- Haskell-Cafe - Haskell.org

Re: [Haskell-cafe] Property checks and PostgreSQL?
by Anthony Clayden 15 Mar '21

15 Mar '21

> PostgreSQL data base, Haskell back end, PureScript front end. And there is a fault somewhere in this stack: sometimes some items present in the data base do not make it to the front end. A small fortune in programmer hours has been spent ... Speaking purely from a SQL-and-stuff point of view, "programmer hours spent" sounds to me like a problem with nulls. Are any of your columns nullable? Are you sure they should be nullable? In particular are any of the columns in the keys nullable? Or any of the columns on which you're joining tables? (Perhaps a proper value in one table, but a null in the table it's supposed to equal?) How are you mapping between Haskell field values to/from nullable columns in the SQL? Is that mapping fool-proof? Is it consistent across different tables using what should be the same value? And in general: SQL nulls are an abomination; don't do that; do normalise your database schema to avoid nulls. AntC (35 years an SQL programmer, never trusted it)

1 0

Let's fight Int, Pt. 2
by Johannes Waldmann 15 Mar '21

15 Mar '21

Dear Cafe, it's bad to use Int where you really mean Nat, because this creates extra work (defensive programming, or mental gymnastics to prove that you don't need it) Another Int anti-feature is arithmetics with silent wrap-around - with respect to an unspecified modulus, as highlighted by a recent example: https://gitlab.haskell.org/ghc/ghc/-/issues/19500 What are the possibilities for a "checked Int"? (also, "checked Nat") A historical precedent is set by Ada http://archive.adaic.com/standards/83lrm/html/lrm-03-05.html#3.5.4 "The exception NUMERIC_ERROR is raised by the execution of an operation ... that cannot deliver the correct result" Another point in the design space is Rust: "Integer operators will panic when they overflow when compiled in debug mode." https://doc.rust-lang.org/reference/expressions/operator-expr.html?highligh… The Haskell standard https://www.haskell.org/onlinereport/haskell2010/haskellch6.html#x13-135000… says that "The results of exceptional conditions (such as overflow or underflow) on the fixed-precision numeric types are undefined;" so any program that relies on Int's wrap-around, is already broken. GHC's base library specifies "modulo 2^width" semantics for signed (Data.Int) and unsigned (Data.Word) integer types. Finally, about "fighting" - I guess it's more about strategies to cope with the present situation. - J.W.

2 1

Is Data.Pool not thread-safe or have I done something silly?
by Cody Gman 14 Mar '21

14 Mar '21

I wrote some code with a Data.Pool that has 1 stripe, 1 max resource, and then basically did `Async.replicateConcurrently_ . withResource $ \res -> f res`. I expect withResource to block in each of those threads until the first thread spawned is done with that Resource and releases. To be clear, I get output like: ```shell /tmp $ stack exec --resolver lts-17.5 --package resource-pool --package stm --package hslogger --package time --package concurrent-output -- ghc -O2 -threaded -fwarn-unused-imports testpool.hs && ./testpool acquire 0 - .81428 acquire 1 - .81431 acquire 2 - .81438 acquire 3 - .81440 acquire 4 - .81448 ThreadId 8: processing 1 - .82460 ThreadId 17: processing 4 - .82461 ThreadId 11: processing 2 - .82464 ThreadId 14: processing 3 - .82464 ThreadId 5: processing 0 - .82465 anything else? release 4 - .14427 release 3 - .14430 release 2 - .14431 release 1 - .14431 release 0 - .14432 anything else? anything else? ``` I expect output like: ```shell /tmp $ stack exec --resolver lts-17.5 --package resource-pool --package stm --package hslogger --package time --package concurrent-output -- ghc -O2 -threaded -fwarn-unused-imports testpool.hs && ./testpool [1 of 1] Compiling Main ( testpool.hs, testpool.o ) Linking testpool ... acquire 0 ThreadId 5: processing 0 - .01129 release 0 acquire 1 -- I thought this would have blocked until 0 was released ThreadId 8: processing 1 - .01120 release 1 acquire 2 ThreadId 11: processing 2 - .01123 release 2 acquire 3 ThreadId 14: processing 3 - .01129 release 3 acquire 4 ThreadId 17: processing 4 - .01129 release 4 anything else? anything else? anything else? ``` Here is the code: ``` #!/usr/bin/env stack -- stack script --resolver lts-17.5 --package resource-pool --package stm --package hslogger --package time --package concurrent-output import Control.Concurrent import Control.Concurrent.Async import Control.Concurrent.STM import Data.Pool import Data.Time import System.Console.Concurrent import System.Clock main :: IO () main = do counter <- newTVarIO 0 let acquire = do k <- atomically $ do k <- readTVar counter writeTVar counter (k + 1) return k now <- getTime Monotonic outputConcurrent ("acquire " ++ show k ++ " - "++ show now ++ "\n") return k release k = do now <- getTime Monotonic outputConcurrent ("acquire " ++ show k ++ " - "++ show now ++ "\n") withConcurrentOutput $ do -- create a pool that only allows 1 resource pool <- createPool acquire release 1 500 1 replicateConcurrently_ 5 $ do useResourceFor (seconds 10) pool -- Why do you need these to see the release messages? putStrLn "anything else?" >> threadDelay (seconds 5) putStrLn "anything else?" >> threadDelay (seconds 5) putStrLn "anything else?" >> threadDelay (seconds 5) useResourceFor waitSeconds pool = withResource pool $ \i -> do threadDelay waitSeconds tid <- myThreadId now <- getTime Monotonic outputConcurrent $ show tid <> ": " <> " processing " <> show i <> " - " <> show now <> "\n" seconds = (* 1000000) ``` I'm hoping someone could help explain what I did wrong or confirm it's a bug. Thanks, Cody

2 1

Re: [Haskell-cafe] naturally, length :: a -> Int
by Olaf Klinke 14 Mar '21

14 Mar '21

Which bugs can be caught at compile-time by having length return natural numbers? Regarding space, Int instead of Word only wastes the sign bit, doesn't it? I like the idea that Johannes Waldmann and Jaro Reinders brought up: Why is length :: Foldable a => a -> Int so convenient? Short answer: Because of "affine" things like `availableSpace - length xs'. There are indeed two types involved here, as the foundation package points out: relative offsets (like a tangent space?) and absolute counts. Think of NominalDiffTime versus UTCTime, or the two interpretations of vectors as points/movements in space. In this light one could regard the current length as "the relative offset of the end of the list" which can readily be subtracted from another relative offset. In mathematical terms: Int the free group over the monoid of cardinal lengths. While we're at it: Can there be a Fractional type permitting only positive numbers, as the positive real numbers are closed under division? Can there be a type of rationals between 0 and 1 (which is closed under multiplication) where (/x) is the right adjoint to (*x)? Olaf

4 6

Property checks and PostgreSQL?
by Ignat Insarov 14 Mar '21

14 Mar '21

Note: I posted this [elsewhere] — I hope it is permissible to also post here. There seems to be no single right venue for Haskell questions at this time. [elsewhere]: https://discourse.haskell.org/t/property-checks-and-postgresql/2078 So, I have a three layer stack: PostgreSQL data base, Haskell back end, PureScript front end. And there is a fault somewhere in this stack: sometimes some items present in the data base do not make it to the front end. A small fortune in programmer hours has been spent searching for the fault but there is no progress. No one can figure out even roughly where the fault is. Our last hope is to property check the whole thing, from end to end, verifying that any things put into the data base can be retrieved by the front end. We have in place a fixture that creates a new PostgreSQL cluster, runs the back end on it and makes some scripted HTTP requests. We use it to run unit tests, such like _«if I make this `PUT` request and then that `GET` request to the same end point, I should get back the same thing as I put in»_. In principle it would not be a problem to make a property check out of this property. Practically, tearing down the cluster, building a new pristine one and initializing the data base anew takes seconds, so this way even a trivial property check would take minutes to run. Theoretically, we can carefully reset the data base back to the initial state after each run of the property check. Given that only a small amount of data is modified, it would take imperceptibly little time. But it is hard to know what exactly should be reset, and it is easy to get it wrong if done by hand! One trick we do use is SQL transactions. We take a single connexion, start a transaction, perform a check and then roll the transaction back. Unfortunately, even this is not completely safe: it looks as though sequences used to draw primary keys from are not reset to their previous values! _(Or maybe we are doing something wrong — I am not really a PostgreSQL guru so I am not too sure of myself.)_ But even ignoring this problem _(it is irrelevant to most checks)_, there is another, more severe problem: transactions guarantee that uncommitted data is only visible inside the same connexion. So, there is no way to request it from the HTTP API. This trick is suitable for fuzzing the data base – back end layer, but nothing beyond that. Another thing I heard being done is to selectively drop and re-create exactly the tables affected by a run of a property check. This seems challenging to automate though. How can I know in a general way if a table has been touched? And then, how can I re-create exactly the affected tables given that the data base initialization code is an opaque blob? I wonder if this problem is already solved. In any case, any advice is welcome!

5 9

Package Takeover: `toml`
by Emily Pillmore 14 Mar '21

14 Mar '21

To whom it may concern, I'd like to take over maintenance of the package `toml`. I have a burning need. I tried to contact Spiros a year ago, but no response has come. The last release was 2017, and it's safe to say the package is totally abandoned. My hackage user is topos. Cheers, E

20 51

data kinds and exhaustiveness of typeclass instances
by Paul Brauner 14 Mar '21

14 Mar '21

Hello, the following program doesn't typecheck in GHC 9: data Tag = A | B data Foo (a :: Tag) = Foo class C a where f :: a -> Int instance C (Foo A) where f x = 1 instance C (Foo B) where f x = 2 g :: Foo a -> Int g = f Yet one could argue that for all a :: Tag, C (Foo a) holds because a :: Tag can only take two values: A or B, and C (Foo A) and C (Foo B) hold. Is there a way to somehow convince GHC of that fact so that g typechecks? Cheers, Paul

6 7

[ANN] singletons-3.0, singletons-th-3.0, and singletons-base-3.0
by Ryan Scott 13 Mar '21

13 Mar '21

I'm happy to announce the 3.0 release of the singletons library, as well as the debut of its companion libraries, singletons-th and singletons-base. There's been a fair bit of improvements, cleanup, and reorganization since the last release, so I thought it appropriate to give this release a super-major version bump. This announcement will go over some highlights of the new releases, but if you want the full details, you can consult the changelog entries for each library below: * singletons-3.0: http://hackage.haskell.org/package/singletons-3.0/changelog * singletons-th-3.0: http://hackage.haskell.org/package/singletons-th-3.0/changelog * singletons-base-3.0: http://hackage.haskell.org/package/singletons-base-3.0/changelog ########################## ## Why three libraries? ## ########################## Instead of there just being one singletons library, there are now _three_ libraries: * singletons: A small, foundational library that defines basic singleton-related types and definitions. The new singletons library supports GHC 8.0 or later, does not depend on any external libraries, and does not make use of Template Haskell. * singletons-th: A library that defines Template Haskell functionality for promoting term-level functions to type-level equivalents and singling functions to dependently typed equivalents. This library requires bleeding-edge GHC features, and as a result, it only supports GHC 9.0 or later. * singletons-base: A library that uses singletons-th to define promoted and singled functions from the base library, including the Prelude. Similarly to singletons-th, this library only supports GHC 9.0 or later. Previous singletons releases combined all of this functionality into a single library. However, discussions with singletons users revealed that this was a frequent source of pain: * The total time it took to compile the old singletons library dissuaded some from using it, especially for those who only wanted to use a small part of the library. * Some users wished to use singletons in environments where Template Haskell is not an option. * Some users wanted to use core definitions from Data.Singletons, but because the old singletons library only ever supported one GHC at a time, this was impractical. The new package structure in the 3.0 release is an attempt to address some of these pain points. In particular, most of the code which results in extensive compile times is now confined to the singletons-base library. For the full discussion that led up to this, see [1]. ########################### ## Module reorganization ## ########################### Splitting up singletons into smaller packages provided a rare opportunity to clean up the module naming conventions, which were confusing and inconsistent in many places. The 3.0 releases now use the following conventions: * All modules in singletons now begin with Data.Singletons.*. * All modules in singletons-th now begin with Data.Singletons.TH.*. * Most modules in singletons-base now reflect the modules from base from which they take inspiration. For example, the Prelude module now has a Prelude.Singletons counterpart in singletons. Similarly, there are also modules like Control.Monad.Singletons, GHC.TypeLits.Singletons, etc. All other modules that do not correspond to something in base now begin with Data.Singletons.Base.*. The convention now used in singletons-base was inspired by the conventions used in the lens library. The end result is that is now somewhat easier to figure out what parts of singletons-base to import. Because this is a significant departure from previous singletons releases, if you want to continue supporting pre-3.0 releases, you will likely need to put something like this in your .cabal file: flag singletons-3-0 description: Use @singletons-3.0@ or later. default: True library ... if flag(singletons-3-0) build-depends: singletons-base >= 3.0 -- You may also need to depend on singletons >= 3.0 or singletons-th >= 3.0 else build-depends: singletons < 3.0 ################### ## Other changes ## ################### Besides the package and module reorganization mentioned above, singletons-{,th,base}-3.0 feature a variety of new quality-of-life improvements. Here are a handful of the more notable improvements: * The OptionsM type, used for configuring how the Template Haskell machinery in singletons-th works, is now an instance of Quote. This avoids the need to use the Control.Monad.Trans.Class.lift function to lift quoted declarations into OptionsM. [2] * Data.Singletons.TH.Options now defines a promotedDataTypeOrConName option. Overriding this option can be useful in situations where one wishes to promote types such as `Nat`, `Symbol`, or data types built on top of them. [3] * The internals of the ShowSing class have been refactored to allow deriving Show instances for Sing types (e.g., deriving instance ShowSing a => Show (SList (z :: [a]))). [4] * Improve the quality of GHCi's output when using :kind on types defined by singletons-th. [5] If you encounter any issues, feel free to leave a bug report at https://github.com/goldfirere/singletons/issues. Happy (type-level) hacking, Ryan S. ----- [1] https://github.com/goldfirere/singletons/issues/420 [2] https://github.com/goldfirere/singletons/pull/484 [3] https://github.com/goldfirere/singletons/pull/462 [4] https://github.com/goldfirere/singletons/pull/486 [5] https://github.com/goldfirere/singletons/pull/446

2 1

ANN: hledger-1.21
by Simon Michael 11 Mar '21

11 Mar '21

I'm very pleased to announce hledger 1.21 ! http://hledger.org/release-notes.html#hledger-1-21 describes the user-visible changes. Highlights include: - About 10% faster/more space-efficient, and much more in some cases - Jump to any topic in the manuals from the command line - More polish for valuation and balance reports, and a new value change report - Improvements to check, print and other commands. Thank you to the following release contributors this time around: Dmitry Astapov, Arnout Engelen, Damien Cassou, aragaer, Doug Goldstein, Caleb Maclennan, Felix Van der Jeugt, heavy first time contributor Vladimir Zhelezov, and especially Stephen Morgan. hledger (http://hledger.org) is a robust, cross-platform, plain text accounting tool, with command-line, terminal and web UIs. It is an actively maintained, largely compatible reimplementation of Ledger CLI with many improvements. You can use it to track money, time, investments, cryptocurrencies, inventory and more. See also the Plain Text Accounting site (http://plaintextaccounting.org) http://hledger.org/download shows all the ways to install hledger on mac, windows or unix (stack, cabal, brew, nix, CI binaries, your package manager..). Or, run this bash script to install or upgrade to the latest release: $ curl -sO https://raw.githubusercontent.com/simonmichael/hledger/master/hledger-insta… $ less hledger-install.sh # security review $ bash hledger-install.sh New users, check out https://hledger.org/quickstart. To get help, see http://hledger.org#help-feedback, and join our chat channel via Freenode (#hledger, http://irc.hledger.org) or Matrix (#freenode_#hledger:matrix.org, http://matrix.hledger.org) Beginners and experts, contributors, sponsors, and all feedback are most welcome! Wishing you health and prosperity, -Simon

1 0

Multitier Haskell?
by Adriaan Leijnse 09 Mar '21

09 Mar '21

Dear Café, Are there any techniques for compiling a single client-server program such that clients are not distributed any server code? I'm currently looking at implementing an FRP library with a type system like Gavial's [0]. The types will make it clear on which side code belongs. My current thinking is to have Client and Server instances of a ClientServer type class which can safely implement the functions of the other side as no-ops/bottom, but I imagine that would at leastrequire a dead code elimination step after compilation to get the desired effect? Adriaan [0] https://arxiv.org/pdf/2002.06188.pdf

1 0