February 2020 - ghc-devs

The future of Phabricator
by Ben Gamari 18 Sep '21

18 Sep '21

TL;DR. For several reasons I think we should consider alternatives to Phabricator. My view is that GitLab seems like the best option. Hello everyone, Over the past year I have been growing increasingly weary of our continued dependence on Phabricator. Without a doubt, its code review interface is the best I have used. However, for a myriad of reasons I am recently questioning whether it is still the best tool for GHC's needs. # The problem There are a number of reasons why I am currently uncertain about Phabricator. For one, at this point we have no options for support in the event that something goes wrong as the company responsible for Phabricator, Phacility, has closed their support channels to non-paying customers. Furthermore, in the past year or two Phacility has been placing their development resources in the parts their customers pay them for, which appear to be much different that the parts that we actively use. For this reason, some parts that we rely on seem oddly half-finished. This concern was recently underscored by some rather unfortunate misfeatures in Harbormaster which resulted in broken CI after the Hadrian merge and now apparent bugs which have made it difficult to migrate to the CircleCI integration we previously prepared. Perhaps most importantly, in our recent development priorities survey our use of Phabricator was the most common complaint by a fair margin, both in case of respondents who have contributed patches and those who have not. On the whole, past contributors and potential future contributors alike have strongly indicated that they want a more Git-like experience. Of course, this wasn't terribly surprising; this is just the most recent case where contributors have made this preference known. Frankly, in a sense it is hard to blame them. The fact that users need to install a PHP tool, Arcanist, to contribute anything but documentation patches has always seemed like unnecessary friction to me and I would be quite happy to be rid of it. Indeed we have had a quite healthy number of GitHub documentation patches since we started accepting them. This makes me thing that there may indeed be potential contributoes that we are leaving on the table. # What to do With Rackspace support ending at the end of year, now may be a good time to consider whether we really want to continue on this road. Phabricator is great at code review but I am less and less certain that it is worth the maintenance uncertainty and potential lost contributors that it costs. Moreover, good alternatives seem closer at-hand than they were when we deployed Phabricator. ## Move to GitHub When people complain about our infrastructure, they often use GitHub as the example of what they would like to see. However, realistically I have a hard time seeing GitHub as a viable option. Its feature set is simply insufficient enough to handle the needs of a larger project like GHC without significant external tooling (as seen in the case of Rust-lang). The concrete reasons have been well-documented in previous discussions but, to summarize, * its review functionality is extremely painful to use with larger patches * rebased patches lead to extreme confusion and lost review comments * it lacks support for pre-receive hooks, which serve as a last line of defense to prevent unintended submodule changes * its inability to deal with external tickets is problematic * there is essentially no possibility that we could eventually migrate GHC's tickets to GitHub's ticket tracker without considerable data loss (much less manage the ticket load that would result), meaning that we would forever be stuck with maintaining Trac. * on a personal note, its search functionality has often left me empty-handed On the whole, these issues seem pretty hard to surmount. ## Move to GitLab In using GitLab for another project over the past months I have been positively surprised by its quality. It handles rebased merge requests far better than GitHub, has functional search, and quite a usable review interface. Furthermore, upstream has been extremely responsive to suggestions for improvement [1]. Even out-of-the-box it seems to be flexible enough to accommodate our needs, supporting integration with external issue trackers, having reasonable release management features, and support for code owners to automate review triaging (replacing much of the functionality of Phabricator's Herald). Finally, other FOSS projects' [3] recent migrations from Phabrictor to GitLab have shown that GitLab-the-company is quite willing to offer help when needed. I took some time this weekend to try setting up a quick GHC instance [2] to play around with. Even after just a few hours of playing around I think the result is surprisingly usable. Out of curiosity I also played around with importing some tickets from Trac (building on Matt Pickering's Trac-to-Maniphest migration tool). With relatively little effort I was even able to get nearly all of our tickets (as of 9 months ago) imported while preserving ticket numbers (although there are naturally a few wrinkles that would need to be worked out). Naturally, I think we should treat the question of ticket tracker migration as an orthogonal one to code review, but it is good to know that this is possible. ## Continue with Phabricator Continuing with Phabricator is of course an option. Its review functionality is great and it has served us reasonably well. However, compared to GitLab and even GitHub of today its features seem less distinguished than they once did. Moreover, the prospect of having to maintain a largely stagnant product with no support strikes me as a slightly dangerous game to play. Working around the issues we have recently encountered has already cost a non-negligible amount of time. # The bottom line If it wasn't clear already, I think that we should strongly consider a move to GitLab. At this point it seems clear that it isn't going to vanish, has a steady pace of development, is featureful, and available. However, these are just my thoughts. What do you think? Cheers, - Ben [1] 11.4 will ship with a file tree view in the code review interface, which I reported (https://gitlab.com/gitlab-org/gitlab-ce/issues/46474) as being is one of the Phabricator features I missed the most during review [2] https://gitlab.home.smart-cactus.org/ghc/ghc/issues/14641 [3] The GNOME and freedesktop.org projects have recently migrated, the former from a hodge-podge of self-hosted services and the latter from Phabricator

17 40

New implementation for `ImpredicativeTypes`
by Alejandro Serrano Mena 08 Sep '21

08 Sep '21

Hi all, As I mentioned some time ago, we have been busy working on a new implementation of `ImpredicativeTypes` for GHC. I am very thankful to everybody who back then sent us examples of impredicativity which would be nice to support, as far as we know this branch supports all of them! :) If you want to try it, at https://gitlab.haskell.org/trupill/ghc/commit/a3f95a0fe0f647702fd7225fa719a… you can find the result of the pipeline, which includes builds for several platforms (click on the "Artifacts" button, the one which looks like a cloud, to get them). The code is being developed at https://gitlab.haskell.org/trupill/ghc. Any code should run *unchanged* except for some eta-expansion required for some specific usage patterns of higher-rank types. Please don't hesitate to ask any questions or clarifications about it. A merge request for tracking this can be found at https://gitlab.haskell.org/ghc/ghc/merge_requests/1659 Kind regards, Alejandro

8 23

Blocking MVar# primops not performing stack checks?
by Shao, Cheng 04 Mar '20

04 Mar '20

Hi all, When an MVar# primop blocks, it jumps to a function in HeapStackCheck.cmm which pushes a RET_SMALL stack frame before returning to the scheduler (e.g. the takeMVar# primop jumps to stg_block_takemvar for stack adjustment). But these functions directly bump Sp without checking for possible stack overflow, I wonder if it is a bug? Cheers, Cheng

3 2

Feasibility of native RTS support for continuations?
by Alexis King 03 Mar '20

03 Mar '20

Hi all, tl;dr: I want to try to implement native support for capturing slices of the RTS stack as a personal experiment; please tell me the obstacles I am likely to run into. Much more context follows. --- I have been working on an implementation of an algebraic effect system that uses unsafe primops to be as performant as possible. However, the unavoidable need to CPS the entire program balloons heap allocation. Consider the following definition: f a b = g a >>= \c -> h (b + c) Assume `g` and `h` are not inlined. If the monad used is IO, this will be compiled efficiently: the result of `g` is returned on the stack, and no closure needs to be allocated for the lambda. However, if the monad supports capturing the continuation, the above definition must be CPS’d. After inlining, we end up with f a b = \k -> let lvl = \c -> h (b + c) k in g a lvl which must allocate a closure on the heap. This is frustrating, as it must happen for every call to a non-inlined monadic operation, even if that operation never captures the continuation. In an algebraic effect system, there are many shortcuts that avoid the need to capture the continuation, and under my implementation, large swaths of code never do so. I’ve managed to exploit that to get some savings, but I can’t escape the need to allocate all these closures. This motivates my question: how difficult would it be to allow capturing a portion of the RTS call stack directly? My requirements are fairly minimal, as continuations go: 1. Capturing a continuation is only legal from within a strict state thread (i.e. IO or strict ST). 2. The continuation is captured up to a prompt, which would be a new kind of RTS stack frame. Prompts are not tagged, so there is only ever exactly one prompt active at any time (which may be the root prompt). 3. Capturing a continuation is unsafe. The behavior of capturing a continuation is undefined if the current prompt was not created by the current state thread (and it is never legal to capture up to the root prompt). 4. Applying a continuation is unsafe. Captured continuations return `Any`, and type safety is the caller’s obligation. 5. Continuations are “functional,” which is to say applying them does not trigger any additional stack unwinding. This minimal support for primitive continuation capturing would be enough to support my efficient, safe delimited control implementation. In my ignorant mind, implementing this ought to be as simple as defining two new primops, reset# :: (State# s -> (# State# s, a #)) -> State# s -> (# State# s, a #) shift# :: ((a -> State# s -> (# State# s, Any #)) -> State# s -> (# State# s, Any #)) -> State# s -> (# State# s, a #) where reset# pushes a new prompt frame and shift# captures a slice of the RTS stack up to that frame and copies it into the heap. Restoring a continuation would copy all the captured frames onto the end of the current stack. Sounds simple enough! I would like to experiment with implementing something like this myself, just to see if it would really work, but somehow I doubt it is actually as simple as it sounds. Minor complications are fine, but what worries me are major obstacles I haven’t found in my limited attempts to learn about the RTS. So far, I’ve read the old “The New GHC/Hugs Runtime System” paper, which still seems mostly accurate from a high level, though I imagine many details have changed since then. I’ve also read the Commentary/RTS/Storage/Stack wiki page, and I’ve peeked at some of the RTS source code. I’ve also skimmed a handful of other papers and wiki pages, but it’s hard to know what I’m looking for. Can anyone point me to better resources or give me some insight into what will be hard? Thanks in advance, Alexis

4 15

Re: [ANNOUNCE] GHC 8.8.3 is now available
by Ben Gamari 29 Feb '20

29 Feb '20

Ben Gamari <ben(a)well-typed.com> writes: > --text follows this line-- > <#part sign=pgpmime> > > Hello everyone, > > The GHC team is proud to announce the release of GHC 8.8.3. The source > distribution, binary distributions, and documentation are available at > > https://downloads.haskell.org/~ghc/8.8.3 > > Release notes are also available [1]. > Unfortunately there has been a slight mix-up in the binary distributions associated with this release and I will need to do a re-spin. I have pulled down the binary distributions for the time being. They should be back in a few hours. Many apologies for the confusion! Cheers, - Ben

2 2

hsinspect-0.0.11 released
by Tseen She 28 Feb '20

28 Feb '20

Hello ghc developers, With many thanks to Daniel Grober, I have released 0.0.11 of hsinspect to Hackage, which now supports ghc versions 8.4.x, 8.6.x and 8.8.x. I would like to bring this tool to the attention of this community of ghc developers because it is using the ghc api exclusively to provide semantic information about .hs files, so it may be of interest to you to see a tool using your api in the wild. And selfishly, I learnt so much from Daniel's reviews of two short functions recently (which were subject to a change in ghc behavour in 8.8) that I would very much welcome further reviews by this group. The source code is split across three parts: The plugin (an alternative to hie-bios, which didn't work well for me): https://gitlab.com/tseenshe/hsinspect/-/blob/master/plugin/GhcFlags/Plugin.… The library, one file per feature: https://gitlab.com/tseenshe/hsinspect/-/tree/master/library/HsInspect and the Executable (setup, then just call the relevant library): https://gitlab.com/tseenshe/hsinspect/-/blob/master/exe/Main.hs My usecase is to provide semantic information for my Emacs mode, which is a completely from-scratch haskell-mode that I've been working on in my free time for the past year: https://gitlab.com/tseenshe/haskell-tng.el For anybody curious about the wider project, there are some screenshots of the features in the README, which also fully justifies why I undertook this endeavour: i.e. why I don't believe I could achieve what I wanted in haskell-mode or LSP. I am content for this tool to be used just by myself, my work colleagues, and anybody who is interested in contributing. Best regards, Tseen She

1 0

8.8.x change in lookupModule / findModule semantics?
by Tseen She 28 Feb '20

28 Feb '20

Hi ghc-devs, When upgrading my dev tool to use the 8.8.x api I have found that calls to lookupModule / findModule are failing on uncompilable code, whereas these calls succeeded on 8.4 and 8.6. Unlike another question that I have asked [1] which seems related, I don't intend to do any typechecking here, I only want access to the ModuleInfo. Has this been an intentional change? Is there a way to recover the original behaviour? If I cannot get this to work then it means that a new feature introduced in https://gitlab.haskell.org/ghc/ghc/merge_requests/1541 that I was quite excited about is unusable :-( [1] https://mail.haskell.org/pipermail/ghc-devs/2020-February/018655.html

2 12

Confused about PAP object layout
by Ömer Sinan Ağacan 27 Feb '20

27 Feb '20

Hi Simon, In this code: (slightly simplified) StgPtr scavenge_PAP (StgPAP *pap) { evacuate(&pap->fun); return scavenge_PAP_payload (pap->fun, pap->payload, pap->n_args); } StgPtr scavenge_PAP_payload (StgClosure *fun, StgClosure **payload, StgWord size) { const StgFunInfoTable *fun_info = get_fun_itbl(UNTAG_CONST_CLOSURE(fun)); StgPtr p = (StgPtr)payload; switch (fun_info->f.fun_type) { case ARG_GEN_BIG: scavenge_large_bitmap(p, GET_FUN_LARGE_BITMAP(fun_info), size); p += size; break; ... } return p; } Here the `size` argument in `scavenge_PAP_payload` is the number of arguments applied to the function in `pap->fun`. But when scavenging the function's bitmap we're using this number as the size of the bitmap which doesn't make sense to me, because I think size of the function's bitmap and size of the PAP's payload may be different. Or in other words I may have the same function used in many PAPs with different n_args, but that'd be buggy if this code is correct. I haven't checked every single place where we build a PAP but for example the `NEW_PAP` macro uses the argument's (another PAP) function directly, without making any bitmap-related changes, but bumps n_args by one. If the code above is right, then this new PAP will be scavenged incorrectly. Am I missing anything? Thanks, Ömer

5 13

ghc 8.8.x change in behaviour when loading in-memory modules
by Tseen She 26 Feb '20

26 Feb '20

Dear ghc devs, I am in the process of updating my dev tool, which uses the ghc api, to support ghc-8.8.x. However, I have noticed that one of my regression tests is failing, and I have isolated the problem to a change in behaviour in typecheckModule when using a Target with an in-memory buffer (i.e. targetContents is a Just). Using debug tracing, I can confirm that in the following code, the ms_hspp_buf of a ModSummary has my in-memory version: https://gitlab.com/tseenshe/hsinspect/-/blob/dfcb50a4e3437fbb4d0a3eb786b99c… minf_rdr_env' :: GHC.GhcMonad m => GHC.ModuleName -> m GlobalRdrEnv minf_rdr_env' m = do modSum <- GHC.getModSummary m pmod <- GHC.parseModule modSum tmod <- GHC.typecheckModule pmod let (tc_gbl_env, _) = GHC.tm_internals_ tmod pure $ tcg_rdr_env tc_gbl_env You can see how I setup the session in https://gitlab.com/tseenshe/hsinspect/-/blob/dfcb50a4e3437fbb4d0a3eb786b99c… And whereas in 8.4.4 and 8.6.5 the typecheckModule call is working on the in-memory representation, in 8.8.1 and 8.8.2 it is falling back to the version on disk and therefore failing, because I have made in-memory changes that are necessary. Aside: I am able to observe this because I have a function that trims everything except the module declaration and import section, and can therefore tolerate typos in the rest of the file. Now my code fails if the entire file does not typecheck, which is also incredibly slow, because I only want access to the import section. Best regards, Tseen She

1 0

Time-profiling the compiler pipeline
by Sergej Jaskiewicz 26 Feb '20

26 Feb '20

Hello everyone, Recently I've been exploring ways to improve the LLVM support in GHC. While doing that, I've encountered an issue that I think is worth looking into. Namely, the compilation times with LLVM are considerably longer than with the native code generator. To be more concrete, on my machine I've managed to build the whole toolchain in just half an hour with NCG, but the same build with LLVM-powered GHC took several hours. If we want to improve compilation times, we should be able to measure it first. As far as I understand, the only way for profiling the compiler pipeline today is building GHC with special flags: https://gitlab.haskell.org/ghc/ghc/wikis/debugging/profiling-ghc This has three important downsides: * We need build the whole compiler ourselves. If I'm just a user interested in why my program is so slow to compile and how I should fix it, this is too much for me to ask to do. * There's just too much irrelevant data in the output. For example, it doesn't tell me which part of my Haskell code is taking so long to compile. It would be nice if the report contained all the metadata that I need to go and fix my code. * If we're using LLVM, we're pretty much out of luck, because we're using an existing LLVM binary which is likely not built for profiling. There have been the same problem with the Clang compiler until recently. In Clang 9, which was released in the last year, there's a new flag '-ftime-trace' that produces a beautiful flame chart with all the metadata needed to identify bottlenecks. This flag tells Clang to produce a JSON file that can be visualized in Chrome's chrome://tracing page, or on https://www.speedscope.app. This is described in more details here: https://aras-p.info/blog/2019/01/16/time-trace-timeline-flame-chart-profile… I'm thinking about implementing similar functionality in GHC. One option would be to utilize the plugin infrastructure. Maybe I'm missing something and there are better options? — Sergej.

2 1