Posts for the month of October 2014

GHC Weekly News - 2014/10/31 (Halloween Edition)

Hello *,

Welcome to the GHC Weekly news. And it's just in time before you go out and eat lots of candy and food.

  • David Feuer and Joachim Brietner spent some time this past week talking about more optimizations for Haskell code for fusing code, and creating better consumers and producers. This work includes optimizations of "One shot lambdas" (lambdas used at most once) and Call Arity, which was implemented by Joachim at Microsoft Research. The thread is here -

The current situation is that Call Arity and One shot analysis tends to have good combined results for exploiting more fusion opportunities, but sometimes these backfire. As a result, Joachim and David have worked on improving the situation - particularly by letting programmers help with a new oneShot primitive (in Phab:D392 & Phab:D393).

  • Herbert Valerio Riedel opened up a discussion about the origins of code contributions. In short, we'd like to have some stronger ground to stand on in the face of contributions from contributors - the origin of a change and its legal status is a bit nebulous. The thread is here:

Overall, there's been a lot of separate points made, including CLAs (unlikely), "Developer Certificates of Origin" a la the Linux Kernel, and reworking how we mark up header files, and keep track of GHC's long history of authors. If you work on a big project where some of these concerns are real, we'd like to hear what you have to say!

  • Gintautas Milauskas has done some fantastic work for GHC on Windows lately, including fixing tests, improving the build, and making things a lot more reasonable to use. With his work, we hope GHC 7.10 will finally ship an updated MinGW compiler (a long requested feature), and have a lot of good fixes for windows. Thank you, Gintautas!
  • And on that note, the call for Windows developers rages on - it looks like Gintautaus, Austin, Simon, and others will be meeting to discuss the best way to tackle our current woes. Are you a Windows user? Please give us input - having input is a crucial part of the decision making process, so let us know.
  • Jan Stolarek had a question about core2core - a lot of questions, in fact. What's the difference between demand, strictness, and cardinality analylsis? Does the demand analyzer change things? And what's going on in some of the implementation? A good read if you're interested in deep GHC optimization magic:
  • Peter Wortmann has put up the new DWARF generation patches for review, in Phab:D396. This is one of the major components we still plan on landing in 7.10, and with a few weeks to spare, it looks like we can make sure it's merged for the STABLE freeze!
  • There have been a lot of good changes in the tree this past week:

Thanks to Michael Orlitzky, we plan on adding doctest examples to more modules in 7.10, and increase that coverage further. This is *really* important work, but very low fruit - thanks a ton Michael!

Data.Bifunctor is now inside base! (Phab:D336)

atomicModifyIORef' has been optimized with excellent speedups (as much as 1.7x to 1.4x, depending on the RTS used), thanks to some older work by Patrick Palka (Phab:D315). GHC's internals have been reworked to unwire Integer from GHC, leading not only to a code cleanup, but laying the foundation for further GMP (and non-GMP!) related Integer improvements (Phab:D351).

David Feuer and Joachim have been relentless in improving fusion opportunities, including the performance of take, isSuffixOf, and more prelude improvements, spread over nearly half a dozen patches. And this doesn't even include the work on improving oneShot or Call Arity!

In a slight change to base semantics, David Feuer also finally fixed #9236. This is a change that can expose latent bugs in your program (as it did for Haddock), so be sure to test thoroughly with 7.10 (Phab:D327).

GHC now has support for a new __GLASGOW_HASKELL_TH__ macro, primarily useful for testing bootstrap compilers, or compilers which don't support GHCi.

And there have been many closed tickets: #9549, #9593, #9720, #9031, #8345, #9439, #9435, #8825, #9006, #9417, #9727, #2104, #9676, #2628, #9510, #9740, #9734, #9367, #9726, #7984, #9230, #9681, #9747, and #9236.

GHC Weekly News - 2014/10/24

Hi *,

Welcome to the weekly GHC news. This one will be short this week, as the preceding one occurred only on Monday - but we'll be going with Fridays from now on, so next week we'll hopefully see a longer list.

  • GHC 7.8.4 tickets have been in waiting, and the RC will be soon after Austin finishes some final merges and tests on his branch. We have not committed a time for the release after the RC, yet we would like people to please seriously test and immediately report any major showstoppers - or alert us of ones we missed.
  • For the GHC 7.10 release, one of the major features we planned to try and merge was DWARF debugging information. This is actually a small component of larger ongoing work, including adding stack traces to Haskell executables. While, unfortunately, not all the work can be merged, we talked with Peter, and made a plan: our hope is to get Phab:D169 merged, which lays all the groundwork, followed by DWARF debugging information in the code generators. This will allow tools like gdb or other extensible debuggers to analyze C-- IR accurately for compiled executables. Peter has written up a wiki page, available at SourceNotes, describing the design. We hope to land all the core infrastructure in Phab:D169 soon, followed by DWARF information for the Native Code Generator, all for 7.10.1
  • This past week, a discussion sort of organically started on the #ghc IRC channel about the future of the LLVM backend. GHC's backend is buggy, has no control over LLVM versions, and breaks frequently with new versions. This all significantly impacts users, and relegates the backend to a second class citizen. After some discussion, Austin wrote up a proposal for a improved backend, and wrangled several other people to help. The current plan is to try an execute this by GHC 7.12, with the goal of making the LLVM backend Tier 1 for major supported platforms.
  • You may notice is now responds slightly faster in some cases - we've activated a caching layer (CloudFlare) on the site, so hopefully things should be a little more smooth.

Closed tickets this week: #9684, #9692, #9038, #9679, #9537, #1473.

GHC Weekly News - 2014/10/20

Hi *,

It's been a few weeks since the last message - and I apologize! We actually are changing the posting time to be Friday now, so hopefully this situation will be corrected preeeeetty quickly from this point forward, and hopefully will give better context over the course of a weekly discussion.

That said, let's begin!

  • We've seen plenty of changes to GHC itself in the past few weeks. Some of the highlights include:
    • Some changes to help make Prelude combinators fuse better. David Feuer has been leading a lot of this work, and it's been quite fruitful, with several new things now fusing (like takeWhile, scanl, scanr, and mapAccumL.
    • Relatedly, Data.List.Inits should be far faster thanks to David Feuer (ref: Phab:D329).
    • The testsuite driver now has preliminary support for Python 3 - which should be useful for platforms out there that sport it, and ones that will use it as the default eventually (such as Fedora 22, possibly).
    • Some of the initial work by Edward Yang to remove HEAP_ALLOCED from the GHC runtime system has landed. Briefly, HEAP_ALLOCED is a check the RTS uses to determine if some address is part of the dynamic heap - but the check is a bit costly. Edward's full patchset hopes to remove this with an 8% speedup or so on average.
    • GHC now has a new macro, __GLASGOW_HASKELL_PATCHLEVEL__, which will allow you to determine the point-level release of the GHC you're using. This has been a requested feature in the past we were a little hesitant of adding, but Herbert went through and did it for us. (Ref: Phab:D66)
    • Template Haskell now supports LINE pragmas, thanks to Eric Mertens (ref: Phab:D299).
    • Sergei Trofimovich revived libbfd debugging support for the runtime system linker, which should be of use to several daring souls out there (ref: Phab:D193).
    • Several patches from Gintautas Miliauskas has improved the usability of msys and the testsuite on Windows - and he's not done yet!
    • A few improvements to the x86 code generator were committed by Reid Barton and Herbert Valerio Riedel, improving size/space for certain cases (ref: Phab:D320, Phab:D163).

and more besides that, including some linker improvements, and general cleanups as usual.

  • The mailing list has been busy (as usual), with some discussions including:
    • Austin posted some discussion about the tentative 7.10.1 plans - we're still hoping these are accurate, so take note! We hope to freeze mid-November, and release Feburary 2015! [1]
    • Austin also called for some feedback: GHC HQ has become convinced a 7.8.4 release is needed to fix some showstoppers - so please let him know soon if you're totally incapable of using 7.8 for something! [2]
    • Alan Zimmerman has asked for some feedback on his proposed "AST Annotations", which will hopefully allow GHC API clients to add richer annotations to GHC's syntactic representations. The motivation is for refactoring tools like HaRe - and I'm sure all input would be appreciated. [3]
    • Chris done sparked off a discussion about making GHCi awesomer, and I'm sure everyone can appreciate that! In particular, Chris wanted to discuss programmatic means of controlling GHCi itself, and naturally we need to ask - is the current API not enough, and why? [4]
    • Yuras Shumovich has implemented a proposal for allowing the Haskell FFI to support C structures natively as return values - this involves interfacing with C ABI rules to properly support structure layout. While Yuras has an initial implementation in Phab:D252, some questions about the feature - including its implementation complexity - remain before it gets merged. [5]
    • Richard Eisenberg made a modest request: can Phabricator patches have a 'roadmap', so people can instruct reviewers how to read a diff? The answer: yes, and it should be relatively easy to implement, and Austin can do so Real Soon Now™. [6]
    • Ben Gamari started a big discussion about one-shot event semantics in the I/O manager, with a lot of replies concerning not only the bug, but machines to test the actual change on. With any luck, Ben's fix for the I/O manager and a test machine should come quickly enough. [7]
    • Herbert Valerio Riedel opened an RFC: Should we look into using AsciiDoc for GHC? Historically, GHC's documentation has been written using DocBook, a verbose but very precise and unambiguous documentation format. However, AsciiDoc offers a much nicer markup language, while retaining DocBook support. In short, it looks like GHC may get a much more clean user manual soon. [8]
    • Yuras opened another discussion: Should we namespace proposals we create on our wiki? What seems uncontroversial can lead to surprising discussion, and the results were mixed this time it seems. [9]
    • Geoff Mainland stepped up and fixed Data Parallel Haskell to work with a new version of vector and GHC. Austin had disabled DPH a few weeks prior due to its difficulty to upgrade, and divergent source trees. With 7.10, GHC will hopefully ship a more modern vector and dph to boot.
    • Austin asks: can we warn on tabs by default for GHC 7.10? It's an easy change and a minor one - but we should at least ask first. Vote now! [10]
    • Philip Hölzenspies opens up a discussion about Uniques in GHC, and their impact on the compilers current design. Philip has a hopeful design to redo Unique values in GHC, and a patch to support it: Phab:D323. [11]
    • Richard Eisenberg asks: can we somehow integrate GitHub into our development process? While GitHub doesn't have as many nice tools as something like Phabricator, it has a very high inertia factor, and Richard is interested in making the 'first step' as easy as possible for newbies. Discussions about Phab<->GitHub integrations were afoot, as well as general discussion about contributor needs. There were a lot of points brought up, but the conversation has slightly dried up now - but will surely be revived again. [12]

And now, look at all these tickets we closed! Including: #9658, #9094, #9356, #9604, #9680, #9689, #9670, #9345, #9695, #9639, #9296, #9377, #9184, #9684.