpts`. Enable `-Wcompat -Werror=compat` in the testsuite. = Motivation 1. `-Werror=compat` would be a convenient shorthand to ensure forwards- compatibility of code. I can imagine Haskell libs enabling it on their CI. 2. Enabling `-Wcompat -Werror=compat` in the testsuite would allow us to easily see the impact that a new warning has on code. It also means that in the period between adding the warning and making the actual breaking change, all new test cases that are being added to the testsuite will be forwards-compatible. This is good because it will make the actual breaking change contain less irrelevant testsuite updates. The idea came up during my work on Phab:D4834 which defines a new warning and adds it to `-Wcompat`. Ryan Scott raised the following concern: > There are likely many tests in the testsuite that will now fail due to this change—I'm unclear if you're intending to fix them by explicitly quantifying their kind variables, or leaving the new error message as-is. My intention was to update the test cases, but it turned out that I couldn't even locate these test cases without `make EXTRA_HC_OPTS="-fwarn- implicit-kind-vars -Werror=implicit-kind-vars"`. And even if I updated the testsuite, nothing would prevent new test cases that are not forwards- compatible, because the warning isn't enabled by default (and without `-Werror=compat`, enabling `-Wcompat` would not cause test failures in all circumstances). = Implementation plan 1. Add new flags: `-Werror=compat`, `-Wno-error=compat`, `-Wwarn=compat`. 2. Document these flags in the User's Guide. 3. Enable `-Wcompat -Werror=compat` in the testsuite by default and fix all tests that break. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15278> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 4