April 2014 - ghc-devs

GHC rewrite rules and constructor wrappers?
by Conal Elliott 06 May '14

06 May '14

I'm trying to sort out the relationship of GHC rewrite rules and constructor wrappers. I have rules like > "reify/(:<)" reifyEP (:<) = kPrim VecSP This rule seems to fire for `reifyEP ($W:<)` rather than `reifyEP (:<)`. If I'm tracking (uncertain), `($W:<)` inlines to `(:<)`. Sometimes I'm able to preserve the `$W` form, but sometimes I get the bare form when inlining a definition from another already-compiled module. In those cases, I don't know how to get my rules to fire. Advice and explanation appreciated, including pointers explanations of the role of these wrappers in GHC. -- Conal

2 1

Branch cleanup coming soon - please take inventory
by Austin Seipp 04 May '14

04 May '14

Hello all, I was talking about this on IRC the other day, and it reminded me I wanted to do this a while back. At the moment, we have a *lot* of branches hanging out in the GHC tree: $ git branch -r | grep origin | wc -l 89 $ So I'd like to start deleting all the old ones that are merged and don't need to hang around. A while back I kept a list that was curated of what branches were dead and could be removed. Here it is: https://ghc.haskell.org/trac/ghc/wiki/ActiveBranches This page is a bit out of date, but still many of the comments are accurate, particularly what can already be deleted. I will likely be *deleting* these merged branches soon unless someone speaks up. Note that: - Anything I cannot be 100% certain is merged will stay. - Anything which is effectively 'in-limbo' (like local-gc, or supercompiler) will stay. We also have a lot of WIP branches here too. If you all would all take inventory of your branches and make sure they're catalogued on that page, it would be really nice. Just add it for now - I'll get rid of the old entries once I delete them later. -- Regards, Austin Seipp, Haskell Consultant Well-Typed LLP, http://www.well-typed.com/

2 1

Relocating (some of) GHC's core-libraries to github.com/haskell
by Herbert Valerio Riedel 02 May '14

02 May '14

Hi GHC devs, In accordance with Edward and Austin, I want to move the primary home of the non-GHC specific core-library packages to the http://github.com/haskell/ organization. Specifically, I plan to move the following package Git repositories to the github.com/haskell organization: - array - directory - filepath - old-locale - old-time - process - stm - unix (N.B.: the 'Win32' package already lives at github.com/haskell) I'm not sure yet about the following packages, those are not officially maintained by the committee (@Simon, maybe you can state your preference where you want your packages to be hosted/issue-tracked in the future?) - parallel - deepseq - hpc - hoopl - haskell2010 - haskell98 The practical effects to GHC developers of this switch would be that: - Issue tracking for those packages moves over to GitHub. (the official maintaing entity is the core-library committee) - In ghc.git those packages will be *turned into Git submodules* (i.e. they will be handled just like the other existing 3rd party upstream packages such as 'Win32' or 'Cabal' are already handled[1], see also next item) - Using pull-requests on GitHub is highly recommended for submitting changes which are not urgent (instead of diverging the respective git.haskell.org lagged mirror repo) -- note also that pull-requests will be validated by Travis-CI for multiple GHC versions to help detect compat-breaking changes. However, GHC developers can easily be given push-rights to the respective repos in the github.com/haskell/ organization should this turn out to be a more appropriate workflow. [1]: https://ghc.haskell.org/trac/ghc/wiki/Repositories/Upstream#Modifiyinglibra… PS: 'haddock.git' is planned to move its issue tracking over to GitHub as well, however it's going to be handled slightly different (mostly because haddock is tightly coupled to the GHC API) and will be explained in more detail in a future separate email. Cheers, hvr

6 10

Overloaded record fields: we ought to change the name of FldTy
by Johan Tibell 01 May '14

01 May '14

Hi, I'm currently watching the Skills Matter talks you gave (great!) and I'm slightly worried that we're going to expose beginners (and intermediate users!) to a name called FldTy, which might be OK for a GHC-internal name, but is not very readable. If this is going to be exposed to users, could we name is something readable, like FieldType, Field, or something similar? -- Johan

3 3

Removing -fext-core
by Austin Seipp 30 Apr '14

30 Apr '14

Hello all, Recently I was wondering something: is there any reason to keep -fext-core around? In particular, it's been broken for a while at this point, see GHC bug #5630[1] As far as I'm aware there really aren't any users of it still around these days, at least none working with a modern GHC. And it seems like if people were to need access to Core, they could use a plugin or some such to get direct access to what they want. Simon mentioned removing ExtCore in face of replacing it with IfaceSyn in the compiler. I'm not entirely sure how much work that would be to bring it all up to scratch if people needed it, but maybe it's worth thinking about. One other issue is a notion of semantics which is present in External Core, but we also now have a documented semantics for GHC's core language as well (which has evolved quite a bit), so I don't know how much that matters. So long story short: I don't think anyone is using it, maintaining it, and it seems subsumed by more recent events. Therefore, if nobody objects, I'd vote to remove -fext-core from GHC, unless someone is willing to step up and really maintain it. If you're using it, you should probably speak up soon I'd imagine... [1] https://ghc.haskell.org/trac/ghc/ticket/5630 -- Regards, Austin Seipp, Haskell Consultant Well-Typed LLP, http://www.well-typed.com/

4 4

RE: [commit: ghc] master: Improve implementation of unSubCo_maybe. (a3896ab)
by Simon Peyton Jones 29 Apr '14

29 Apr '14

Richard, Looks good, but *surely* worth a Note [unSubCo] in the file? It's manifestly non-obvious or you'd have done this from day 1. Moreover, the *reason* Conal needed it, and the new optimsations it enables, are also non-obvious and deserve examples! Evernyone: my obsession with Notes is born of the hours I have spent staring at code that has cases I don't understand; or "fixing" something turned out to be there for a good but undocumented reason. Going back to unSubCo, I think the start of the documentation could be if g :: t1 ~R t2 then unSubCo g :: t ~N t2 (if it succeeds at all). But why do we need it? And what optimisations do we get? Thanks Simon | -----Original Message----- | From: ghc-commits [mailto:ghc-commits-bounces@haskell.org] On Behalf Of | git(a)git.haskell.org | Sent: 29 April 2014 00:51 | To: ghc-commits(a)haskell.org | Subject: [commit: ghc] master: Improve implementation of unSubCo_maybe. | (a3896ab) | | Repository : ssh://git@git.haskell.org/ghc | | On branch : master | Link : | http://ghc.haskell.org/trac/ghc/changeset/a3896ab5d2dc88160f710705bf23e6 | e25e327da5/ghc | | >--------------------------------------------------------------- | | commit a3896ab5d2dc88160f710705bf23e6e25e327da5 | Author: Richard Eisenberg <eir(a)cis.upenn.edu> | Date: Mon Apr 28 13:33:13 2014 -0400 | | Improve implementation of unSubCo_maybe. | | This is the result of an email conversation (off list) with | Conal Elliott, who needed a stronger unSubCo_maybe. This | commit adds cases to upgrade the role of a coercion when | recursion is necessary to do say (for example, for a use of | TransCo). As a side effect, more coercion optimizations are | now possible. | | This was not done previously because unSubCo_maybe was used | only during coercion optimization, and the recursive cases | looked to be unlikely. However, adding them can cause no harm. | | unSubCo_maybe is now also exported from Coercion, for use | cases like Conal's. | | | >--------------------------------------------------------------- | | a3896ab5d2dc88160f710705bf23e6e25e327da5 | compiler/types/Coercion.lhs | 20 +++++++++++++++----- | 1 file changed, 15 insertions(+), 5 deletions(-) | | diff --git a/compiler/types/Coercion.lhs b/compiler/types/Coercion.lhs | index af2b2fa..f60fcbd 100644 | --- a/compiler/types/Coercion.lhs | +++ b/compiler/types/Coercion.lhs | @@ -38,7 +38,7 @@ module Coercion ( | splitAppCo_maybe, | splitForAllCo_maybe, | nthRole, tyConRolesX, | - nextRole, | + nextRole, unSubCo_maybe, | | -- ** Coercion variables | mkCoVar, isCoVar, isCoVarType, coVarName, setCoVarName, | setCoVarUnique, @@ -1051,16 +1051,26 @@ maybeSubCo2 r1 r2 co | | -- if co is Nominal, returns it; otherwise, unwraps a SubCo; otherwise, | fails unSubCo_maybe :: Coercion -> Maybe Coercion | +unSubCo_maybe co | + | Nominal <- coercionRole co = Just co | unSubCo_maybe (SubCo co) = Just co | unSubCo_maybe (Refl _ ty) = Just $ Refl Nominal ty -unSubCo_maybe | (TyConAppCo Representational tc cos) | - = do { cos' <- mapM unSubCo_maybe cos | +unSubCo_maybe (TyConAppCo Representational tc coes) | + = do { cos' <- mapM unSubCo_maybe coes | ; return $ TyConAppCo Nominal tc cos' } unSubCo_maybe (UnivCo | Representational ty1 ty2) = Just $ UnivCo Nominal ty1 ty2 | -- We do *not* promote UnivCo Phantom, as that's unsafe. | -- UnivCo Nominal is no more unsafe than UnivCo Representational - | unSubCo_maybe co | - | Nominal <- coercionRole co = Just co | +unSubCo_maybe (TransCo co1 co2) | + = TransCo <$> unSubCo_maybe co1 <*> unSubCo_maybe co2 unSubCo_maybe | +(AppCo co1 co2) | + = AppCo <$> unSubCo_maybe co1 <*> pure co2 unSubCo_maybe (ForAllCo tv | +co) | + = ForAllCo tv <$> unSubCo_maybe co | +unSubCo_maybe (NthCo n co) | + = NthCo n <$> unSubCo_maybe co | +unSubCo_maybe (InstCo co ty) | + = InstCo <$> unSubCo_maybe co <*> pure ty | unSubCo_maybe _ = Nothing | | -- takes any coercion and turns it into a Phantom coercion | | _______________________________________________ | ghc-commits mailing list | ghc-commits(a)haskell.org | http://www.haskell.org/mailman/listinfo/ghc-commits

3 4

optllvm failures
by Daniel Fischer 29 Apr '14

29 Apr '14

I have done a bit of testing with regards to my optllvm failures in the testsuite. Some tests compile but segfault when they are run, others fail to compile, both with HEAD and 7.8.2, but those I tested all work with 7.6.3, although 7.6 gives the warning: You are using a new version of LLVM that hasn't been tested yet! We will try though... I suspect that the LLVM backend thinks I have a different version than I actually have, or maybe the LLVM tools think I have a different CPU than I have: dafis@schwartz:~/JNthPrime> llc --version LLVM (http://llvm.org/) LLVM version 3.2svn Optimized build. Default target: x86_64-unknown-linux-gnu Host CPU: corei7-avx Registered Targets: arm - ARM ppc32 - PowerPC 32 ppc64 - PowerPC 64 thumb - Thumb x86 - 32-bit X86: Pentium-Pro and above x86-64 - 64-bit X86: EM64T and AMD64 dafis@schwartz:~/JNthPrime> opt --version LLVM (http://llvm.org/) LLVM version 3.2svn Optimized build. Default target: x86_64-unknown-linux-gnu Host CPU: corei7-avx But my CPU is a Core i5, not i7. Can somebody confirm or deny that that may be the cause?

4 7

GHC builder: out of memory from ghc during make
by Alain O'Dea 29 Apr '14

29 Apr '14

I am trying to get GHC builder-clients running for x86 and x86_64 GHC on SmartOS. I have been trying to get to the bottom of this memory issue and have had no luck. I've set it aside for the past couple of days to try to get a clear head to attack it afresh tonight. The three latest builds die with the following message: ghc: out of memory (requested 2097152 bytes) compiler/ghc.mk:641: recipe for target 'compiler/stage1/build/DynFlags.o' failed http://haskell.inf.elte.hu/builders/smartos-x86_64-head/16/10.html http://haskell.inf.elte.hu/builders/smartos-x86_64-head/17/10.html http://haskell.inf.elte.hu/builders/smartos-x86_64-head/18/10.html All of these were run interactively from the builder directory: builder-client --do-build Here's my zone setup: image: SmartOS base64 13.4.2 max_physical_memory: 4GB max_swap: 8GB disk_quota: 200GB # ulimit -a core file size (blocks, -c) unlimited data seg size (kbytes, -d) unlimited file size (blocks, -f) unlimited open files (-n) 65536 pipe size (512 bytes, -p) 10 stack size (kbytes, -s) 10240 cpu time (seconds, -t) unlimited max user processes (-u) 12261 virtual memory (kbytes, -v) unlimited It doesn't use anywhere near the 4GB of memory or 8GB of swap allocated to the zone. Total zone RSS (physical memory usage) stays under 200MB for the entire builder-client run. I took regular snaps of memory during the build run and saw no evidence of memory saturation or errors. It's nowhere near hitting swap limits either. I think I've done all I can do to remove system imposed resource limits. This doesn't happen when I run make directly from a shell myself. It looks like a defect in the GHC builder-client itself. If POSIX resources are available on the host (which they are on SmartOS) builder-client sets a 1GB ResourceTotalMemory limit using System.Posix.Resource.setResourceLimit: https://github.com/haskell/ghc-builder/blob/master/client/client.hs#L54 I'm trying a build with the resource limit code commented out. Hopefully that will finally put this to bed. Speaking of bed, I should probably sleep too. If anyone has any other ideas I'd be very happy for the help. Best, Alain

2 1

[Urgent] Please fill out the 2014 HCAR entry for GHC!
by Austin Seipp 28 Apr '14

28 Apr '14

Hello all, I hate to pester so late, but this completely slipped my mind *twice* so I'm afraid I have to! The HCAR entry deadling is rapidly approaching (~May 1st), and before I send off what we have to Mihai, I'd like everyone to pitch in as much as possible. I've already gone ahead and filled out the skeleton. Now I need y'all to fill in details. To that end, I've directly CC'd interested parties, and replicating the message here - if your name is below, please take a quick look over, it shouldn't take you long. * Richard - I copied over the note about the old Explicit Type Application work - do you have any update on what its current status is? I also was under the impression the new kind equalities work *may* go into 7.10, but I haven't heard anything yet. Do please confirm and edit as you see fit. * Iavor - I know you, Eric, and Trevor had worked on Kinds without Data before. Do you know of its current status? I copied the current notes into the page - I also believe you have some recent work involving using an SMT solver in the type-checker, which is quite interesting! If you feel like it, please do mention it, I'm sure people would like to hear. * Thomas - everyone is excited about PartialTypeSignatures I think. Please edit the page and write what you'd like - there's a tiny stub there already. * Edward, Simon, Johan - I'm not sure what you're working on, but I imagine you may have some runtime system or optimization changes up your sleeve. Do feel free to add in things you feel you should mention. * SimonM - I've added a small note for ApplicativeDo. Please feel free to expand or tweak it as you see fit. * Herbert - Do please make the new repository changes clear, and be sure to mention anything else you might be working on (perhaps more integer-gmp improvements?) Please spread the word (by mouth somehow or IRC) - I'll be monitoring the page closely for the next few days and send it to Mihail once it's been expanded upon. -- Regards, Austin Seipp, Haskell Consultant Well-Typed LLP, http://www.well-typed.com/

3 3

Status updates
by Austin Seipp 28 Apr '14

28 Apr '14

Hello all, Here's what's gone on this week: - The HCAR is happening! I sent an email about this earlier, but PLEASE, if you're working on something, add it to the page for the May 2014 report! I won't forget the link this time: https://ghc.haskell.org/trac/ghc/wiki/Status/May14 - ORF is still waiting for a merge. I believe Simon was looking over it this past week - I'm sure we'll talk about it later today. - AMP is still waiting for merge, I think we'll have to coordinate a Happy release as well to match (the fix I have is slightly unsatisfactory). I'll talk to Simon about this today. - My extended-rep-movsb patch is on wip/ermsb, but it needs to be benchmarked and tested before I merge it. Unordered-containers will be my baseline, since Johan tells me it's sensitive to copies like this (it may also need alignment tweaking, on this note). - Many of the Coverity bugs have been fixed, hooray! We will probably fix some more of them - all of the current fixes have gone into 7.8 as well. - I triaged the bug tracker a lot this morning as I'm sure you can all see. I went through a bunch of them, closed tickets, punted, lowered priority, etc. - I've proposed to remove External Core - if you need it, you should speak up *soon*, otherwise I will probably nuke it sometime this week: http://www.haskell.org/pipermail/ghc-devs/2014-April/004791.html The work to do this is on wip/kill-extcore, so again, speak soon or forever hold your peace. - The 7.8.3 milestone has been shaped up. Please take a look at the outstanding tickets: https://ghc.haskell.org/trac/ghc/milestone/7.8.3 https://ghc.haskell.org/trac/ghc/query?status=infoneeded&status=merge&statu… Note that *not* all of these will be fixed - but these are the ones we will pay attention to for the 7.8.3 release. So do have a look over and if you think you can fix something, please do so! ------------- Some things I've been thinking about: - One really annoying thing about the way we generate source-distributions is that we must do a build first, but this practically negates the need for a sdist! Joachim in particular finds this annoying too for the Debian builds. Ideally we could build a source dist immediately from the repository, and then build from that. This has the benefit the CI system is always testing something that a user can build and make themselves. The problem here is that we need to run Happy and Alex on the parser, which is part of the stage1/stage2 build. Although it's kind of a hack, I think we can do this differently, but I haven't tried it. In particular, they do not depend on any CPP, so we could just immediately generate them up front in the build process. This is kind of a gross hack from the build-system's point of view, but it would make this task much easier. - For right now, I'd like to go with Stackage for building a subset of packages against HEAD I think. Herbert recently uploaded fresh PPAs for GHC HEAD recently, so I think we can already begin the process of testing things and seeing how/why they break. Of course, not everyone will always need to pay attention to this, and sometimes it *will* be broken (c.f. upcoming AMP changes). But in the long run I think this will be very useful, and we can also set it up for the 7.8 branch. I'll be spinning up some new machines today and reply with some updates on this later. -- Regards, Austin Seipp, Haskell Consultant Well-Typed LLP, http://www.well-typed.com/

1 0